xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision adb57164)
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 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
90 	} else {
91 		vcpu->arch.guest_supported_xcr0 =
92 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
93 		vcpu->arch.guest_xstate_size = best->ebx =
94 			xstate_required_size(vcpu->arch.xcr0, false);
95 	}
96 
97 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
98 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
99 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
100 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
101 
102 	/*
103 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
104 	 * canonical address checks; exit if it is ever changed.
105 	 */
106 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
107 	if (best) {
108 		int vaddr_bits = (best->eax & 0xff00) >> 8;
109 
110 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
111 			return -EINVAL;
112 	}
113 
114 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
115 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
116 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
117 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
118 
119 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
120 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
121 		if (best)
122 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
123 					   vcpu->arch.ia32_misc_enable_msr &
124 					   MSR_IA32_MISC_ENABLE_MWAIT);
125 	}
126 
127 	/* Update physical-address width */
128 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
129 	kvm_mmu_reset_context(vcpu);
130 
131 	kvm_pmu_refresh(vcpu);
132 	return 0;
133 }
134 
135 static int is_efer_nx(void)
136 {
137 	return host_efer & EFER_NX;
138 }
139 
140 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
141 {
142 	int i;
143 	struct kvm_cpuid_entry2 *e, *entry;
144 
145 	entry = NULL;
146 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
147 		e = &vcpu->arch.cpuid_entries[i];
148 		if (e->function == 0x80000001) {
149 			entry = e;
150 			break;
151 		}
152 	}
153 	if (entry && cpuid_entry_has(entry, X86_FEATURE_NX) && !is_efer_nx()) {
154 		cpuid_entry_clear(entry, X86_FEATURE_NX);
155 		printk(KERN_INFO "kvm: guest NX capability removed\n");
156 	}
157 }
158 
159 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
160 {
161 	struct kvm_cpuid_entry2 *best;
162 
163 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
164 	if (!best || best->eax < 0x80000008)
165 		goto not_found;
166 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
167 	if (best)
168 		return best->eax & 0xff;
169 not_found:
170 	return 36;
171 }
172 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
173 
174 /* when an old userspace process fills a new kernel module */
175 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
176 			     struct kvm_cpuid *cpuid,
177 			     struct kvm_cpuid_entry __user *entries)
178 {
179 	int r, i;
180 	struct kvm_cpuid_entry *cpuid_entries = NULL;
181 
182 	r = -E2BIG;
183 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
184 		goto out;
185 	r = -ENOMEM;
186 	if (cpuid->nent) {
187 		cpuid_entries =
188 			vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
189 					   cpuid->nent));
190 		if (!cpuid_entries)
191 			goto out;
192 		r = -EFAULT;
193 		if (copy_from_user(cpuid_entries, entries,
194 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
195 			goto out;
196 	}
197 	for (i = 0; i < cpuid->nent; i++) {
198 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
199 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
200 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
201 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
202 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
203 		vcpu->arch.cpuid_entries[i].index = 0;
204 		vcpu->arch.cpuid_entries[i].flags = 0;
205 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
206 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
207 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
208 	}
209 	vcpu->arch.cpuid_nent = cpuid->nent;
210 	cpuid_fix_nx_cap(vcpu);
211 	kvm_apic_set_version(vcpu);
212 	kvm_x86_ops.cpuid_update(vcpu);
213 	r = kvm_update_cpuid(vcpu);
214 
215 out:
216 	vfree(cpuid_entries);
217 	return r;
218 }
219 
220 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
221 			      struct kvm_cpuid2 *cpuid,
222 			      struct kvm_cpuid_entry2 __user *entries)
223 {
224 	int r;
225 
226 	r = -E2BIG;
227 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
228 		goto out;
229 	r = -EFAULT;
230 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
231 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
232 		goto out;
233 	vcpu->arch.cpuid_nent = cpuid->nent;
234 	kvm_apic_set_version(vcpu);
235 	kvm_x86_ops.cpuid_update(vcpu);
236 	r = kvm_update_cpuid(vcpu);
237 out:
238 	return r;
239 }
240 
241 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
242 			      struct kvm_cpuid2 *cpuid,
243 			      struct kvm_cpuid_entry2 __user *entries)
244 {
245 	int r;
246 
247 	r = -E2BIG;
248 	if (cpuid->nent < vcpu->arch.cpuid_nent)
249 		goto out;
250 	r = -EFAULT;
251 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
252 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
253 		goto out;
254 	return 0;
255 
256 out:
257 	cpuid->nent = vcpu->arch.cpuid_nent;
258 	return r;
259 }
260 
261 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
262 {
263 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
264 	struct kvm_cpuid_entry2 entry;
265 
266 	reverse_cpuid_check(leaf);
267 	kvm_cpu_caps[leaf] &= mask;
268 
269 	cpuid_count(cpuid.function, cpuid.index,
270 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
271 
272 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
273 }
274 
275 void kvm_set_cpu_caps(void)
276 {
277 	unsigned int f_nx = is_efer_nx() ? F(NX) : 0;
278 #ifdef CONFIG_X86_64
279 	unsigned int f_gbpages = F(GBPAGES);
280 	unsigned int f_lm = F(LM);
281 #else
282 	unsigned int f_gbpages = 0;
283 	unsigned int f_lm = 0;
284 #endif
285 
286 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) >
287 		     sizeof(boot_cpu_data.x86_capability));
288 
289 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
290 	       sizeof(kvm_cpu_caps));
291 
292 	kvm_cpu_cap_mask(CPUID_1_ECX,
293 		/*
294 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
295 		 * advertised to guests via CPUID!
296 		 */
297 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
298 		0 /* DS-CPL, VMX, SMX, EST */ |
299 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
300 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
301 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
302 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
303 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
304 		F(F16C) | F(RDRAND)
305 	);
306 	/* KVM emulates x2apic in software irrespective of host support. */
307 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
308 
309 	kvm_cpu_cap_mask(CPUID_1_EDX,
310 		F(FPU) | F(VME) | F(DE) | F(PSE) |
311 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
312 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
313 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
314 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
315 		0 /* Reserved, DS, ACPI */ | F(MMX) |
316 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
317 		0 /* HTT, TM, Reserved, PBE */
318 	);
319 
320 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
321 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
322 		F(BMI2) | F(ERMS) | 0 /*INVPCID*/ | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
323 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
324 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
325 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
326 	);
327 
328 	kvm_cpu_cap_mask(CPUID_7_ECX,
329 		F(AVX512VBMI) | F(LA57) | 0 /*PKU*/ | 0 /*OSPKE*/ | F(RDPID) |
330 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
331 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
332 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/
333 	);
334 	/* Set LA57 based on hardware capability. */
335 	if (cpuid_ecx(7) & F(LA57))
336 		kvm_cpu_cap_set(X86_FEATURE_LA57);
337 
338 	kvm_cpu_cap_mask(CPUID_7_EDX,
339 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
340 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
341 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM)
342 	);
343 
344 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
345 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
346 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
347 
348 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
349 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
350 	if (boot_cpu_has(X86_FEATURE_STIBP))
351 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
352 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
353 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
354 
355 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
356 		F(AVX512_BF16)
357 	);
358 
359 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
360 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
361 	);
362 
363 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
364 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
365 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
366 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
367 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
368 		F(TOPOEXT) | F(PERFCTR_CORE)
369 	);
370 
371 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
372 		F(FPU) | F(VME) | F(DE) | F(PSE) |
373 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
374 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
375 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
376 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
377 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
378 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
379 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
380 	);
381 
382 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
383 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
384 
385 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
386 		F(CLZERO) | F(XSAVEERPTR) |
387 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
388 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON)
389 	);
390 
391 	/*
392 	 * AMD has separate bits for each SPEC_CTRL bit.
393 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
394 	 * record that in cpufeatures so use them.
395 	 */
396 	if (boot_cpu_has(X86_FEATURE_IBPB))
397 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
398 	if (boot_cpu_has(X86_FEATURE_IBRS))
399 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
400 	if (boot_cpu_has(X86_FEATURE_STIBP))
401 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
402 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
403 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
404 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
405 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
406 	/*
407 	 * The preference is to use SPEC CTRL MSR instead of the
408 	 * VIRT_SPEC MSR.
409 	 */
410 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
411 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
412 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
413 
414 	/*
415 	 * Hide all SVM features by default, SVM will set the cap bits for
416 	 * features it emulates and/or exposes for L1.
417 	 */
418 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
419 
420 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
421 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
422 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
423 		F(PMM) | F(PMM_EN)
424 	);
425 }
426 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
427 
428 struct kvm_cpuid_array {
429 	struct kvm_cpuid_entry2 *entries;
430 	const int maxnent;
431 	int nent;
432 };
433 
434 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
435 					      u32 function, u32 index)
436 {
437 	struct kvm_cpuid_entry2 *entry;
438 
439 	if (array->nent >= array->maxnent)
440 		return NULL;
441 
442 	entry = &array->entries[array->nent++];
443 
444 	entry->function = function;
445 	entry->index = index;
446 	entry->flags = 0;
447 
448 	cpuid_count(entry->function, entry->index,
449 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
450 
451 	switch (function) {
452 	case 4:
453 	case 7:
454 	case 0xb:
455 	case 0xd:
456 	case 0xf:
457 	case 0x10:
458 	case 0x12:
459 	case 0x14:
460 	case 0x17:
461 	case 0x18:
462 	case 0x1f:
463 	case 0x8000001d:
464 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
465 		break;
466 	}
467 
468 	return entry;
469 }
470 
471 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
472 {
473 	struct kvm_cpuid_entry2 *entry;
474 
475 	if (array->nent >= array->maxnent)
476 		return -E2BIG;
477 
478 	entry = &array->entries[array->nent];
479 	entry->function = func;
480 	entry->index = 0;
481 	entry->flags = 0;
482 
483 	switch (func) {
484 	case 0:
485 		entry->eax = 7;
486 		++array->nent;
487 		break;
488 	case 1:
489 		entry->ecx = F(MOVBE);
490 		++array->nent;
491 		break;
492 	case 7:
493 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
494 		entry->eax = 0;
495 		entry->ecx = F(RDPID);
496 		++array->nent;
497 	default:
498 		break;
499 	}
500 
501 	return 0;
502 }
503 
504 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
505 {
506 	struct kvm_cpuid_entry2 *entry;
507 	int r, i, max_idx;
508 
509 	/* all calls to cpuid_count() should be made on the same cpu */
510 	get_cpu();
511 
512 	r = -E2BIG;
513 
514 	entry = do_host_cpuid(array, function, 0);
515 	if (!entry)
516 		goto out;
517 
518 	switch (function) {
519 	case 0:
520 		/* Limited to the highest leaf implemented in KVM. */
521 		entry->eax = min(entry->eax, 0x1fU);
522 		break;
523 	case 1:
524 		cpuid_entry_override(entry, CPUID_1_EDX);
525 		cpuid_entry_override(entry, CPUID_1_ECX);
526 		break;
527 	case 2:
528 		/*
529 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
530 		 * CPUID(function=2, index=0) may return different results each
531 		 * time, with the least-significant byte in EAX enumerating the
532 		 * number of times software should do CPUID(2, 0).
533 		 *
534 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
535 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
536 		 * return 01H. Software should ignore this value and not
537 		 * interpret it as an informational descriptor", while AMD's
538 		 * APM states that CPUID(2) is reserved.
539 		 *
540 		 * WARN if a frankenstein CPU that supports virtualization and
541 		 * a stateful CPUID.0x2 is encountered.
542 		 */
543 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
544 		break;
545 	/* functions 4 and 0x8000001d have additional index. */
546 	case 4:
547 	case 0x8000001d:
548 		/*
549 		 * Read entries until the cache type in the previous entry is
550 		 * zero, i.e. indicates an invalid entry.
551 		 */
552 		for (i = 1; entry->eax & 0x1f; ++i) {
553 			entry = do_host_cpuid(array, function, i);
554 			if (!entry)
555 				goto out;
556 		}
557 		break;
558 	case 6: /* Thermal management */
559 		entry->eax = 0x4; /* allow ARAT */
560 		entry->ebx = 0;
561 		entry->ecx = 0;
562 		entry->edx = 0;
563 		break;
564 	/* function 7 has additional index. */
565 	case 7:
566 		entry->eax = min(entry->eax, 1u);
567 		cpuid_entry_override(entry, CPUID_7_0_EBX);
568 		cpuid_entry_override(entry, CPUID_7_ECX);
569 		cpuid_entry_override(entry, CPUID_7_EDX);
570 
571 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
572 		if (entry->eax == 1) {
573 			entry = do_host_cpuid(array, function, 1);
574 			if (!entry)
575 				goto out;
576 
577 			cpuid_entry_override(entry, CPUID_7_1_EAX);
578 			entry->ebx = 0;
579 			entry->ecx = 0;
580 			entry->edx = 0;
581 		}
582 		break;
583 	case 9:
584 		break;
585 	case 0xa: { /* Architectural Performance Monitoring */
586 		struct x86_pmu_capability cap;
587 		union cpuid10_eax eax;
588 		union cpuid10_edx edx;
589 
590 		perf_get_x86_pmu_capability(&cap);
591 
592 		/*
593 		 * Only support guest architectural pmu on a host
594 		 * with architectural pmu.
595 		 */
596 		if (!cap.version)
597 			memset(&cap, 0, sizeof(cap));
598 
599 		eax.split.version_id = min(cap.version, 2);
600 		eax.split.num_counters = cap.num_counters_gp;
601 		eax.split.bit_width = cap.bit_width_gp;
602 		eax.split.mask_length = cap.events_mask_len;
603 
604 		edx.split.num_counters_fixed = cap.num_counters_fixed;
605 		edx.split.bit_width_fixed = cap.bit_width_fixed;
606 		edx.split.reserved = 0;
607 
608 		entry->eax = eax.full;
609 		entry->ebx = cap.events_mask;
610 		entry->ecx = 0;
611 		entry->edx = edx.full;
612 		break;
613 	}
614 	/*
615 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
616 	 * thus they can be handled by common code.
617 	 */
618 	case 0x1f:
619 	case 0xb:
620 		/*
621 		 * Populate entries until the level type (ECX[15:8]) of the
622 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
623 		 * the starting entry, filled by the primary do_host_cpuid().
624 		 */
625 		for (i = 1; entry->ecx & 0xff00; ++i) {
626 			entry = do_host_cpuid(array, function, i);
627 			if (!entry)
628 				goto out;
629 		}
630 		break;
631 	case 0xd:
632 		entry->eax &= supported_xcr0;
633 		entry->ebx = xstate_required_size(supported_xcr0, false);
634 		entry->ecx = entry->ebx;
635 		entry->edx &= supported_xcr0 >> 32;
636 		if (!supported_xcr0)
637 			break;
638 
639 		entry = do_host_cpuid(array, function, 1);
640 		if (!entry)
641 			goto out;
642 
643 		cpuid_entry_override(entry, CPUID_D_1_EAX);
644 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
645 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
646 							  true);
647 		else {
648 			WARN_ON_ONCE(supported_xss != 0);
649 			entry->ebx = 0;
650 		}
651 		entry->ecx &= supported_xss;
652 		entry->edx &= supported_xss >> 32;
653 
654 		for (i = 2; i < 64; ++i) {
655 			bool s_state;
656 			if (supported_xcr0 & BIT_ULL(i))
657 				s_state = false;
658 			else if (supported_xss & BIT_ULL(i))
659 				s_state = true;
660 			else
661 				continue;
662 
663 			entry = do_host_cpuid(array, function, i);
664 			if (!entry)
665 				goto out;
666 
667 			/*
668 			 * The supported check above should have filtered out
669 			 * invalid sub-leafs.  Only valid sub-leafs should
670 			 * reach this point, and they should have a non-zero
671 			 * save state size.  Furthermore, check whether the
672 			 * processor agrees with supported_xcr0/supported_xss
673 			 * on whether this is an XCR0- or IA32_XSS-managed area.
674 			 */
675 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
676 				--array->nent;
677 				continue;
678 			}
679 			entry->edx = 0;
680 		}
681 		break;
682 	/* Intel PT */
683 	case 0x14:
684 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
685 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
686 			break;
687 		}
688 
689 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
690 			if (!do_host_cpuid(array, function, i))
691 				goto out;
692 		}
693 		break;
694 	case KVM_CPUID_SIGNATURE: {
695 		static const char signature[12] = "KVMKVMKVM\0\0";
696 		const u32 *sigptr = (const u32 *)signature;
697 		entry->eax = KVM_CPUID_FEATURES;
698 		entry->ebx = sigptr[0];
699 		entry->ecx = sigptr[1];
700 		entry->edx = sigptr[2];
701 		break;
702 	}
703 	case KVM_CPUID_FEATURES:
704 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
705 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
706 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
707 			     (1 << KVM_FEATURE_ASYNC_PF) |
708 			     (1 << KVM_FEATURE_PV_EOI) |
709 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
710 			     (1 << KVM_FEATURE_PV_UNHALT) |
711 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
712 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
713 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
714 			     (1 << KVM_FEATURE_POLL_CONTROL) |
715 			     (1 << KVM_FEATURE_PV_SCHED_YIELD);
716 
717 		if (sched_info_on())
718 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
719 
720 		entry->ebx = 0;
721 		entry->ecx = 0;
722 		entry->edx = 0;
723 		break;
724 	case 0x80000000:
725 		entry->eax = min(entry->eax, 0x8000001f);
726 		break;
727 	case 0x80000001:
728 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
729 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
730 		break;
731 	case 0x80000007: /* Advanced power management */
732 		/* invariant TSC is CPUID.80000007H:EDX[8] */
733 		entry->edx &= (1 << 8);
734 		/* mask against host */
735 		entry->edx &= boot_cpu_data.x86_power;
736 		entry->eax = entry->ebx = entry->ecx = 0;
737 		break;
738 	case 0x80000008: {
739 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
740 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
741 		unsigned phys_as = entry->eax & 0xff;
742 
743 		if (!g_phys_as)
744 			g_phys_as = phys_as;
745 		entry->eax = g_phys_as | (virt_as << 8);
746 		entry->edx = 0;
747 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
748 		break;
749 	}
750 	case 0x8000000A:
751 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
752 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
753 			break;
754 		}
755 		entry->eax = 1; /* SVM revision 1 */
756 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
757 				   ASID emulation to nested SVM */
758 		entry->ecx = 0; /* Reserved */
759 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
760 		break;
761 	case 0x80000019:
762 		entry->ecx = entry->edx = 0;
763 		break;
764 	case 0x8000001a:
765 	case 0x8000001e:
766 		break;
767 	/* Support memory encryption cpuid if host supports it */
768 	case 0x8000001F:
769 		if (!boot_cpu_has(X86_FEATURE_SEV))
770 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
771 		break;
772 	/*Add support for Centaur's CPUID instruction*/
773 	case 0xC0000000:
774 		/*Just support up to 0xC0000004 now*/
775 		entry->eax = min(entry->eax, 0xC0000004);
776 		break;
777 	case 0xC0000001:
778 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
779 		break;
780 	case 3: /* Processor serial number */
781 	case 5: /* MONITOR/MWAIT */
782 	case 0xC0000002:
783 	case 0xC0000003:
784 	case 0xC0000004:
785 	default:
786 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
787 		break;
788 	}
789 
790 	r = 0;
791 
792 out:
793 	put_cpu();
794 
795 	return r;
796 }
797 
798 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
799 			 unsigned int type)
800 {
801 	if (type == KVM_GET_EMULATED_CPUID)
802 		return __do_cpuid_func_emulated(array, func);
803 
804 	return __do_cpuid_func(array, func);
805 }
806 
807 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
808 
809 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
810 			  unsigned int type)
811 {
812 	u32 limit;
813 	int r;
814 
815 	if (func == CENTAUR_CPUID_SIGNATURE &&
816 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
817 		return 0;
818 
819 	r = do_cpuid_func(array, func, type);
820 	if (r)
821 		return r;
822 
823 	limit = array->entries[array->nent - 1].eax;
824 	for (func = func + 1; func <= limit; ++func) {
825 		r = do_cpuid_func(array, func, type);
826 		if (r)
827 			break;
828 	}
829 
830 	return r;
831 }
832 
833 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
834 				 __u32 num_entries, unsigned int ioctl_type)
835 {
836 	int i;
837 	__u32 pad[3];
838 
839 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
840 		return false;
841 
842 	/*
843 	 * We want to make sure that ->padding is being passed clean from
844 	 * userspace in case we want to use it for something in the future.
845 	 *
846 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
847 	 * have to give ourselves satisfied only with the emulated side. /me
848 	 * sheds a tear.
849 	 */
850 	for (i = 0; i < num_entries; i++) {
851 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
852 			return true;
853 
854 		if (pad[0] || pad[1] || pad[2])
855 			return true;
856 	}
857 	return false;
858 }
859 
860 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
861 			    struct kvm_cpuid_entry2 __user *entries,
862 			    unsigned int type)
863 {
864 	static const u32 funcs[] = {
865 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
866 	};
867 
868 	struct kvm_cpuid_array array = {
869 		.nent = 0,
870 		.maxnent = cpuid->nent,
871 	};
872 	int r, i;
873 
874 	if (cpuid->nent < 1)
875 		return -E2BIG;
876 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
877 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
878 
879 	if (sanity_check_entries(entries, cpuid->nent, type))
880 		return -EINVAL;
881 
882 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
883 					   cpuid->nent));
884 	if (!array.entries)
885 		return -ENOMEM;
886 
887 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
888 		r = get_cpuid_func(&array, funcs[i], type);
889 		if (r)
890 			goto out_free;
891 	}
892 	cpuid->nent = array.nent;
893 
894 	if (copy_to_user(entries, array.entries,
895 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
896 		r = -EFAULT;
897 
898 out_free:
899 	vfree(array.entries);
900 	return r;
901 }
902 
903 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
904 					      u32 function, u32 index)
905 {
906 	struct kvm_cpuid_entry2 *e;
907 	int i;
908 
909 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
910 		e = &vcpu->arch.cpuid_entries[i];
911 
912 		if (e->function == function && (e->index == index ||
913 		    !(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX)))
914 			return e;
915 	}
916 	return NULL;
917 }
918 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
919 
920 /*
921  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
922  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
923  * returns all zeroes for any undefined leaf, whether or not the leaf is in
924  * range.  Centaur/VIA follows Intel semantics.
925  *
926  * A leaf is considered out-of-range if its function is higher than the maximum
927  * supported leaf of its associated class or if its associated class does not
928  * exist.
929  *
930  * There are three primary classes to be considered, with their respective
931  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
932  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
933  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
934  *
935  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
936  *  - Hypervisor: 0x40000000 - 0x4fffffff
937  *  - Extended:   0x80000000 - 0xbfffffff
938  *  - Centaur:    0xc0000000 - 0xcfffffff
939  *
940  * The Hypervisor class is further subdivided into sub-classes that each act as
941  * their own indepdent class associated with a 0x100 byte range.  E.g. if Qemu
942  * is advertising support for both HyperV and KVM, the resulting Hypervisor
943  * CPUID sub-classes are:
944  *
945  *  - HyperV:     0x40000000 - 0x400000ff
946  *  - KVM:        0x40000100 - 0x400001ff
947  */
948 static struct kvm_cpuid_entry2 *
949 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
950 {
951 	struct kvm_cpuid_entry2 *basic, *class;
952 	u32 function = *fn_ptr;
953 
954 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
955 	if (!basic)
956 		return NULL;
957 
958 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
959 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
960 		return NULL;
961 
962 	if (function >= 0x40000000 && function <= 0x4fffffff)
963 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
964 	else if (function >= 0xc0000000)
965 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
966 	else
967 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
968 
969 	if (class && function <= class->eax)
970 		return NULL;
971 
972 	/*
973 	 * Leaf specific adjustments are also applied when redirecting to the
974 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
975 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
976 	 * needs to be pulled from CPUID.0xb.1.
977 	 */
978 	*fn_ptr = basic->eax;
979 
980 	/*
981 	 * The class does not exist or the requested function is out of range;
982 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
983 	 * the original requested leaf is observed!
984 	 */
985 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
986 }
987 
988 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
989 	       u32 *ecx, u32 *edx, bool exact_only)
990 {
991 	u32 orig_function = *eax, function = *eax, index = *ecx;
992 	struct kvm_cpuid_entry2 *entry;
993 	bool exact, used_max_basic = false;
994 
995 	entry = kvm_find_cpuid_entry(vcpu, function, index);
996 	exact = !!entry;
997 
998 	if (!entry && !exact_only) {
999 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1000 		used_max_basic = !!entry;
1001 	}
1002 
1003 	if (entry) {
1004 		*eax = entry->eax;
1005 		*ebx = entry->ebx;
1006 		*ecx = entry->ecx;
1007 		*edx = entry->edx;
1008 		if (function == 7 && index == 0) {
1009 			u64 data;
1010 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1011 			    (data & TSX_CTRL_CPUID_CLEAR))
1012 				*ebx &= ~(F(RTM) | F(HLE));
1013 		}
1014 	} else {
1015 		*eax = *ebx = *ecx = *edx = 0;
1016 		/*
1017 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1018 		 * and EDX is always the x2APIC ID, even for undefined
1019 		 * subleaves. Index 1 will exist iff the leaf is
1020 		 * implemented, so we pass through CL iff leaf 1
1021 		 * exists. EDX can be copied from any existing index.
1022 		 */
1023 		if (function == 0xb || function == 0x1f) {
1024 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1025 			if (entry) {
1026 				*ecx = index & 0xff;
1027 				*edx = entry->edx;
1028 			}
1029 		}
1030 	}
1031 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1032 			used_max_basic);
1033 	return exact;
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_cpuid);
1036 
1037 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1038 {
1039 	u32 eax, ebx, ecx, edx;
1040 
1041 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1042 		return 1;
1043 
1044 	eax = kvm_rax_read(vcpu);
1045 	ecx = kvm_rcx_read(vcpu);
1046 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1047 	kvm_rax_write(vcpu, eax);
1048 	kvm_rbx_write(vcpu, ebx);
1049 	kvm_rcx_write(vcpu, ecx);
1050 	kvm_rdx_write(vcpu, edx);
1051 	return kvm_skip_emulated_instruction(vcpu);
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1054