xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 0661cb2a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  * cpuid support routines
5  *
6  * derived from arch/x86/kvm/x86.c
7  *
8  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
9  * Copyright IBM Corporation, 2008
10  */
11 
12 #include <linux/kvm_host.h>
13 #include <linux/export.h>
14 #include <linux/vmalloc.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/stat.h>
17 
18 #include <asm/processor.h>
19 #include <asm/user.h>
20 #include <asm/fpu/xstate.h>
21 #include <asm/sgx.h>
22 #include "cpuid.h"
23 #include "lapic.h"
24 #include "mmu.h"
25 #include "trace.h"
26 #include "pmu.h"
27 
28 /*
29  * Unlike "struct cpuinfo_x86.x86_capability", kvm_cpu_caps doesn't need to be
30  * aligned to sizeof(unsigned long) because it's not accessed via bitops.
31  */
32 u32 kvm_cpu_caps[NR_KVM_CPU_CAPS] __read_mostly;
33 EXPORT_SYMBOL_GPL(kvm_cpu_caps);
34 
35 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
36 {
37 	int feature_bit = 0;
38 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
39 
40 	xstate_bv &= XFEATURE_MASK_EXTEND;
41 	while (xstate_bv) {
42 		if (xstate_bv & 0x1) {
43 		        u32 eax, ebx, ecx, edx, offset;
44 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
45 			offset = compacted ? ret : ebx;
46 			ret = max(ret, offset + eax);
47 		}
48 
49 		xstate_bv >>= 1;
50 		feature_bit++;
51 	}
52 
53 	return ret;
54 }
55 
56 #define F feature_bit
57 #define SF(name) (boot_cpu_has(X86_FEATURE_##name) ? F(name) : 0)
58 
59 static inline struct kvm_cpuid_entry2 *cpuid_entry2_find(
60 	struct kvm_cpuid_entry2 *entries, int nent, u32 function, u32 index)
61 {
62 	struct kvm_cpuid_entry2 *e;
63 	int i;
64 
65 	for (i = 0; i < nent; i++) {
66 		e = &entries[i];
67 
68 		if (e->function == function && (e->index == index ||
69 		    !(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX)))
70 			return e;
71 	}
72 
73 	return NULL;
74 }
75 
76 static int kvm_check_cpuid(struct kvm_cpuid_entry2 *entries, int nent)
77 {
78 	struct kvm_cpuid_entry2 *best;
79 
80 	/*
81 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
82 	 * canonical address checks; exit if it is ever changed.
83 	 */
84 	best = cpuid_entry2_find(entries, nent, 0x80000008, 0);
85 	if (best) {
86 		int vaddr_bits = (best->eax & 0xff00) >> 8;
87 
88 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
89 			return -EINVAL;
90 	}
91 
92 	return 0;
93 }
94 
95 void kvm_update_pv_runtime(struct kvm_vcpu *vcpu)
96 {
97 	struct kvm_cpuid_entry2 *best;
98 
99 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
100 
101 	/*
102 	 * save the feature bitmap to avoid cpuid lookup for every PV
103 	 * operation
104 	 */
105 	if (best)
106 		vcpu->arch.pv_cpuid.features = best->eax;
107 }
108 
109 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
110 {
111 	struct kvm_cpuid_entry2 *best;
112 
113 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
114 	if (best) {
115 		/* Update OSXSAVE bit */
116 		if (boot_cpu_has(X86_FEATURE_XSAVE))
117 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
118 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
119 
120 		cpuid_entry_change(best, X86_FEATURE_APIC,
121 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
122 	}
123 
124 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
125 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
126 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
127 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
128 
129 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
130 	if (best)
131 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
132 
133 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
134 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
135 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
136 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
137 
138 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
139 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
140 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
141 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
142 
143 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
144 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
145 		if (best)
146 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
147 					   vcpu->arch.ia32_misc_enable_msr &
148 					   MSR_IA32_MISC_ENABLE_MWAIT);
149 	}
150 }
151 EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
152 
153 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
154 {
155 	struct kvm_lapic *apic = vcpu->arch.apic;
156 	struct kvm_cpuid_entry2 *best;
157 
158 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
159 	if (best && apic) {
160 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
161 			apic->lapic_timer.timer_mode_mask = 3 << 17;
162 		else
163 			apic->lapic_timer.timer_mode_mask = 1 << 17;
164 
165 		kvm_apic_set_version(vcpu);
166 	}
167 
168 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
169 	if (!best)
170 		vcpu->arch.guest_supported_xcr0 = 0;
171 	else
172 		vcpu->arch.guest_supported_xcr0 =
173 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
174 
175 	/*
176 	 * Bits 127:0 of the allowed SECS.ATTRIBUTES (CPUID.0x12.0x1) enumerate
177 	 * the supported XSAVE Feature Request Mask (XFRM), i.e. the enclave's
178 	 * requested XCR0 value.  The enclave's XFRM must be a subset of XCRO
179 	 * at the time of EENTER, thus adjust the allowed XFRM by the guest's
180 	 * supported XCR0.  Similar to XCR0 handling, FP and SSE are forced to
181 	 * '1' even on CPUs that don't support XSAVE.
182 	 */
183 	best = kvm_find_cpuid_entry(vcpu, 0x12, 0x1);
184 	if (best) {
185 		best->ecx &= vcpu->arch.guest_supported_xcr0 & 0xffffffff;
186 		best->edx &= vcpu->arch.guest_supported_xcr0 >> 32;
187 		best->ecx |= XFEATURE_MASK_FPSSE;
188 	}
189 
190 	kvm_update_pv_runtime(vcpu);
191 
192 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
193 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
194 
195 	kvm_pmu_refresh(vcpu);
196 	vcpu->arch.cr4_guest_rsvd_bits =
197 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
198 
199 	kvm_hv_set_cpuid(vcpu);
200 
201 	/* Invoke the vendor callback only after the above state is updated. */
202 	static_call(kvm_x86_vcpu_after_set_cpuid)(vcpu);
203 
204 	/*
205 	 * Except for the MMU, which needs to do its thing any vendor specific
206 	 * adjustments to the reserved GPA bits.
207 	 */
208 	kvm_mmu_after_set_cpuid(vcpu);
209 }
210 
211 static int is_efer_nx(void)
212 {
213 	return host_efer & EFER_NX;
214 }
215 
216 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
217 {
218 	int i;
219 	struct kvm_cpuid_entry2 *e, *entry;
220 
221 	entry = NULL;
222 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
223 		e = &vcpu->arch.cpuid_entries[i];
224 		if (e->function == 0x80000001) {
225 			entry = e;
226 			break;
227 		}
228 	}
229 	if (entry && cpuid_entry_has(entry, X86_FEATURE_NX) && !is_efer_nx()) {
230 		cpuid_entry_clear(entry, X86_FEATURE_NX);
231 		printk(KERN_INFO "kvm: guest NX capability removed\n");
232 	}
233 }
234 
235 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
236 {
237 	struct kvm_cpuid_entry2 *best;
238 
239 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
240 	if (!best || best->eax < 0x80000008)
241 		goto not_found;
242 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
243 	if (best)
244 		return best->eax & 0xff;
245 not_found:
246 	return 36;
247 }
248 
249 /*
250  * This "raw" version returns the reserved GPA bits without any adjustments for
251  * encryption technologies that usurp bits.  The raw mask should be used if and
252  * only if hardware does _not_ strip the usurped bits, e.g. in virtual MTRRs.
253  */
254 u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
255 {
256 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
257 }
258 
259 /* when an old userspace process fills a new kernel module */
260 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
261 			     struct kvm_cpuid *cpuid,
262 			     struct kvm_cpuid_entry __user *entries)
263 {
264 	int r, i;
265 	struct kvm_cpuid_entry *e = NULL;
266 	struct kvm_cpuid_entry2 *e2 = NULL;
267 
268 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
269 		return -E2BIG;
270 
271 	if (cpuid->nent) {
272 		e = vmemdup_user(entries, array_size(sizeof(*e), cpuid->nent));
273 		if (IS_ERR(e))
274 			return PTR_ERR(e);
275 
276 		e2 = kvmalloc_array(cpuid->nent, sizeof(*e2), GFP_KERNEL_ACCOUNT);
277 		if (!e2) {
278 			r = -ENOMEM;
279 			goto out_free_cpuid;
280 		}
281 	}
282 	for (i = 0; i < cpuid->nent; i++) {
283 		e2[i].function = e[i].function;
284 		e2[i].eax = e[i].eax;
285 		e2[i].ebx = e[i].ebx;
286 		e2[i].ecx = e[i].ecx;
287 		e2[i].edx = e[i].edx;
288 		e2[i].index = 0;
289 		e2[i].flags = 0;
290 		e2[i].padding[0] = 0;
291 		e2[i].padding[1] = 0;
292 		e2[i].padding[2] = 0;
293 	}
294 
295 	r = kvm_check_cpuid(e2, cpuid->nent);
296 	if (r) {
297 		kvfree(e2);
298 		goto out_free_cpuid;
299 	}
300 
301 	kvfree(vcpu->arch.cpuid_entries);
302 	vcpu->arch.cpuid_entries = e2;
303 	vcpu->arch.cpuid_nent = cpuid->nent;
304 
305 	cpuid_fix_nx_cap(vcpu);
306 	kvm_update_cpuid_runtime(vcpu);
307 	kvm_vcpu_after_set_cpuid(vcpu);
308 
309 out_free_cpuid:
310 	kvfree(e);
311 
312 	return r;
313 }
314 
315 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
316 			      struct kvm_cpuid2 *cpuid,
317 			      struct kvm_cpuid_entry2 __user *entries)
318 {
319 	struct kvm_cpuid_entry2 *e2 = NULL;
320 	int r;
321 
322 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
323 		return -E2BIG;
324 
325 	if (cpuid->nent) {
326 		e2 = vmemdup_user(entries, array_size(sizeof(*e2), cpuid->nent));
327 		if (IS_ERR(e2))
328 			return PTR_ERR(e2);
329 	}
330 
331 	r = kvm_check_cpuid(e2, cpuid->nent);
332 	if (r) {
333 		kvfree(e2);
334 		return r;
335 	}
336 
337 	kvfree(vcpu->arch.cpuid_entries);
338 	vcpu->arch.cpuid_entries = e2;
339 	vcpu->arch.cpuid_nent = cpuid->nent;
340 
341 	kvm_update_cpuid_runtime(vcpu);
342 	kvm_vcpu_after_set_cpuid(vcpu);
343 
344 	return 0;
345 }
346 
347 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
348 			      struct kvm_cpuid2 *cpuid,
349 			      struct kvm_cpuid_entry2 __user *entries)
350 {
351 	int r;
352 
353 	r = -E2BIG;
354 	if (cpuid->nent < vcpu->arch.cpuid_nent)
355 		goto out;
356 	r = -EFAULT;
357 	if (copy_to_user(entries, vcpu->arch.cpuid_entries,
358 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
359 		goto out;
360 	return 0;
361 
362 out:
363 	cpuid->nent = vcpu->arch.cpuid_nent;
364 	return r;
365 }
366 
367 /* Mask kvm_cpu_caps for @leaf with the raw CPUID capabilities of this CPU. */
368 static __always_inline void __kvm_cpu_cap_mask(unsigned int leaf)
369 {
370 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
371 	struct kvm_cpuid_entry2 entry;
372 
373 	reverse_cpuid_check(leaf);
374 
375 	cpuid_count(cpuid.function, cpuid.index,
376 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
377 
378 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
379 }
380 
381 static __always_inline
382 void kvm_cpu_cap_init_scattered(enum kvm_only_cpuid_leafs leaf, u32 mask)
383 {
384 	/* Use kvm_cpu_cap_mask for non-scattered leafs. */
385 	BUILD_BUG_ON(leaf < NCAPINTS);
386 
387 	kvm_cpu_caps[leaf] = mask;
388 
389 	__kvm_cpu_cap_mask(leaf);
390 }
391 
392 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
393 {
394 	/* Use kvm_cpu_cap_init_scattered for scattered leafs. */
395 	BUILD_BUG_ON(leaf >= NCAPINTS);
396 
397 	kvm_cpu_caps[leaf] &= mask;
398 
399 	__kvm_cpu_cap_mask(leaf);
400 }
401 
402 void kvm_set_cpu_caps(void)
403 {
404 	unsigned int f_nx = is_efer_nx() ? F(NX) : 0;
405 #ifdef CONFIG_X86_64
406 	unsigned int f_gbpages = F(GBPAGES);
407 	unsigned int f_lm = F(LM);
408 #else
409 	unsigned int f_gbpages = 0;
410 	unsigned int f_lm = 0;
411 #endif
412 	memset(kvm_cpu_caps, 0, sizeof(kvm_cpu_caps));
413 
414 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)) >
415 		     sizeof(boot_cpu_data.x86_capability));
416 
417 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
418 	       sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)));
419 
420 	kvm_cpu_cap_mask(CPUID_1_ECX,
421 		/*
422 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
423 		 * advertised to guests via CPUID!
424 		 */
425 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
426 		0 /* DS-CPL, VMX, SMX, EST */ |
427 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
428 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
429 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
430 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
431 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
432 		F(F16C) | F(RDRAND)
433 	);
434 	/* KVM emulates x2apic in software irrespective of host support. */
435 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
436 
437 	kvm_cpu_cap_mask(CPUID_1_EDX,
438 		F(FPU) | F(VME) | F(DE) | F(PSE) |
439 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
440 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
441 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
442 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
443 		0 /* Reserved, DS, ACPI */ | F(MMX) |
444 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
445 		0 /* HTT, TM, Reserved, PBE */
446 	);
447 
448 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
449 		F(FSGSBASE) | F(SGX) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
450 		F(BMI2) | F(ERMS) | F(INVPCID) | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
451 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
452 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
453 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
454 	);
455 
456 	kvm_cpu_cap_mask(CPUID_7_ECX,
457 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
458 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
459 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
460 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/ |
461 		F(SGX_LC) | F(BUS_LOCK_DETECT)
462 	);
463 	/* Set LA57 based on hardware capability. */
464 	if (cpuid_ecx(7) & F(LA57))
465 		kvm_cpu_cap_set(X86_FEATURE_LA57);
466 
467 	/*
468 	 * PKU not yet implemented for shadow paging and requires OSPKE
469 	 * to be set on the host. Clear it if that is not the case
470 	 */
471 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
472 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
473 
474 	kvm_cpu_cap_mask(CPUID_7_EDX,
475 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
476 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
477 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM) |
478 		F(SERIALIZE) | F(TSXLDTRK) | F(AVX512_FP16)
479 	);
480 
481 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
482 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
483 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
484 
485 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
486 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
487 	if (boot_cpu_has(X86_FEATURE_STIBP))
488 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
489 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
490 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
491 
492 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
493 		F(AVX_VNNI) | F(AVX512_BF16)
494 	);
495 
496 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
497 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
498 	);
499 
500 	kvm_cpu_cap_init_scattered(CPUID_12_EAX,
501 		SF(SGX1) | SF(SGX2)
502 	);
503 
504 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
505 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
506 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
507 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
508 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
509 		F(TOPOEXT) | F(PERFCTR_CORE)
510 	);
511 
512 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
513 		F(FPU) | F(VME) | F(DE) | F(PSE) |
514 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
515 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
516 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
517 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
518 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
519 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
520 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
521 	);
522 
523 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
524 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
525 
526 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
527 		F(CLZERO) | F(XSAVEERPTR) |
528 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
529 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON)
530 	);
531 
532 	/*
533 	 * AMD has separate bits for each SPEC_CTRL bit.
534 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
535 	 * record that in cpufeatures so use them.
536 	 */
537 	if (boot_cpu_has(X86_FEATURE_IBPB))
538 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
539 	if (boot_cpu_has(X86_FEATURE_IBRS))
540 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
541 	if (boot_cpu_has(X86_FEATURE_STIBP))
542 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
543 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
544 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
545 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
546 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
547 	/*
548 	 * The preference is to use SPEC CTRL MSR instead of the
549 	 * VIRT_SPEC MSR.
550 	 */
551 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
552 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
553 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
554 
555 	/*
556 	 * Hide all SVM features by default, SVM will set the cap bits for
557 	 * features it emulates and/or exposes for L1.
558 	 */
559 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
560 
561 	kvm_cpu_cap_mask(CPUID_8000_001F_EAX,
562 		0 /* SME */ | F(SEV) | 0 /* VM_PAGE_FLUSH */ | F(SEV_ES) |
563 		F(SME_COHERENT));
564 
565 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
566 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
567 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
568 		F(PMM) | F(PMM_EN)
569 	);
570 
571 	/*
572 	 * Hide RDTSCP and RDPID if either feature is reported as supported but
573 	 * probing MSR_TSC_AUX failed.  This is purely a sanity check and
574 	 * should never happen, but the guest will likely crash if RDTSCP or
575 	 * RDPID is misreported, and KVM has botched MSR_TSC_AUX emulation in
576 	 * the past.  For example, the sanity check may fire if this instance of
577 	 * KVM is running as L1 on top of an older, broken KVM.
578 	 */
579 	if (WARN_ON((kvm_cpu_cap_has(X86_FEATURE_RDTSCP) ||
580 		     kvm_cpu_cap_has(X86_FEATURE_RDPID)) &&
581 		     !kvm_is_supported_user_return_msr(MSR_TSC_AUX))) {
582 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
583 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
584 	}
585 }
586 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
587 
588 struct kvm_cpuid_array {
589 	struct kvm_cpuid_entry2 *entries;
590 	int maxnent;
591 	int nent;
592 };
593 
594 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
595 					      u32 function, u32 index)
596 {
597 	struct kvm_cpuid_entry2 *entry;
598 
599 	if (array->nent >= array->maxnent)
600 		return NULL;
601 
602 	entry = &array->entries[array->nent++];
603 
604 	entry->function = function;
605 	entry->index = index;
606 	entry->flags = 0;
607 
608 	cpuid_count(entry->function, entry->index,
609 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
610 
611 	switch (function) {
612 	case 4:
613 	case 7:
614 	case 0xb:
615 	case 0xd:
616 	case 0xf:
617 	case 0x10:
618 	case 0x12:
619 	case 0x14:
620 	case 0x17:
621 	case 0x18:
622 	case 0x1f:
623 	case 0x8000001d:
624 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
625 		break;
626 	}
627 
628 	return entry;
629 }
630 
631 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
632 {
633 	struct kvm_cpuid_entry2 *entry;
634 
635 	if (array->nent >= array->maxnent)
636 		return -E2BIG;
637 
638 	entry = &array->entries[array->nent];
639 	entry->function = func;
640 	entry->index = 0;
641 	entry->flags = 0;
642 
643 	switch (func) {
644 	case 0:
645 		entry->eax = 7;
646 		++array->nent;
647 		break;
648 	case 1:
649 		entry->ecx = F(MOVBE);
650 		++array->nent;
651 		break;
652 	case 7:
653 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
654 		entry->eax = 0;
655 		if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
656 			entry->ecx = F(RDPID);
657 		++array->nent;
658 		break;
659 	default:
660 		break;
661 	}
662 
663 	return 0;
664 }
665 
666 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
667 {
668 	struct kvm_cpuid_entry2 *entry;
669 	int r, i, max_idx;
670 
671 	/* all calls to cpuid_count() should be made on the same cpu */
672 	get_cpu();
673 
674 	r = -E2BIG;
675 
676 	entry = do_host_cpuid(array, function, 0);
677 	if (!entry)
678 		goto out;
679 
680 	switch (function) {
681 	case 0:
682 		/* Limited to the highest leaf implemented in KVM. */
683 		entry->eax = min(entry->eax, 0x1fU);
684 		break;
685 	case 1:
686 		cpuid_entry_override(entry, CPUID_1_EDX);
687 		cpuid_entry_override(entry, CPUID_1_ECX);
688 		break;
689 	case 2:
690 		/*
691 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
692 		 * CPUID(function=2, index=0) may return different results each
693 		 * time, with the least-significant byte in EAX enumerating the
694 		 * number of times software should do CPUID(2, 0).
695 		 *
696 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
697 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
698 		 * return 01H. Software should ignore this value and not
699 		 * interpret it as an informational descriptor", while AMD's
700 		 * APM states that CPUID(2) is reserved.
701 		 *
702 		 * WARN if a frankenstein CPU that supports virtualization and
703 		 * a stateful CPUID.0x2 is encountered.
704 		 */
705 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
706 		break;
707 	/* functions 4 and 0x8000001d have additional index. */
708 	case 4:
709 	case 0x8000001d:
710 		/*
711 		 * Read entries until the cache type in the previous entry is
712 		 * zero, i.e. indicates an invalid entry.
713 		 */
714 		for (i = 1; entry->eax & 0x1f; ++i) {
715 			entry = do_host_cpuid(array, function, i);
716 			if (!entry)
717 				goto out;
718 		}
719 		break;
720 	case 6: /* Thermal management */
721 		entry->eax = 0x4; /* allow ARAT */
722 		entry->ebx = 0;
723 		entry->ecx = 0;
724 		entry->edx = 0;
725 		break;
726 	/* function 7 has additional index. */
727 	case 7:
728 		entry->eax = min(entry->eax, 1u);
729 		cpuid_entry_override(entry, CPUID_7_0_EBX);
730 		cpuid_entry_override(entry, CPUID_7_ECX);
731 		cpuid_entry_override(entry, CPUID_7_EDX);
732 
733 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
734 		if (entry->eax == 1) {
735 			entry = do_host_cpuid(array, function, 1);
736 			if (!entry)
737 				goto out;
738 
739 			cpuid_entry_override(entry, CPUID_7_1_EAX);
740 			entry->ebx = 0;
741 			entry->ecx = 0;
742 			entry->edx = 0;
743 		}
744 		break;
745 	case 9:
746 		break;
747 	case 0xa: { /* Architectural Performance Monitoring */
748 		struct x86_pmu_capability cap;
749 		union cpuid10_eax eax;
750 		union cpuid10_edx edx;
751 
752 		perf_get_x86_pmu_capability(&cap);
753 
754 		/*
755 		 * Only support guest architectural pmu on a host
756 		 * with architectural pmu.
757 		 */
758 		if (!cap.version)
759 			memset(&cap, 0, sizeof(cap));
760 
761 		eax.split.version_id = min(cap.version, 2);
762 		eax.split.num_counters = cap.num_counters_gp;
763 		eax.split.bit_width = cap.bit_width_gp;
764 		eax.split.mask_length = cap.events_mask_len;
765 
766 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
767 		edx.split.bit_width_fixed = cap.bit_width_fixed;
768 		if (cap.version)
769 			edx.split.anythread_deprecated = 1;
770 		edx.split.reserved1 = 0;
771 		edx.split.reserved2 = 0;
772 
773 		entry->eax = eax.full;
774 		entry->ebx = cap.events_mask;
775 		entry->ecx = 0;
776 		entry->edx = edx.full;
777 		break;
778 	}
779 	/*
780 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
781 	 * thus they can be handled by common code.
782 	 */
783 	case 0x1f:
784 	case 0xb:
785 		/*
786 		 * Populate entries until the level type (ECX[15:8]) of the
787 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
788 		 * the starting entry, filled by the primary do_host_cpuid().
789 		 */
790 		for (i = 1; entry->ecx & 0xff00; ++i) {
791 			entry = do_host_cpuid(array, function, i);
792 			if (!entry)
793 				goto out;
794 		}
795 		break;
796 	case 0xd:
797 		entry->eax &= supported_xcr0;
798 		entry->ebx = xstate_required_size(supported_xcr0, false);
799 		entry->ecx = entry->ebx;
800 		entry->edx &= supported_xcr0 >> 32;
801 		if (!supported_xcr0)
802 			break;
803 
804 		entry = do_host_cpuid(array, function, 1);
805 		if (!entry)
806 			goto out;
807 
808 		cpuid_entry_override(entry, CPUID_D_1_EAX);
809 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
810 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
811 							  true);
812 		else {
813 			WARN_ON_ONCE(supported_xss != 0);
814 			entry->ebx = 0;
815 		}
816 		entry->ecx &= supported_xss;
817 		entry->edx &= supported_xss >> 32;
818 
819 		for (i = 2; i < 64; ++i) {
820 			bool s_state;
821 			if (supported_xcr0 & BIT_ULL(i))
822 				s_state = false;
823 			else if (supported_xss & BIT_ULL(i))
824 				s_state = true;
825 			else
826 				continue;
827 
828 			entry = do_host_cpuid(array, function, i);
829 			if (!entry)
830 				goto out;
831 
832 			/*
833 			 * The supported check above should have filtered out
834 			 * invalid sub-leafs.  Only valid sub-leafs should
835 			 * reach this point, and they should have a non-zero
836 			 * save state size.  Furthermore, check whether the
837 			 * processor agrees with supported_xcr0/supported_xss
838 			 * on whether this is an XCR0- or IA32_XSS-managed area.
839 			 */
840 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
841 				--array->nent;
842 				continue;
843 			}
844 			entry->edx = 0;
845 		}
846 		break;
847 	case 0x12:
848 		/* Intel SGX */
849 		if (!kvm_cpu_cap_has(X86_FEATURE_SGX)) {
850 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
851 			break;
852 		}
853 
854 		/*
855 		 * Index 0: Sub-features, MISCSELECT (a.k.a extended features)
856 		 * and max enclave sizes.   The SGX sub-features and MISCSELECT
857 		 * are restricted by kernel and KVM capabilities (like most
858 		 * feature flags), while enclave size is unrestricted.
859 		 */
860 		cpuid_entry_override(entry, CPUID_12_EAX);
861 		entry->ebx &= SGX_MISC_EXINFO;
862 
863 		entry = do_host_cpuid(array, function, 1);
864 		if (!entry)
865 			goto out;
866 
867 		/*
868 		 * Index 1: SECS.ATTRIBUTES.  ATTRIBUTES are restricted a la
869 		 * feature flags.  Advertise all supported flags, including
870 		 * privileged attributes that require explicit opt-in from
871 		 * userspace.  ATTRIBUTES.XFRM is not adjusted as userspace is
872 		 * expected to derive it from supported XCR0.
873 		 */
874 		entry->eax &= SGX_ATTR_DEBUG | SGX_ATTR_MODE64BIT |
875 			      SGX_ATTR_PROVISIONKEY | SGX_ATTR_EINITTOKENKEY |
876 			      SGX_ATTR_KSS;
877 		entry->ebx &= 0;
878 		break;
879 	/* Intel PT */
880 	case 0x14:
881 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
882 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
883 			break;
884 		}
885 
886 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
887 			if (!do_host_cpuid(array, function, i))
888 				goto out;
889 		}
890 		break;
891 	case KVM_CPUID_SIGNATURE: {
892 		static const char signature[12] = "KVMKVMKVM\0\0";
893 		const u32 *sigptr = (const u32 *)signature;
894 		entry->eax = KVM_CPUID_FEATURES;
895 		entry->ebx = sigptr[0];
896 		entry->ecx = sigptr[1];
897 		entry->edx = sigptr[2];
898 		break;
899 	}
900 	case KVM_CPUID_FEATURES:
901 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
902 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
903 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
904 			     (1 << KVM_FEATURE_ASYNC_PF) |
905 			     (1 << KVM_FEATURE_PV_EOI) |
906 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
907 			     (1 << KVM_FEATURE_PV_UNHALT) |
908 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
909 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
910 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
911 			     (1 << KVM_FEATURE_POLL_CONTROL) |
912 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
913 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
914 
915 		if (sched_info_on())
916 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
917 
918 		entry->ebx = 0;
919 		entry->ecx = 0;
920 		entry->edx = 0;
921 		break;
922 	case 0x80000000:
923 		entry->eax = min(entry->eax, 0x8000001f);
924 		break;
925 	case 0x80000001:
926 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
927 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
928 		break;
929 	case 0x80000006:
930 		/* L2 cache and TLB: pass through host info. */
931 		break;
932 	case 0x80000007: /* Advanced power management */
933 		/* invariant TSC is CPUID.80000007H:EDX[8] */
934 		entry->edx &= (1 << 8);
935 		/* mask against host */
936 		entry->edx &= boot_cpu_data.x86_power;
937 		entry->eax = entry->ebx = entry->ecx = 0;
938 		break;
939 	case 0x80000008: {
940 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
941 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
942 		unsigned phys_as = entry->eax & 0xff;
943 
944 		/*
945 		 * If TDP (NPT) is disabled use the adjusted host MAXPHYADDR as
946 		 * the guest operates in the same PA space as the host, i.e.
947 		 * reductions in MAXPHYADDR for memory encryption affect shadow
948 		 * paging, too.
949 		 *
950 		 * If TDP is enabled but an explicit guest MAXPHYADDR is not
951 		 * provided, use the raw bare metal MAXPHYADDR as reductions to
952 		 * the HPAs do not affect GPAs.
953 		 */
954 		if (!tdp_enabled)
955 			g_phys_as = boot_cpu_data.x86_phys_bits;
956 		else if (!g_phys_as)
957 			g_phys_as = phys_as;
958 
959 		entry->eax = g_phys_as | (virt_as << 8);
960 		entry->edx = 0;
961 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
962 		break;
963 	}
964 	case 0x8000000A:
965 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
966 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
967 			break;
968 		}
969 		entry->eax = 1; /* SVM revision 1 */
970 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
971 				   ASID emulation to nested SVM */
972 		entry->ecx = 0; /* Reserved */
973 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
974 		break;
975 	case 0x80000019:
976 		entry->ecx = entry->edx = 0;
977 		break;
978 	case 0x8000001a:
979 	case 0x8000001e:
980 		break;
981 	case 0x8000001F:
982 		if (!kvm_cpu_cap_has(X86_FEATURE_SEV)) {
983 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
984 		} else {
985 			cpuid_entry_override(entry, CPUID_8000_001F_EAX);
986 
987 			/*
988 			 * Enumerate '0' for "PA bits reduction", the adjusted
989 			 * MAXPHYADDR is enumerated directly (see 0x80000008).
990 			 */
991 			entry->ebx &= ~GENMASK(11, 6);
992 		}
993 		break;
994 	/*Add support for Centaur's CPUID instruction*/
995 	case 0xC0000000:
996 		/*Just support up to 0xC0000004 now*/
997 		entry->eax = min(entry->eax, 0xC0000004);
998 		break;
999 	case 0xC0000001:
1000 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
1001 		break;
1002 	case 3: /* Processor serial number */
1003 	case 5: /* MONITOR/MWAIT */
1004 	case 0xC0000002:
1005 	case 0xC0000003:
1006 	case 0xC0000004:
1007 	default:
1008 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1009 		break;
1010 	}
1011 
1012 	r = 0;
1013 
1014 out:
1015 	put_cpu();
1016 
1017 	return r;
1018 }
1019 
1020 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1021 			 unsigned int type)
1022 {
1023 	if (type == KVM_GET_EMULATED_CPUID)
1024 		return __do_cpuid_func_emulated(array, func);
1025 
1026 	return __do_cpuid_func(array, func);
1027 }
1028 
1029 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
1030 
1031 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1032 			  unsigned int type)
1033 {
1034 	u32 limit;
1035 	int r;
1036 
1037 	if (func == CENTAUR_CPUID_SIGNATURE &&
1038 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
1039 		return 0;
1040 
1041 	r = do_cpuid_func(array, func, type);
1042 	if (r)
1043 		return r;
1044 
1045 	limit = array->entries[array->nent - 1].eax;
1046 	for (func = func + 1; func <= limit; ++func) {
1047 		r = do_cpuid_func(array, func, type);
1048 		if (r)
1049 			break;
1050 	}
1051 
1052 	return r;
1053 }
1054 
1055 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
1056 				 __u32 num_entries, unsigned int ioctl_type)
1057 {
1058 	int i;
1059 	__u32 pad[3];
1060 
1061 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
1062 		return false;
1063 
1064 	/*
1065 	 * We want to make sure that ->padding is being passed clean from
1066 	 * userspace in case we want to use it for something in the future.
1067 	 *
1068 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
1069 	 * have to give ourselves satisfied only with the emulated side. /me
1070 	 * sheds a tear.
1071 	 */
1072 	for (i = 0; i < num_entries; i++) {
1073 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
1074 			return true;
1075 
1076 		if (pad[0] || pad[1] || pad[2])
1077 			return true;
1078 	}
1079 	return false;
1080 }
1081 
1082 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
1083 			    struct kvm_cpuid_entry2 __user *entries,
1084 			    unsigned int type)
1085 {
1086 	static const u32 funcs[] = {
1087 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
1088 	};
1089 
1090 	struct kvm_cpuid_array array = {
1091 		.nent = 0,
1092 	};
1093 	int r, i;
1094 
1095 	if (cpuid->nent < 1)
1096 		return -E2BIG;
1097 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1098 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1099 
1100 	if (sanity_check_entries(entries, cpuid->nent, type))
1101 		return -EINVAL;
1102 
1103 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
1104 					   cpuid->nent));
1105 	if (!array.entries)
1106 		return -ENOMEM;
1107 
1108 	array.maxnent = cpuid->nent;
1109 
1110 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
1111 		r = get_cpuid_func(&array, funcs[i], type);
1112 		if (r)
1113 			goto out_free;
1114 	}
1115 	cpuid->nent = array.nent;
1116 
1117 	if (copy_to_user(entries, array.entries,
1118 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
1119 		r = -EFAULT;
1120 
1121 out_free:
1122 	vfree(array.entries);
1123 	return r;
1124 }
1125 
1126 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
1127 					      u32 function, u32 index)
1128 {
1129 	return cpuid_entry2_find(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent,
1130 				 function, index);
1131 }
1132 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
1133 
1134 /*
1135  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
1136  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
1137  * returns all zeroes for any undefined leaf, whether or not the leaf is in
1138  * range.  Centaur/VIA follows Intel semantics.
1139  *
1140  * A leaf is considered out-of-range if its function is higher than the maximum
1141  * supported leaf of its associated class or if its associated class does not
1142  * exist.
1143  *
1144  * There are three primary classes to be considered, with their respective
1145  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
1146  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
1147  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
1148  *
1149  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
1150  *  - Hypervisor: 0x40000000 - 0x4fffffff
1151  *  - Extended:   0x80000000 - 0xbfffffff
1152  *  - Centaur:    0xc0000000 - 0xcfffffff
1153  *
1154  * The Hypervisor class is further subdivided into sub-classes that each act as
1155  * their own independent class associated with a 0x100 byte range.  E.g. if Qemu
1156  * is advertising support for both HyperV and KVM, the resulting Hypervisor
1157  * CPUID sub-classes are:
1158  *
1159  *  - HyperV:     0x40000000 - 0x400000ff
1160  *  - KVM:        0x40000100 - 0x400001ff
1161  */
1162 static struct kvm_cpuid_entry2 *
1163 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
1164 {
1165 	struct kvm_cpuid_entry2 *basic, *class;
1166 	u32 function = *fn_ptr;
1167 
1168 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
1169 	if (!basic)
1170 		return NULL;
1171 
1172 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
1173 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
1174 		return NULL;
1175 
1176 	if (function >= 0x40000000 && function <= 0x4fffffff)
1177 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1178 	else if (function >= 0xc0000000)
1179 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1180 	else
1181 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1182 
1183 	if (class && function <= class->eax)
1184 		return NULL;
1185 
1186 	/*
1187 	 * Leaf specific adjustments are also applied when redirecting to the
1188 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1189 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1190 	 * needs to be pulled from CPUID.0xb.1.
1191 	 */
1192 	*fn_ptr = basic->eax;
1193 
1194 	/*
1195 	 * The class does not exist or the requested function is out of range;
1196 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1197 	 * the original requested leaf is observed!
1198 	 */
1199 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1200 }
1201 
1202 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1203 	       u32 *ecx, u32 *edx, bool exact_only)
1204 {
1205 	u32 orig_function = *eax, function = *eax, index = *ecx;
1206 	struct kvm_cpuid_entry2 *entry;
1207 	bool exact, used_max_basic = false;
1208 
1209 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1210 	exact = !!entry;
1211 
1212 	if (!entry && !exact_only) {
1213 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1214 		used_max_basic = !!entry;
1215 	}
1216 
1217 	if (entry) {
1218 		*eax = entry->eax;
1219 		*ebx = entry->ebx;
1220 		*ecx = entry->ecx;
1221 		*edx = entry->edx;
1222 		if (function == 7 && index == 0) {
1223 			u64 data;
1224 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1225 			    (data & TSX_CTRL_CPUID_CLEAR))
1226 				*ebx &= ~(F(RTM) | F(HLE));
1227 		}
1228 	} else {
1229 		*eax = *ebx = *ecx = *edx = 0;
1230 		/*
1231 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1232 		 * and EDX is always the x2APIC ID, even for undefined
1233 		 * subleaves. Index 1 will exist iff the leaf is
1234 		 * implemented, so we pass through CL iff leaf 1
1235 		 * exists. EDX can be copied from any existing index.
1236 		 */
1237 		if (function == 0xb || function == 0x1f) {
1238 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1239 			if (entry) {
1240 				*ecx = index & 0xff;
1241 				*edx = entry->edx;
1242 			}
1243 		}
1244 	}
1245 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1246 			used_max_basic);
1247 	return exact;
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_cpuid);
1250 
1251 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1252 {
1253 	u32 eax, ebx, ecx, edx;
1254 
1255 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1256 		return 1;
1257 
1258 	eax = kvm_rax_read(vcpu);
1259 	ecx = kvm_rcx_read(vcpu);
1260 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1261 	kvm_rax_write(vcpu, eax);
1262 	kvm_rbx_write(vcpu, ebx);
1263 	kvm_rcx_write(vcpu, ecx);
1264 	kvm_rdx_write(vcpu, edx);
1265 	return kvm_skip_emulated_instruction(vcpu);
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1268