xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision d89775fc)
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 static int kvm_check_cpuid(struct kvm_vcpu *vcpu)
58 {
59 	struct kvm_cpuid_entry2 *best;
60 
61 	/*
62 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
63 	 * canonical address checks; exit if it is ever changed.
64 	 */
65 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
66 	if (best) {
67 		int vaddr_bits = (best->eax & 0xff00) >> 8;
68 
69 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
70 			return -EINVAL;
71 	}
72 
73 	return 0;
74 }
75 
76 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
77 {
78 	struct kvm_cpuid_entry2 *best;
79 
80 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
81 	if (best) {
82 		/* Update OSXSAVE bit */
83 		if (boot_cpu_has(X86_FEATURE_XSAVE))
84 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
85 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
86 
87 		cpuid_entry_change(best, X86_FEATURE_APIC,
88 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
89 	}
90 
91 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
92 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
93 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
94 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
95 
96 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
97 	if (best)
98 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
99 
100 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
101 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
102 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
103 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
104 
105 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
106 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
107 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
108 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
109 
110 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
111 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
112 		if (best)
113 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
114 					   vcpu->arch.ia32_misc_enable_msr &
115 					   MSR_IA32_MISC_ENABLE_MWAIT);
116 	}
117 }
118 
119 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
120 {
121 	struct kvm_lapic *apic = vcpu->arch.apic;
122 	struct kvm_cpuid_entry2 *best;
123 
124 	kvm_x86_ops.vcpu_after_set_cpuid(vcpu);
125 
126 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
127 	if (best && apic) {
128 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
129 			apic->lapic_timer.timer_mode_mask = 3 << 17;
130 		else
131 			apic->lapic_timer.timer_mode_mask = 1 << 17;
132 
133 		kvm_apic_set_version(vcpu);
134 	}
135 
136 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
137 	if (!best)
138 		vcpu->arch.guest_supported_xcr0 = 0;
139 	else
140 		vcpu->arch.guest_supported_xcr0 =
141 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
142 
143 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
144 	kvm_mmu_reset_context(vcpu);
145 
146 	kvm_pmu_refresh(vcpu);
147 	vcpu->arch.cr4_guest_rsvd_bits =
148 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
149 	kvm_x86_ops.update_exception_bitmap(vcpu);
150 }
151 
152 static int is_efer_nx(void)
153 {
154 	return host_efer & EFER_NX;
155 }
156 
157 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
158 {
159 	int i;
160 	struct kvm_cpuid_entry2 *e, *entry;
161 
162 	entry = NULL;
163 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
164 		e = &vcpu->arch.cpuid_entries[i];
165 		if (e->function == 0x80000001) {
166 			entry = e;
167 			break;
168 		}
169 	}
170 	if (entry && cpuid_entry_has(entry, X86_FEATURE_NX) && !is_efer_nx()) {
171 		cpuid_entry_clear(entry, X86_FEATURE_NX);
172 		printk(KERN_INFO "kvm: guest NX capability removed\n");
173 	}
174 }
175 
176 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
177 {
178 	struct kvm_cpuid_entry2 *best;
179 
180 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
181 	if (!best || best->eax < 0x80000008)
182 		goto not_found;
183 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
184 	if (best)
185 		return best->eax & 0xff;
186 not_found:
187 	return 36;
188 }
189 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
190 
191 /* when an old userspace process fills a new kernel module */
192 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
193 			     struct kvm_cpuid *cpuid,
194 			     struct kvm_cpuid_entry __user *entries)
195 {
196 	int r, i;
197 	struct kvm_cpuid_entry *cpuid_entries = NULL;
198 
199 	r = -E2BIG;
200 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
201 		goto out;
202 	if (cpuid->nent) {
203 		cpuid_entries = vmemdup_user(entries,
204 					     array_size(sizeof(struct kvm_cpuid_entry),
205 							cpuid->nent));
206 		if (IS_ERR(cpuid_entries)) {
207 			r = PTR_ERR(cpuid_entries);
208 			goto out;
209 		}
210 	}
211 	for (i = 0; i < cpuid->nent; i++) {
212 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
213 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
214 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
215 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
216 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
217 		vcpu->arch.cpuid_entries[i].index = 0;
218 		vcpu->arch.cpuid_entries[i].flags = 0;
219 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
220 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
221 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
222 	}
223 	vcpu->arch.cpuid_nent = cpuid->nent;
224 	r = kvm_check_cpuid(vcpu);
225 	if (r) {
226 		vcpu->arch.cpuid_nent = 0;
227 		kvfree(cpuid_entries);
228 		goto out;
229 	}
230 
231 	cpuid_fix_nx_cap(vcpu);
232 	kvm_update_cpuid_runtime(vcpu);
233 	kvm_vcpu_after_set_cpuid(vcpu);
234 
235 	kvfree(cpuid_entries);
236 out:
237 	return r;
238 }
239 
240 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
241 			      struct kvm_cpuid2 *cpuid,
242 			      struct kvm_cpuid_entry2 __user *entries)
243 {
244 	int r;
245 
246 	r = -E2BIG;
247 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
248 		goto out;
249 	r = -EFAULT;
250 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
251 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
252 		goto out;
253 	vcpu->arch.cpuid_nent = cpuid->nent;
254 	r = kvm_check_cpuid(vcpu);
255 	if (r) {
256 		vcpu->arch.cpuid_nent = 0;
257 		goto out;
258 	}
259 
260 	kvm_update_cpuid_runtime(vcpu);
261 	kvm_vcpu_after_set_cpuid(vcpu);
262 out:
263 	return r;
264 }
265 
266 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
267 			      struct kvm_cpuid2 *cpuid,
268 			      struct kvm_cpuid_entry2 __user *entries)
269 {
270 	int r;
271 
272 	r = -E2BIG;
273 	if (cpuid->nent < vcpu->arch.cpuid_nent)
274 		goto out;
275 	r = -EFAULT;
276 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
277 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
278 		goto out;
279 	return 0;
280 
281 out:
282 	cpuid->nent = vcpu->arch.cpuid_nent;
283 	return r;
284 }
285 
286 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
287 {
288 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
289 	struct kvm_cpuid_entry2 entry;
290 
291 	reverse_cpuid_check(leaf);
292 	kvm_cpu_caps[leaf] &= mask;
293 
294 	cpuid_count(cpuid.function, cpuid.index,
295 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
296 
297 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
298 }
299 
300 void kvm_set_cpu_caps(void)
301 {
302 	unsigned int f_nx = is_efer_nx() ? F(NX) : 0;
303 #ifdef CONFIG_X86_64
304 	unsigned int f_gbpages = F(GBPAGES);
305 	unsigned int f_lm = F(LM);
306 #else
307 	unsigned int f_gbpages = 0;
308 	unsigned int f_lm = 0;
309 #endif
310 
311 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) >
312 		     sizeof(boot_cpu_data.x86_capability));
313 
314 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
315 	       sizeof(kvm_cpu_caps));
316 
317 	kvm_cpu_cap_mask(CPUID_1_ECX,
318 		/*
319 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
320 		 * advertised to guests via CPUID!
321 		 */
322 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
323 		0 /* DS-CPL, VMX, SMX, EST */ |
324 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
325 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
326 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
327 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
328 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
329 		F(F16C) | F(RDRAND)
330 	);
331 	/* KVM emulates x2apic in software irrespective of host support. */
332 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
333 
334 	kvm_cpu_cap_mask(CPUID_1_EDX,
335 		F(FPU) | F(VME) | F(DE) | F(PSE) |
336 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
337 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
338 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
339 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
340 		0 /* Reserved, DS, ACPI */ | F(MMX) |
341 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
342 		0 /* HTT, TM, Reserved, PBE */
343 	);
344 
345 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
346 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
347 		F(BMI2) | F(ERMS) | 0 /*INVPCID*/ | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
348 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
349 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
350 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
351 	);
352 
353 	kvm_cpu_cap_mask(CPUID_7_ECX,
354 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
355 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
356 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
357 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/
358 	);
359 	/* Set LA57 based on hardware capability. */
360 	if (cpuid_ecx(7) & F(LA57))
361 		kvm_cpu_cap_set(X86_FEATURE_LA57);
362 
363 	/*
364 	 * PKU not yet implemented for shadow paging and requires OSPKE
365 	 * to be set on the host. Clear it if that is not the case
366 	 */
367 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
368 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
369 
370 	kvm_cpu_cap_mask(CPUID_7_EDX,
371 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
372 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
373 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM)
374 	);
375 
376 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
377 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
378 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
379 
380 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
381 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
382 	if (boot_cpu_has(X86_FEATURE_STIBP))
383 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
384 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
385 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
386 
387 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
388 		F(AVX512_BF16)
389 	);
390 
391 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
392 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
393 	);
394 
395 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
396 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
397 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
398 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
399 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
400 		F(TOPOEXT) | F(PERFCTR_CORE)
401 	);
402 
403 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
404 		F(FPU) | F(VME) | F(DE) | F(PSE) |
405 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
406 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
407 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
408 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
409 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
410 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
411 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
412 	);
413 
414 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
415 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
416 
417 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
418 		F(CLZERO) | F(XSAVEERPTR) |
419 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
420 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON)
421 	);
422 
423 	/*
424 	 * AMD has separate bits for each SPEC_CTRL bit.
425 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
426 	 * record that in cpufeatures so use them.
427 	 */
428 	if (boot_cpu_has(X86_FEATURE_IBPB))
429 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
430 	if (boot_cpu_has(X86_FEATURE_IBRS))
431 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
432 	if (boot_cpu_has(X86_FEATURE_STIBP))
433 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
434 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
435 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
436 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
437 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
438 	/*
439 	 * The preference is to use SPEC CTRL MSR instead of the
440 	 * VIRT_SPEC MSR.
441 	 */
442 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
443 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
444 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
445 
446 	/*
447 	 * Hide all SVM features by default, SVM will set the cap bits for
448 	 * features it emulates and/or exposes for L1.
449 	 */
450 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
451 
452 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
453 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
454 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
455 		F(PMM) | F(PMM_EN)
456 	);
457 }
458 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
459 
460 struct kvm_cpuid_array {
461 	struct kvm_cpuid_entry2 *entries;
462 	int maxnent;
463 	int nent;
464 };
465 
466 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
467 					      u32 function, u32 index)
468 {
469 	struct kvm_cpuid_entry2 *entry;
470 
471 	if (array->nent >= array->maxnent)
472 		return NULL;
473 
474 	entry = &array->entries[array->nent++];
475 
476 	entry->function = function;
477 	entry->index = index;
478 	entry->flags = 0;
479 
480 	cpuid_count(entry->function, entry->index,
481 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
482 
483 	switch (function) {
484 	case 4:
485 	case 7:
486 	case 0xb:
487 	case 0xd:
488 	case 0xf:
489 	case 0x10:
490 	case 0x12:
491 	case 0x14:
492 	case 0x17:
493 	case 0x18:
494 	case 0x1f:
495 	case 0x8000001d:
496 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
497 		break;
498 	}
499 
500 	return entry;
501 }
502 
503 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
504 {
505 	struct kvm_cpuid_entry2 *entry;
506 
507 	if (array->nent >= array->maxnent)
508 		return -E2BIG;
509 
510 	entry = &array->entries[array->nent];
511 	entry->function = func;
512 	entry->index = 0;
513 	entry->flags = 0;
514 
515 	switch (func) {
516 	case 0:
517 		entry->eax = 7;
518 		++array->nent;
519 		break;
520 	case 1:
521 		entry->ecx = F(MOVBE);
522 		++array->nent;
523 		break;
524 	case 7:
525 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
526 		entry->eax = 0;
527 		entry->ecx = F(RDPID);
528 		++array->nent;
529 	default:
530 		break;
531 	}
532 
533 	return 0;
534 }
535 
536 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
537 {
538 	struct kvm_cpuid_entry2 *entry;
539 	int r, i, max_idx;
540 
541 	/* all calls to cpuid_count() should be made on the same cpu */
542 	get_cpu();
543 
544 	r = -E2BIG;
545 
546 	entry = do_host_cpuid(array, function, 0);
547 	if (!entry)
548 		goto out;
549 
550 	switch (function) {
551 	case 0:
552 		/* Limited to the highest leaf implemented in KVM. */
553 		entry->eax = min(entry->eax, 0x1fU);
554 		break;
555 	case 1:
556 		cpuid_entry_override(entry, CPUID_1_EDX);
557 		cpuid_entry_override(entry, CPUID_1_ECX);
558 		break;
559 	case 2:
560 		/*
561 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
562 		 * CPUID(function=2, index=0) may return different results each
563 		 * time, with the least-significant byte in EAX enumerating the
564 		 * number of times software should do CPUID(2, 0).
565 		 *
566 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
567 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
568 		 * return 01H. Software should ignore this value and not
569 		 * interpret it as an informational descriptor", while AMD's
570 		 * APM states that CPUID(2) is reserved.
571 		 *
572 		 * WARN if a frankenstein CPU that supports virtualization and
573 		 * a stateful CPUID.0x2 is encountered.
574 		 */
575 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
576 		break;
577 	/* functions 4 and 0x8000001d have additional index. */
578 	case 4:
579 	case 0x8000001d:
580 		/*
581 		 * Read entries until the cache type in the previous entry is
582 		 * zero, i.e. indicates an invalid entry.
583 		 */
584 		for (i = 1; entry->eax & 0x1f; ++i) {
585 			entry = do_host_cpuid(array, function, i);
586 			if (!entry)
587 				goto out;
588 		}
589 		break;
590 	case 6: /* Thermal management */
591 		entry->eax = 0x4; /* allow ARAT */
592 		entry->ebx = 0;
593 		entry->ecx = 0;
594 		entry->edx = 0;
595 		break;
596 	/* function 7 has additional index. */
597 	case 7:
598 		entry->eax = min(entry->eax, 1u);
599 		cpuid_entry_override(entry, CPUID_7_0_EBX);
600 		cpuid_entry_override(entry, CPUID_7_ECX);
601 		cpuid_entry_override(entry, CPUID_7_EDX);
602 
603 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
604 		if (entry->eax == 1) {
605 			entry = do_host_cpuid(array, function, 1);
606 			if (!entry)
607 				goto out;
608 
609 			cpuid_entry_override(entry, CPUID_7_1_EAX);
610 			entry->ebx = 0;
611 			entry->ecx = 0;
612 			entry->edx = 0;
613 		}
614 		break;
615 	case 9:
616 		break;
617 	case 0xa: { /* Architectural Performance Monitoring */
618 		struct x86_pmu_capability cap;
619 		union cpuid10_eax eax;
620 		union cpuid10_edx edx;
621 
622 		perf_get_x86_pmu_capability(&cap);
623 
624 		/*
625 		 * Only support guest architectural pmu on a host
626 		 * with architectural pmu.
627 		 */
628 		if (!cap.version)
629 			memset(&cap, 0, sizeof(cap));
630 
631 		eax.split.version_id = min(cap.version, 2);
632 		eax.split.num_counters = cap.num_counters_gp;
633 		eax.split.bit_width = cap.bit_width_gp;
634 		eax.split.mask_length = cap.events_mask_len;
635 
636 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
637 		edx.split.bit_width_fixed = cap.bit_width_fixed;
638 		edx.split.reserved = 0;
639 
640 		entry->eax = eax.full;
641 		entry->ebx = cap.events_mask;
642 		entry->ecx = 0;
643 		entry->edx = edx.full;
644 		break;
645 	}
646 	/*
647 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
648 	 * thus they can be handled by common code.
649 	 */
650 	case 0x1f:
651 	case 0xb:
652 		/*
653 		 * Populate entries until the level type (ECX[15:8]) of the
654 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
655 		 * the starting entry, filled by the primary do_host_cpuid().
656 		 */
657 		for (i = 1; entry->ecx & 0xff00; ++i) {
658 			entry = do_host_cpuid(array, function, i);
659 			if (!entry)
660 				goto out;
661 		}
662 		break;
663 	case 0xd:
664 		entry->eax &= supported_xcr0;
665 		entry->ebx = xstate_required_size(supported_xcr0, false);
666 		entry->ecx = entry->ebx;
667 		entry->edx &= supported_xcr0 >> 32;
668 		if (!supported_xcr0)
669 			break;
670 
671 		entry = do_host_cpuid(array, function, 1);
672 		if (!entry)
673 			goto out;
674 
675 		cpuid_entry_override(entry, CPUID_D_1_EAX);
676 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
677 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
678 							  true);
679 		else {
680 			WARN_ON_ONCE(supported_xss != 0);
681 			entry->ebx = 0;
682 		}
683 		entry->ecx &= supported_xss;
684 		entry->edx &= supported_xss >> 32;
685 
686 		for (i = 2; i < 64; ++i) {
687 			bool s_state;
688 			if (supported_xcr0 & BIT_ULL(i))
689 				s_state = false;
690 			else if (supported_xss & BIT_ULL(i))
691 				s_state = true;
692 			else
693 				continue;
694 
695 			entry = do_host_cpuid(array, function, i);
696 			if (!entry)
697 				goto out;
698 
699 			/*
700 			 * The supported check above should have filtered out
701 			 * invalid sub-leafs.  Only valid sub-leafs should
702 			 * reach this point, and they should have a non-zero
703 			 * save state size.  Furthermore, check whether the
704 			 * processor agrees with supported_xcr0/supported_xss
705 			 * on whether this is an XCR0- or IA32_XSS-managed area.
706 			 */
707 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
708 				--array->nent;
709 				continue;
710 			}
711 			entry->edx = 0;
712 		}
713 		break;
714 	/* Intel PT */
715 	case 0x14:
716 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
717 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
718 			break;
719 		}
720 
721 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
722 			if (!do_host_cpuid(array, function, i))
723 				goto out;
724 		}
725 		break;
726 	case KVM_CPUID_SIGNATURE: {
727 		static const char signature[12] = "KVMKVMKVM\0\0";
728 		const u32 *sigptr = (const u32 *)signature;
729 		entry->eax = KVM_CPUID_FEATURES;
730 		entry->ebx = sigptr[0];
731 		entry->ecx = sigptr[1];
732 		entry->edx = sigptr[2];
733 		break;
734 	}
735 	case KVM_CPUID_FEATURES:
736 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
737 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
738 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
739 			     (1 << KVM_FEATURE_ASYNC_PF) |
740 			     (1 << KVM_FEATURE_PV_EOI) |
741 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
742 			     (1 << KVM_FEATURE_PV_UNHALT) |
743 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
744 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
745 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
746 			     (1 << KVM_FEATURE_POLL_CONTROL) |
747 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
748 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
749 
750 		if (sched_info_on())
751 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
752 
753 		entry->ebx = 0;
754 		entry->ecx = 0;
755 		entry->edx = 0;
756 		break;
757 	case 0x80000000:
758 		entry->eax = min(entry->eax, 0x8000001f);
759 		break;
760 	case 0x80000001:
761 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
762 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
763 		break;
764 	case 0x80000006:
765 		/* L2 cache and TLB: pass through host info. */
766 		break;
767 	case 0x80000007: /* Advanced power management */
768 		/* invariant TSC is CPUID.80000007H:EDX[8] */
769 		entry->edx &= (1 << 8);
770 		/* mask against host */
771 		entry->edx &= boot_cpu_data.x86_power;
772 		entry->eax = entry->ebx = entry->ecx = 0;
773 		break;
774 	case 0x80000008: {
775 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
776 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
777 		unsigned phys_as = entry->eax & 0xff;
778 
779 		if (!g_phys_as)
780 			g_phys_as = phys_as;
781 		entry->eax = g_phys_as | (virt_as << 8);
782 		entry->edx = 0;
783 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
784 		break;
785 	}
786 	case 0x8000000A:
787 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
788 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
789 			break;
790 		}
791 		entry->eax = 1; /* SVM revision 1 */
792 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
793 				   ASID emulation to nested SVM */
794 		entry->ecx = 0; /* Reserved */
795 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
796 		break;
797 	case 0x80000019:
798 		entry->ecx = entry->edx = 0;
799 		break;
800 	case 0x8000001a:
801 	case 0x8000001e:
802 		break;
803 	/* Support memory encryption cpuid if host supports it */
804 	case 0x8000001F:
805 		if (!boot_cpu_has(X86_FEATURE_SEV))
806 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
807 		break;
808 	/*Add support for Centaur's CPUID instruction*/
809 	case 0xC0000000:
810 		/*Just support up to 0xC0000004 now*/
811 		entry->eax = min(entry->eax, 0xC0000004);
812 		break;
813 	case 0xC0000001:
814 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
815 		break;
816 	case 3: /* Processor serial number */
817 	case 5: /* MONITOR/MWAIT */
818 	case 0xC0000002:
819 	case 0xC0000003:
820 	case 0xC0000004:
821 	default:
822 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
823 		break;
824 	}
825 
826 	r = 0;
827 
828 out:
829 	put_cpu();
830 
831 	return r;
832 }
833 
834 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
835 			 unsigned int type)
836 {
837 	if (type == KVM_GET_EMULATED_CPUID)
838 		return __do_cpuid_func_emulated(array, func);
839 
840 	return __do_cpuid_func(array, func);
841 }
842 
843 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
844 
845 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
846 			  unsigned int type)
847 {
848 	u32 limit;
849 	int r;
850 
851 	if (func == CENTAUR_CPUID_SIGNATURE &&
852 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
853 		return 0;
854 
855 	r = do_cpuid_func(array, func, type);
856 	if (r)
857 		return r;
858 
859 	limit = array->entries[array->nent - 1].eax;
860 	for (func = func + 1; func <= limit; ++func) {
861 		r = do_cpuid_func(array, func, type);
862 		if (r)
863 			break;
864 	}
865 
866 	return r;
867 }
868 
869 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
870 				 __u32 num_entries, unsigned int ioctl_type)
871 {
872 	int i;
873 	__u32 pad[3];
874 
875 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
876 		return false;
877 
878 	/*
879 	 * We want to make sure that ->padding is being passed clean from
880 	 * userspace in case we want to use it for something in the future.
881 	 *
882 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
883 	 * have to give ourselves satisfied only with the emulated side. /me
884 	 * sheds a tear.
885 	 */
886 	for (i = 0; i < num_entries; i++) {
887 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
888 			return true;
889 
890 		if (pad[0] || pad[1] || pad[2])
891 			return true;
892 	}
893 	return false;
894 }
895 
896 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
897 			    struct kvm_cpuid_entry2 __user *entries,
898 			    unsigned int type)
899 {
900 	static const u32 funcs[] = {
901 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
902 	};
903 
904 	struct kvm_cpuid_array array = {
905 		.nent = 0,
906 	};
907 	int r, i;
908 
909 	if (cpuid->nent < 1)
910 		return -E2BIG;
911 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
912 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
913 
914 	if (sanity_check_entries(entries, cpuid->nent, type))
915 		return -EINVAL;
916 
917 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
918 					   cpuid->nent));
919 	if (!array.entries)
920 		return -ENOMEM;
921 
922 	array.maxnent = cpuid->nent;
923 
924 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
925 		r = get_cpuid_func(&array, funcs[i], type);
926 		if (r)
927 			goto out_free;
928 	}
929 	cpuid->nent = array.nent;
930 
931 	if (copy_to_user(entries, array.entries,
932 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
933 		r = -EFAULT;
934 
935 out_free:
936 	vfree(array.entries);
937 	return r;
938 }
939 
940 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
941 					      u32 function, u32 index)
942 {
943 	struct kvm_cpuid_entry2 *e;
944 	int i;
945 
946 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
947 		e = &vcpu->arch.cpuid_entries[i];
948 
949 		if (e->function == function && (e->index == index ||
950 		    !(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX)))
951 			return e;
952 	}
953 	return NULL;
954 }
955 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
956 
957 /*
958  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
959  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
960  * returns all zeroes for any undefined leaf, whether or not the leaf is in
961  * range.  Centaur/VIA follows Intel semantics.
962  *
963  * A leaf is considered out-of-range if its function is higher than the maximum
964  * supported leaf of its associated class or if its associated class does not
965  * exist.
966  *
967  * There are three primary classes to be considered, with their respective
968  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
969  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
970  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
971  *
972  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
973  *  - Hypervisor: 0x40000000 - 0x4fffffff
974  *  - Extended:   0x80000000 - 0xbfffffff
975  *  - Centaur:    0xc0000000 - 0xcfffffff
976  *
977  * The Hypervisor class is further subdivided into sub-classes that each act as
978  * their own indepdent class associated with a 0x100 byte range.  E.g. if Qemu
979  * is advertising support for both HyperV and KVM, the resulting Hypervisor
980  * CPUID sub-classes are:
981  *
982  *  - HyperV:     0x40000000 - 0x400000ff
983  *  - KVM:        0x40000100 - 0x400001ff
984  */
985 static struct kvm_cpuid_entry2 *
986 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
987 {
988 	struct kvm_cpuid_entry2 *basic, *class;
989 	u32 function = *fn_ptr;
990 
991 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
992 	if (!basic)
993 		return NULL;
994 
995 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
996 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
997 		return NULL;
998 
999 	if (function >= 0x40000000 && function <= 0x4fffffff)
1000 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1001 	else if (function >= 0xc0000000)
1002 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1003 	else
1004 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1005 
1006 	if (class && function <= class->eax)
1007 		return NULL;
1008 
1009 	/*
1010 	 * Leaf specific adjustments are also applied when redirecting to the
1011 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1012 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1013 	 * needs to be pulled from CPUID.0xb.1.
1014 	 */
1015 	*fn_ptr = basic->eax;
1016 
1017 	/*
1018 	 * The class does not exist or the requested function is out of range;
1019 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1020 	 * the original requested leaf is observed!
1021 	 */
1022 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1023 }
1024 
1025 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1026 	       u32 *ecx, u32 *edx, bool exact_only)
1027 {
1028 	u32 orig_function = *eax, function = *eax, index = *ecx;
1029 	struct kvm_cpuid_entry2 *entry;
1030 	bool exact, used_max_basic = false;
1031 
1032 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1033 	exact = !!entry;
1034 
1035 	if (!entry && !exact_only) {
1036 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1037 		used_max_basic = !!entry;
1038 	}
1039 
1040 	if (entry) {
1041 		*eax = entry->eax;
1042 		*ebx = entry->ebx;
1043 		*ecx = entry->ecx;
1044 		*edx = entry->edx;
1045 		if (function == 7 && index == 0) {
1046 			u64 data;
1047 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1048 			    (data & TSX_CTRL_CPUID_CLEAR))
1049 				*ebx &= ~(F(RTM) | F(HLE));
1050 		}
1051 	} else {
1052 		*eax = *ebx = *ecx = *edx = 0;
1053 		/*
1054 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1055 		 * and EDX is always the x2APIC ID, even for undefined
1056 		 * subleaves. Index 1 will exist iff the leaf is
1057 		 * implemented, so we pass through CL iff leaf 1
1058 		 * exists. EDX can be copied from any existing index.
1059 		 */
1060 		if (function == 0xb || function == 0x1f) {
1061 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1062 			if (entry) {
1063 				*ecx = index & 0xff;
1064 				*edx = entry->edx;
1065 			}
1066 		}
1067 	}
1068 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1069 			used_max_basic);
1070 	return exact;
1071 }
1072 EXPORT_SYMBOL_GPL(kvm_cpuid);
1073 
1074 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1075 {
1076 	u32 eax, ebx, ecx, edx;
1077 
1078 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1079 		return 1;
1080 
1081 	eax = kvm_rax_read(vcpu);
1082 	ecx = kvm_rcx_read(vcpu);
1083 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1084 	kvm_rax_write(vcpu, eax);
1085 	kvm_rbx_write(vcpu, ebx);
1086 	kvm_rcx_write(vcpu, ecx);
1087 	kvm_rdx_write(vcpu, edx);
1088 	return kvm_skip_emulated_instruction(vcpu);
1089 }
1090 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1091