xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 800e26b8)
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 "cpuid.h"
22 #include "lapic.h"
23 #include "mmu.h"
24 #include "trace.h"
25 #include "pmu.h"
26 
27 /*
28  * Unlike "struct cpuinfo_x86.x86_capability", kvm_cpu_caps doesn't need to be
29  * aligned to sizeof(unsigned long) because it's not accessed via bitops.
30  */
31 u32 kvm_cpu_caps[NCAPINTS] __read_mostly;
32 EXPORT_SYMBOL_GPL(kvm_cpu_caps);
33 
34 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
35 {
36 	int feature_bit = 0;
37 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
38 
39 	xstate_bv &= XFEATURE_MASK_EXTEND;
40 	while (xstate_bv) {
41 		if (xstate_bv & 0x1) {
42 		        u32 eax, ebx, ecx, edx, offset;
43 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
44 			offset = compacted ? ret : ebx;
45 			ret = max(ret, offset + eax);
46 		}
47 
48 		xstate_bv >>= 1;
49 		feature_bit++;
50 	}
51 
52 	return ret;
53 }
54 
55 #define F feature_bit
56 
57 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
58 {
59 	struct kvm_cpuid_entry2 *best;
60 	struct kvm_lapic *apic = vcpu->arch.apic;
61 
62 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
63 	if (!best)
64 		return 0;
65 
66 	/* Update OSXSAVE bit */
67 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1)
68 		cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
69 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
70 
71 	cpuid_entry_change(best, X86_FEATURE_APIC,
72 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
73 
74 	if (apic) {
75 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
76 			apic->lapic_timer.timer_mode_mask = 3 << 17;
77 		else
78 			apic->lapic_timer.timer_mode_mask = 1 << 17;
79 	}
80 
81 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
82 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
83 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
84 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
85 
86 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
87 	if (!best) {
88 		vcpu->arch.guest_supported_xcr0 = 0;
89 	} else {
90 		vcpu->arch.guest_supported_xcr0 =
91 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
92 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
93 	}
94 
95 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
96 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
97 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
98 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
99 
100 	/*
101 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
102 	 * canonical address checks; exit if it is ever changed.
103 	 */
104 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
105 	if (best) {
106 		int vaddr_bits = (best->eax & 0xff00) >> 8;
107 
108 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
109 			return -EINVAL;
110 	}
111 
112 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
113 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
114 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
115 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
116 
117 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
118 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
119 		if (best)
120 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
121 					   vcpu->arch.ia32_misc_enable_msr &
122 					   MSR_IA32_MISC_ENABLE_MWAIT);
123 	}
124 
125 	/* Note, maxphyaddr must be updated before tdp_level. */
126 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
127 	vcpu->arch.tdp_level = kvm_x86_ops.get_tdp_level(vcpu);
128 	kvm_mmu_reset_context(vcpu);
129 
130 	kvm_pmu_refresh(vcpu);
131 	return 0;
132 }
133 
134 static int is_efer_nx(void)
135 {
136 	return host_efer & EFER_NX;
137 }
138 
139 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
140 {
141 	int i;
142 	struct kvm_cpuid_entry2 *e, *entry;
143 
144 	entry = NULL;
145 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
146 		e = &vcpu->arch.cpuid_entries[i];
147 		if (e->function == 0x80000001) {
148 			entry = e;
149 			break;
150 		}
151 	}
152 	if (entry && cpuid_entry_has(entry, X86_FEATURE_NX) && !is_efer_nx()) {
153 		cpuid_entry_clear(entry, X86_FEATURE_NX);
154 		printk(KERN_INFO "kvm: guest NX capability removed\n");
155 	}
156 }
157 
158 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
159 {
160 	struct kvm_cpuid_entry2 *best;
161 
162 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
163 	if (!best || best->eax < 0x80000008)
164 		goto not_found;
165 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
166 	if (best)
167 		return best->eax & 0xff;
168 not_found:
169 	return 36;
170 }
171 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
172 
173 /* when an old userspace process fills a new kernel module */
174 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
175 			     struct kvm_cpuid *cpuid,
176 			     struct kvm_cpuid_entry __user *entries)
177 {
178 	int r, i;
179 	struct kvm_cpuid_entry *cpuid_entries = NULL;
180 
181 	r = -E2BIG;
182 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
183 		goto out;
184 	if (cpuid->nent) {
185 		cpuid_entries = vmemdup_user(entries,
186 					     array_size(sizeof(struct kvm_cpuid_entry),
187 							cpuid->nent));
188 		if (IS_ERR(cpuid_entries)) {
189 			r = PTR_ERR(cpuid_entries);
190 			goto out;
191 		}
192 	}
193 	for (i = 0; i < cpuid->nent; i++) {
194 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
195 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
196 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
197 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
198 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
199 		vcpu->arch.cpuid_entries[i].index = 0;
200 		vcpu->arch.cpuid_entries[i].flags = 0;
201 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
202 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
203 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
204 	}
205 	vcpu->arch.cpuid_nent = cpuid->nent;
206 	cpuid_fix_nx_cap(vcpu);
207 	kvm_apic_set_version(vcpu);
208 	kvm_x86_ops.cpuid_update(vcpu);
209 	r = kvm_update_cpuid(vcpu);
210 
211 	kvfree(cpuid_entries);
212 out:
213 	return r;
214 }
215 
216 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
217 			      struct kvm_cpuid2 *cpuid,
218 			      struct kvm_cpuid_entry2 __user *entries)
219 {
220 	int r;
221 
222 	r = -E2BIG;
223 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
224 		goto out;
225 	r = -EFAULT;
226 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
227 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
228 		goto out;
229 	vcpu->arch.cpuid_nent = cpuid->nent;
230 	kvm_apic_set_version(vcpu);
231 	kvm_x86_ops.cpuid_update(vcpu);
232 	r = kvm_update_cpuid(vcpu);
233 out:
234 	return r;
235 }
236 
237 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
238 			      struct kvm_cpuid2 *cpuid,
239 			      struct kvm_cpuid_entry2 __user *entries)
240 {
241 	int r;
242 
243 	r = -E2BIG;
244 	if (cpuid->nent < vcpu->arch.cpuid_nent)
245 		goto out;
246 	r = -EFAULT;
247 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
248 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
249 		goto out;
250 	return 0;
251 
252 out:
253 	cpuid->nent = vcpu->arch.cpuid_nent;
254 	return r;
255 }
256 
257 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
258 {
259 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
260 	struct kvm_cpuid_entry2 entry;
261 
262 	reverse_cpuid_check(leaf);
263 	kvm_cpu_caps[leaf] &= mask;
264 
265 	cpuid_count(cpuid.function, cpuid.index,
266 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
267 
268 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
269 }
270 
271 void kvm_set_cpu_caps(void)
272 {
273 	unsigned int f_nx = is_efer_nx() ? F(NX) : 0;
274 #ifdef CONFIG_X86_64
275 	unsigned int f_gbpages = F(GBPAGES);
276 	unsigned int f_lm = F(LM);
277 #else
278 	unsigned int f_gbpages = 0;
279 	unsigned int f_lm = 0;
280 #endif
281 
282 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) >
283 		     sizeof(boot_cpu_data.x86_capability));
284 
285 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
286 	       sizeof(kvm_cpu_caps));
287 
288 	kvm_cpu_cap_mask(CPUID_1_ECX,
289 		/*
290 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
291 		 * advertised to guests via CPUID!
292 		 */
293 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
294 		0 /* DS-CPL, VMX, SMX, EST */ |
295 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
296 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
297 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
298 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
299 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
300 		F(F16C) | F(RDRAND)
301 	);
302 	/* KVM emulates x2apic in software irrespective of host support. */
303 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
304 
305 	kvm_cpu_cap_mask(CPUID_1_EDX,
306 		F(FPU) | F(VME) | F(DE) | F(PSE) |
307 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
308 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
309 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
310 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
311 		0 /* Reserved, DS, ACPI */ | F(MMX) |
312 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
313 		0 /* HTT, TM, Reserved, PBE */
314 	);
315 
316 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
317 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
318 		F(BMI2) | F(ERMS) | 0 /*INVPCID*/ | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
319 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
320 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
321 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
322 	);
323 
324 	kvm_cpu_cap_mask(CPUID_7_ECX,
325 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
326 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
327 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
328 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/
329 	);
330 	/* Set LA57 based on hardware capability. */
331 	if (cpuid_ecx(7) & F(LA57))
332 		kvm_cpu_cap_set(X86_FEATURE_LA57);
333 
334 	/*
335 	 * PKU not yet implemented for shadow paging and requires OSPKE
336 	 * to be set on the host. Clear it if that is not the case
337 	 */
338 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
339 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
340 
341 	kvm_cpu_cap_mask(CPUID_7_EDX,
342 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
343 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
344 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM)
345 	);
346 
347 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
348 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
349 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
350 
351 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
352 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
353 	if (boot_cpu_has(X86_FEATURE_STIBP))
354 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
355 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
356 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
357 
358 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
359 		F(AVX512_BF16)
360 	);
361 
362 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
363 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
364 	);
365 
366 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
367 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
368 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
369 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
370 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
371 		F(TOPOEXT) | F(PERFCTR_CORE)
372 	);
373 
374 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
375 		F(FPU) | F(VME) | F(DE) | F(PSE) |
376 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
377 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
378 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
379 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
380 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
381 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
382 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
383 	);
384 
385 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
386 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
387 
388 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
389 		F(CLZERO) | F(XSAVEERPTR) |
390 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
391 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON)
392 	);
393 
394 	/*
395 	 * AMD has separate bits for each SPEC_CTRL bit.
396 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
397 	 * record that in cpufeatures so use them.
398 	 */
399 	if (boot_cpu_has(X86_FEATURE_IBPB))
400 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
401 	if (boot_cpu_has(X86_FEATURE_IBRS))
402 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
403 	if (boot_cpu_has(X86_FEATURE_STIBP))
404 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
405 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
406 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
407 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
408 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
409 	/*
410 	 * The preference is to use SPEC CTRL MSR instead of the
411 	 * VIRT_SPEC MSR.
412 	 */
413 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
414 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
415 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
416 
417 	/*
418 	 * Hide all SVM features by default, SVM will set the cap bits for
419 	 * features it emulates and/or exposes for L1.
420 	 */
421 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
422 
423 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
424 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
425 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
426 		F(PMM) | F(PMM_EN)
427 	);
428 }
429 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
430 
431 struct kvm_cpuid_array {
432 	struct kvm_cpuid_entry2 *entries;
433 	int maxnent;
434 	int nent;
435 };
436 
437 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
438 					      u32 function, u32 index)
439 {
440 	struct kvm_cpuid_entry2 *entry;
441 
442 	if (array->nent >= array->maxnent)
443 		return NULL;
444 
445 	entry = &array->entries[array->nent++];
446 
447 	entry->function = function;
448 	entry->index = index;
449 	entry->flags = 0;
450 
451 	cpuid_count(entry->function, entry->index,
452 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
453 
454 	switch (function) {
455 	case 4:
456 	case 7:
457 	case 0xb:
458 	case 0xd:
459 	case 0xf:
460 	case 0x10:
461 	case 0x12:
462 	case 0x14:
463 	case 0x17:
464 	case 0x18:
465 	case 0x1f:
466 	case 0x8000001d:
467 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
468 		break;
469 	}
470 
471 	return entry;
472 }
473 
474 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
475 {
476 	struct kvm_cpuid_entry2 *entry;
477 
478 	if (array->nent >= array->maxnent)
479 		return -E2BIG;
480 
481 	entry = &array->entries[array->nent];
482 	entry->function = func;
483 	entry->index = 0;
484 	entry->flags = 0;
485 
486 	switch (func) {
487 	case 0:
488 		entry->eax = 7;
489 		++array->nent;
490 		break;
491 	case 1:
492 		entry->ecx = F(MOVBE);
493 		++array->nent;
494 		break;
495 	case 7:
496 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
497 		entry->eax = 0;
498 		entry->ecx = F(RDPID);
499 		++array->nent;
500 	default:
501 		break;
502 	}
503 
504 	return 0;
505 }
506 
507 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
508 {
509 	struct kvm_cpuid_entry2 *entry;
510 	int r, i, max_idx;
511 
512 	/* all calls to cpuid_count() should be made on the same cpu */
513 	get_cpu();
514 
515 	r = -E2BIG;
516 
517 	entry = do_host_cpuid(array, function, 0);
518 	if (!entry)
519 		goto out;
520 
521 	switch (function) {
522 	case 0:
523 		/* Limited to the highest leaf implemented in KVM. */
524 		entry->eax = min(entry->eax, 0x1fU);
525 		break;
526 	case 1:
527 		cpuid_entry_override(entry, CPUID_1_EDX);
528 		cpuid_entry_override(entry, CPUID_1_ECX);
529 		break;
530 	case 2:
531 		/*
532 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
533 		 * CPUID(function=2, index=0) may return different results each
534 		 * time, with the least-significant byte in EAX enumerating the
535 		 * number of times software should do CPUID(2, 0).
536 		 *
537 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
538 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
539 		 * return 01H. Software should ignore this value and not
540 		 * interpret it as an informational descriptor", while AMD's
541 		 * APM states that CPUID(2) is reserved.
542 		 *
543 		 * WARN if a frankenstein CPU that supports virtualization and
544 		 * a stateful CPUID.0x2 is encountered.
545 		 */
546 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
547 		break;
548 	/* functions 4 and 0x8000001d have additional index. */
549 	case 4:
550 	case 0x8000001d:
551 		/*
552 		 * Read entries until the cache type in the previous entry is
553 		 * zero, i.e. indicates an invalid entry.
554 		 */
555 		for (i = 1; entry->eax & 0x1f; ++i) {
556 			entry = do_host_cpuid(array, function, i);
557 			if (!entry)
558 				goto out;
559 		}
560 		break;
561 	case 6: /* Thermal management */
562 		entry->eax = 0x4; /* allow ARAT */
563 		entry->ebx = 0;
564 		entry->ecx = 0;
565 		entry->edx = 0;
566 		break;
567 	/* function 7 has additional index. */
568 	case 7:
569 		entry->eax = min(entry->eax, 1u);
570 		cpuid_entry_override(entry, CPUID_7_0_EBX);
571 		cpuid_entry_override(entry, CPUID_7_ECX);
572 		cpuid_entry_override(entry, CPUID_7_EDX);
573 
574 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
575 		if (entry->eax == 1) {
576 			entry = do_host_cpuid(array, function, 1);
577 			if (!entry)
578 				goto out;
579 
580 			cpuid_entry_override(entry, CPUID_7_1_EAX);
581 			entry->ebx = 0;
582 			entry->ecx = 0;
583 			entry->edx = 0;
584 		}
585 		break;
586 	case 9:
587 		break;
588 	case 0xa: { /* Architectural Performance Monitoring */
589 		struct x86_pmu_capability cap;
590 		union cpuid10_eax eax;
591 		union cpuid10_edx edx;
592 
593 		perf_get_x86_pmu_capability(&cap);
594 
595 		/*
596 		 * Only support guest architectural pmu on a host
597 		 * with architectural pmu.
598 		 */
599 		if (!cap.version)
600 			memset(&cap, 0, sizeof(cap));
601 
602 		eax.split.version_id = min(cap.version, 2);
603 		eax.split.num_counters = cap.num_counters_gp;
604 		eax.split.bit_width = cap.bit_width_gp;
605 		eax.split.mask_length = cap.events_mask_len;
606 
607 		edx.split.num_counters_fixed = cap.num_counters_fixed;
608 		edx.split.bit_width_fixed = cap.bit_width_fixed;
609 		edx.split.reserved = 0;
610 
611 		entry->eax = eax.full;
612 		entry->ebx = cap.events_mask;
613 		entry->ecx = 0;
614 		entry->edx = edx.full;
615 		break;
616 	}
617 	/*
618 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
619 	 * thus they can be handled by common code.
620 	 */
621 	case 0x1f:
622 	case 0xb:
623 		/*
624 		 * Populate entries until the level type (ECX[15:8]) of the
625 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
626 		 * the starting entry, filled by the primary do_host_cpuid().
627 		 */
628 		for (i = 1; entry->ecx & 0xff00; ++i) {
629 			entry = do_host_cpuid(array, function, i);
630 			if (!entry)
631 				goto out;
632 		}
633 		break;
634 	case 0xd:
635 		entry->eax &= supported_xcr0;
636 		entry->ebx = xstate_required_size(supported_xcr0, false);
637 		entry->ecx = entry->ebx;
638 		entry->edx &= supported_xcr0 >> 32;
639 		if (!supported_xcr0)
640 			break;
641 
642 		entry = do_host_cpuid(array, function, 1);
643 		if (!entry)
644 			goto out;
645 
646 		cpuid_entry_override(entry, CPUID_D_1_EAX);
647 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
648 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
649 							  true);
650 		else {
651 			WARN_ON_ONCE(supported_xss != 0);
652 			entry->ebx = 0;
653 		}
654 		entry->ecx &= supported_xss;
655 		entry->edx &= supported_xss >> 32;
656 
657 		for (i = 2; i < 64; ++i) {
658 			bool s_state;
659 			if (supported_xcr0 & BIT_ULL(i))
660 				s_state = false;
661 			else if (supported_xss & BIT_ULL(i))
662 				s_state = true;
663 			else
664 				continue;
665 
666 			entry = do_host_cpuid(array, function, i);
667 			if (!entry)
668 				goto out;
669 
670 			/*
671 			 * The supported check above should have filtered out
672 			 * invalid sub-leafs.  Only valid sub-leafs should
673 			 * reach this point, and they should have a non-zero
674 			 * save state size.  Furthermore, check whether the
675 			 * processor agrees with supported_xcr0/supported_xss
676 			 * on whether this is an XCR0- or IA32_XSS-managed area.
677 			 */
678 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
679 				--array->nent;
680 				continue;
681 			}
682 			entry->edx = 0;
683 		}
684 		break;
685 	/* Intel PT */
686 	case 0x14:
687 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
688 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
689 			break;
690 		}
691 
692 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
693 			if (!do_host_cpuid(array, function, i))
694 				goto out;
695 		}
696 		break;
697 	case KVM_CPUID_SIGNATURE: {
698 		static const char signature[12] = "KVMKVMKVM\0\0";
699 		const u32 *sigptr = (const u32 *)signature;
700 		entry->eax = KVM_CPUID_FEATURES;
701 		entry->ebx = sigptr[0];
702 		entry->ecx = sigptr[1];
703 		entry->edx = sigptr[2];
704 		break;
705 	}
706 	case KVM_CPUID_FEATURES:
707 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
708 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
709 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
710 			     (1 << KVM_FEATURE_ASYNC_PF) |
711 			     (1 << KVM_FEATURE_PV_EOI) |
712 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
713 			     (1 << KVM_FEATURE_PV_UNHALT) |
714 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
715 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
716 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
717 			     (1 << KVM_FEATURE_POLL_CONTROL) |
718 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
719 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
720 
721 		if (sched_info_on())
722 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
723 
724 		entry->ebx = 0;
725 		entry->ecx = 0;
726 		entry->edx = 0;
727 		break;
728 	case 0x80000000:
729 		entry->eax = min(entry->eax, 0x8000001f);
730 		break;
731 	case 0x80000001:
732 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
733 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
734 		break;
735 	case 0x80000006:
736 		/* L2 cache and TLB: pass through host info. */
737 		break;
738 	case 0x80000007: /* Advanced power management */
739 		/* invariant TSC is CPUID.80000007H:EDX[8] */
740 		entry->edx &= (1 << 8);
741 		/* mask against host */
742 		entry->edx &= boot_cpu_data.x86_power;
743 		entry->eax = entry->ebx = entry->ecx = 0;
744 		break;
745 	case 0x80000008: {
746 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
747 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
748 		unsigned phys_as = entry->eax & 0xff;
749 
750 		if (!g_phys_as)
751 			g_phys_as = phys_as;
752 		entry->eax = g_phys_as | (virt_as << 8);
753 		entry->edx = 0;
754 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
755 		break;
756 	}
757 	case 0x8000000A:
758 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
759 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
760 			break;
761 		}
762 		entry->eax = 1; /* SVM revision 1 */
763 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
764 				   ASID emulation to nested SVM */
765 		entry->ecx = 0; /* Reserved */
766 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
767 		break;
768 	case 0x80000019:
769 		entry->ecx = entry->edx = 0;
770 		break;
771 	case 0x8000001a:
772 	case 0x8000001e:
773 		break;
774 	/* Support memory encryption cpuid if host supports it */
775 	case 0x8000001F:
776 		if (!boot_cpu_has(X86_FEATURE_SEV))
777 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
778 		break;
779 	/*Add support for Centaur's CPUID instruction*/
780 	case 0xC0000000:
781 		/*Just support up to 0xC0000004 now*/
782 		entry->eax = min(entry->eax, 0xC0000004);
783 		break;
784 	case 0xC0000001:
785 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
786 		break;
787 	case 3: /* Processor serial number */
788 	case 5: /* MONITOR/MWAIT */
789 	case 0xC0000002:
790 	case 0xC0000003:
791 	case 0xC0000004:
792 	default:
793 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
794 		break;
795 	}
796 
797 	r = 0;
798 
799 out:
800 	put_cpu();
801 
802 	return r;
803 }
804 
805 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
806 			 unsigned int type)
807 {
808 	if (type == KVM_GET_EMULATED_CPUID)
809 		return __do_cpuid_func_emulated(array, func);
810 
811 	return __do_cpuid_func(array, func);
812 }
813 
814 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
815 
816 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
817 			  unsigned int type)
818 {
819 	u32 limit;
820 	int r;
821 
822 	if (func == CENTAUR_CPUID_SIGNATURE &&
823 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
824 		return 0;
825 
826 	r = do_cpuid_func(array, func, type);
827 	if (r)
828 		return r;
829 
830 	limit = array->entries[array->nent - 1].eax;
831 	for (func = func + 1; func <= limit; ++func) {
832 		r = do_cpuid_func(array, func, type);
833 		if (r)
834 			break;
835 	}
836 
837 	return r;
838 }
839 
840 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
841 				 __u32 num_entries, unsigned int ioctl_type)
842 {
843 	int i;
844 	__u32 pad[3];
845 
846 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
847 		return false;
848 
849 	/*
850 	 * We want to make sure that ->padding is being passed clean from
851 	 * userspace in case we want to use it for something in the future.
852 	 *
853 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
854 	 * have to give ourselves satisfied only with the emulated side. /me
855 	 * sheds a tear.
856 	 */
857 	for (i = 0; i < num_entries; i++) {
858 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
859 			return true;
860 
861 		if (pad[0] || pad[1] || pad[2])
862 			return true;
863 	}
864 	return false;
865 }
866 
867 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
868 			    struct kvm_cpuid_entry2 __user *entries,
869 			    unsigned int type)
870 {
871 	static const u32 funcs[] = {
872 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
873 	};
874 
875 	struct kvm_cpuid_array array = {
876 		.nent = 0,
877 	};
878 	int r, i;
879 
880 	if (cpuid->nent < 1)
881 		return -E2BIG;
882 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
883 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
884 
885 	if (sanity_check_entries(entries, cpuid->nent, type))
886 		return -EINVAL;
887 
888 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
889 					   cpuid->nent));
890 	if (!array.entries)
891 		return -ENOMEM;
892 
893 	array.maxnent = cpuid->nent;
894 
895 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
896 		r = get_cpuid_func(&array, funcs[i], type);
897 		if (r)
898 			goto out_free;
899 	}
900 	cpuid->nent = array.nent;
901 
902 	if (copy_to_user(entries, array.entries,
903 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
904 		r = -EFAULT;
905 
906 out_free:
907 	vfree(array.entries);
908 	return r;
909 }
910 
911 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
912 					      u32 function, u32 index)
913 {
914 	struct kvm_cpuid_entry2 *e;
915 	int i;
916 
917 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
918 		e = &vcpu->arch.cpuid_entries[i];
919 
920 		if (e->function == function && (e->index == index ||
921 		    !(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX)))
922 			return e;
923 	}
924 	return NULL;
925 }
926 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
927 
928 /*
929  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
930  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
931  * returns all zeroes for any undefined leaf, whether or not the leaf is in
932  * range.  Centaur/VIA follows Intel semantics.
933  *
934  * A leaf is considered out-of-range if its function is higher than the maximum
935  * supported leaf of its associated class or if its associated class does not
936  * exist.
937  *
938  * There are three primary classes to be considered, with their respective
939  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
940  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
941  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
942  *
943  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
944  *  - Hypervisor: 0x40000000 - 0x4fffffff
945  *  - Extended:   0x80000000 - 0xbfffffff
946  *  - Centaur:    0xc0000000 - 0xcfffffff
947  *
948  * The Hypervisor class is further subdivided into sub-classes that each act as
949  * their own indepdent class associated with a 0x100 byte range.  E.g. if Qemu
950  * is advertising support for both HyperV and KVM, the resulting Hypervisor
951  * CPUID sub-classes are:
952  *
953  *  - HyperV:     0x40000000 - 0x400000ff
954  *  - KVM:        0x40000100 - 0x400001ff
955  */
956 static struct kvm_cpuid_entry2 *
957 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
958 {
959 	struct kvm_cpuid_entry2 *basic, *class;
960 	u32 function = *fn_ptr;
961 
962 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
963 	if (!basic)
964 		return NULL;
965 
966 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
967 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
968 		return NULL;
969 
970 	if (function >= 0x40000000 && function <= 0x4fffffff)
971 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
972 	else if (function >= 0xc0000000)
973 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
974 	else
975 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
976 
977 	if (class && function <= class->eax)
978 		return NULL;
979 
980 	/*
981 	 * Leaf specific adjustments are also applied when redirecting to the
982 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
983 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
984 	 * needs to be pulled from CPUID.0xb.1.
985 	 */
986 	*fn_ptr = basic->eax;
987 
988 	/*
989 	 * The class does not exist or the requested function is out of range;
990 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
991 	 * the original requested leaf is observed!
992 	 */
993 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
994 }
995 
996 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
997 	       u32 *ecx, u32 *edx, bool exact_only)
998 {
999 	u32 orig_function = *eax, function = *eax, index = *ecx;
1000 	struct kvm_cpuid_entry2 *entry;
1001 	bool exact, used_max_basic = false;
1002 
1003 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1004 	exact = !!entry;
1005 
1006 	if (!entry && !exact_only) {
1007 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1008 		used_max_basic = !!entry;
1009 	}
1010 
1011 	if (entry) {
1012 		*eax = entry->eax;
1013 		*ebx = entry->ebx;
1014 		*ecx = entry->ecx;
1015 		*edx = entry->edx;
1016 		if (function == 7 && index == 0) {
1017 			u64 data;
1018 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1019 			    (data & TSX_CTRL_CPUID_CLEAR))
1020 				*ebx &= ~(F(RTM) | F(HLE));
1021 		}
1022 	} else {
1023 		*eax = *ebx = *ecx = *edx = 0;
1024 		/*
1025 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1026 		 * and EDX is always the x2APIC ID, even for undefined
1027 		 * subleaves. Index 1 will exist iff the leaf is
1028 		 * implemented, so we pass through CL iff leaf 1
1029 		 * exists. EDX can be copied from any existing index.
1030 		 */
1031 		if (function == 0xb || function == 0x1f) {
1032 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1033 			if (entry) {
1034 				*ecx = index & 0xff;
1035 				*edx = entry->edx;
1036 			}
1037 		}
1038 	}
1039 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1040 			used_max_basic);
1041 	return exact;
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_cpuid);
1044 
1045 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1046 {
1047 	u32 eax, ebx, ecx, edx;
1048 
1049 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1050 		return 1;
1051 
1052 	eax = kvm_rax_read(vcpu);
1053 	ecx = kvm_rcx_read(vcpu);
1054 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1055 	kvm_rax_write(vcpu, eax);
1056 	kvm_rbx_write(vcpu, ebx);
1057 	kvm_rcx_write(vcpu, ecx);
1058 	kvm_rdx_write(vcpu, edx);
1059 	return kvm_skip_emulated_instruction(vcpu);
1060 }
1061 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1062