xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision f125e2d4)
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 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
28 {
29 	int feature_bit = 0;
30 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
31 
32 	xstate_bv &= XFEATURE_MASK_EXTEND;
33 	while (xstate_bv) {
34 		if (xstate_bv & 0x1) {
35 		        u32 eax, ebx, ecx, edx, offset;
36 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
37 			offset = compacted ? ret : ebx;
38 			ret = max(ret, offset + eax);
39 		}
40 
41 		xstate_bv >>= 1;
42 		feature_bit++;
43 	}
44 
45 	return ret;
46 }
47 
48 bool kvm_mpx_supported(void)
49 {
50 	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
51 		 && kvm_x86_ops->mpx_supported());
52 }
53 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
54 
55 u64 kvm_supported_xcr0(void)
56 {
57 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
58 
59 	if (!kvm_mpx_supported())
60 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
61 
62 	return xcr0;
63 }
64 
65 #define F feature_bit
66 
67 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
68 {
69 	struct kvm_cpuid_entry2 *best;
70 	struct kvm_lapic *apic = vcpu->arch.apic;
71 
72 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
73 	if (!best)
74 		return 0;
75 
76 	/* Update OSXSAVE bit */
77 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
78 		best->ecx &= ~F(OSXSAVE);
79 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
80 			best->ecx |= F(OSXSAVE);
81 	}
82 
83 	best->edx &= ~F(APIC);
84 	if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
85 		best->edx |= F(APIC);
86 
87 	if (apic) {
88 		if (best->ecx & F(TSC_DEADLINE_TIMER))
89 			apic->lapic_timer.timer_mode_mask = 3 << 17;
90 		else
91 			apic->lapic_timer.timer_mode_mask = 1 << 17;
92 	}
93 
94 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
95 	if (best) {
96 		/* Update OSPKE bit */
97 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
98 			best->ecx &= ~F(OSPKE);
99 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
100 				best->ecx |= F(OSPKE);
101 		}
102 	}
103 
104 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
105 	if (!best) {
106 		vcpu->arch.guest_supported_xcr0 = 0;
107 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
108 	} else {
109 		vcpu->arch.guest_supported_xcr0 =
110 			(best->eax | ((u64)best->edx << 32)) &
111 			kvm_supported_xcr0();
112 		vcpu->arch.guest_xstate_size = best->ebx =
113 			xstate_required_size(vcpu->arch.xcr0, false);
114 	}
115 
116 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
117 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
118 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
119 
120 	/*
121 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
122 	 * canonical address checks; exit if it is ever changed.
123 	 */
124 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
125 	if (best) {
126 		int vaddr_bits = (best->eax & 0xff00) >> 8;
127 
128 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
129 			return -EINVAL;
130 	}
131 
132 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
133 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
134 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
135 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
136 
137 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
138 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
139 		if (best) {
140 			if (vcpu->arch.ia32_misc_enable_msr & MSR_IA32_MISC_ENABLE_MWAIT)
141 				best->ecx |= F(MWAIT);
142 			else
143 				best->ecx &= ~F(MWAIT);
144 		}
145 	}
146 
147 	/* Update physical-address width */
148 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
149 	kvm_mmu_reset_context(vcpu);
150 
151 	kvm_pmu_refresh(vcpu);
152 	return 0;
153 }
154 
155 static int is_efer_nx(void)
156 {
157 	unsigned long long efer = 0;
158 
159 	rdmsrl_safe(MSR_EFER, &efer);
160 	return efer & EFER_NX;
161 }
162 
163 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
164 {
165 	int i;
166 	struct kvm_cpuid_entry2 *e, *entry;
167 
168 	entry = NULL;
169 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
170 		e = &vcpu->arch.cpuid_entries[i];
171 		if (e->function == 0x80000001) {
172 			entry = e;
173 			break;
174 		}
175 	}
176 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
177 		entry->edx &= ~F(NX);
178 		printk(KERN_INFO "kvm: guest NX capability removed\n");
179 	}
180 }
181 
182 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
183 {
184 	struct kvm_cpuid_entry2 *best;
185 
186 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
187 	if (!best || best->eax < 0x80000008)
188 		goto not_found;
189 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
190 	if (best)
191 		return best->eax & 0xff;
192 not_found:
193 	return 36;
194 }
195 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
196 
197 /* when an old userspace process fills a new kernel module */
198 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
199 			     struct kvm_cpuid *cpuid,
200 			     struct kvm_cpuid_entry __user *entries)
201 {
202 	int r, i;
203 	struct kvm_cpuid_entry *cpuid_entries = NULL;
204 
205 	r = -E2BIG;
206 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
207 		goto out;
208 	r = -ENOMEM;
209 	if (cpuid->nent) {
210 		cpuid_entries =
211 			vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
212 					   cpuid->nent));
213 		if (!cpuid_entries)
214 			goto out;
215 		r = -EFAULT;
216 		if (copy_from_user(cpuid_entries, entries,
217 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
218 			goto out;
219 	}
220 	for (i = 0; i < cpuid->nent; i++) {
221 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
222 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
223 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
224 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
225 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
226 		vcpu->arch.cpuid_entries[i].index = 0;
227 		vcpu->arch.cpuid_entries[i].flags = 0;
228 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
229 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
230 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
231 	}
232 	vcpu->arch.cpuid_nent = cpuid->nent;
233 	cpuid_fix_nx_cap(vcpu);
234 	kvm_apic_set_version(vcpu);
235 	kvm_x86_ops->cpuid_update(vcpu);
236 	r = kvm_update_cpuid(vcpu);
237 
238 out:
239 	vfree(cpuid_entries);
240 	return r;
241 }
242 
243 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
244 			      struct kvm_cpuid2 *cpuid,
245 			      struct kvm_cpuid_entry2 __user *entries)
246 {
247 	int r;
248 
249 	r = -E2BIG;
250 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
251 		goto out;
252 	r = -EFAULT;
253 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
254 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
255 		goto out;
256 	vcpu->arch.cpuid_nent = cpuid->nent;
257 	kvm_apic_set_version(vcpu);
258 	kvm_x86_ops->cpuid_update(vcpu);
259 	r = kvm_update_cpuid(vcpu);
260 out:
261 	return r;
262 }
263 
264 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
265 			      struct kvm_cpuid2 *cpuid,
266 			      struct kvm_cpuid_entry2 __user *entries)
267 {
268 	int r;
269 
270 	r = -E2BIG;
271 	if (cpuid->nent < vcpu->arch.cpuid_nent)
272 		goto out;
273 	r = -EFAULT;
274 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
275 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
276 		goto out;
277 	return 0;
278 
279 out:
280 	cpuid->nent = vcpu->arch.cpuid_nent;
281 	return r;
282 }
283 
284 static __always_inline void cpuid_mask(u32 *word, int wordnum)
285 {
286 	reverse_cpuid_check(wordnum);
287 	*word &= boot_cpu_data.x86_capability[wordnum];
288 }
289 
290 static void do_host_cpuid(struct kvm_cpuid_entry2 *entry, u32 function,
291 			   u32 index)
292 {
293 	entry->function = function;
294 	entry->index = index;
295 	entry->flags = 0;
296 
297 	cpuid_count(entry->function, entry->index,
298 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
299 
300 	switch (function) {
301 	case 2:
302 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
303 		break;
304 	case 4:
305 	case 7:
306 	case 0xb:
307 	case 0xd:
308 	case 0xf:
309 	case 0x10:
310 	case 0x12:
311 	case 0x14:
312 	case 0x17:
313 	case 0x18:
314 	case 0x1f:
315 	case 0x8000001d:
316 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
317 		break;
318 	}
319 }
320 
321 static int __do_cpuid_func_emulated(struct kvm_cpuid_entry2 *entry,
322 				    u32 func, int *nent, int maxnent)
323 {
324 	entry->function = func;
325 	entry->index = 0;
326 	entry->flags = 0;
327 
328 	switch (func) {
329 	case 0:
330 		entry->eax = 7;
331 		++*nent;
332 		break;
333 	case 1:
334 		entry->ecx = F(MOVBE);
335 		++*nent;
336 		break;
337 	case 7:
338 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
339 		entry->eax = 0;
340 		entry->ecx = F(RDPID);
341 		++*nent;
342 	default:
343 		break;
344 	}
345 
346 	return 0;
347 }
348 
349 static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index)
350 {
351 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
352 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
353 	unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
354 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
355 	unsigned f_la57;
356 	unsigned f_pku = kvm_x86_ops->pku_supported() ? F(PKU) : 0;
357 
358 	/* cpuid 7.0.ebx */
359 	const u32 kvm_cpuid_7_0_ebx_x86_features =
360 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
361 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
362 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
363 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
364 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
365 
366 	/* cpuid 7.0.ecx*/
367 	const u32 kvm_cpuid_7_0_ecx_x86_features =
368 		F(AVX512VBMI) | F(LA57) | 0 /*PKU*/ | 0 /*OSPKE*/ | F(RDPID) |
369 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
370 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
371 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/;
372 
373 	/* cpuid 7.0.edx*/
374 	const u32 kvm_cpuid_7_0_edx_x86_features =
375 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
376 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
377 		F(MD_CLEAR);
378 
379 	/* cpuid 7.1.eax */
380 	const u32 kvm_cpuid_7_1_eax_x86_features =
381 		F(AVX512_BF16);
382 
383 	switch (index) {
384 	case 0:
385 		entry->eax = min(entry->eax, 1u);
386 		entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
387 		cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
388 		/* TSC_ADJUST is emulated */
389 		entry->ebx |= F(TSC_ADJUST);
390 
391 		entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
392 		f_la57 = entry->ecx & F(LA57);
393 		cpuid_mask(&entry->ecx, CPUID_7_ECX);
394 		/* Set LA57 based on hardware capability. */
395 		entry->ecx |= f_la57;
396 		entry->ecx |= f_umip;
397 		entry->ecx |= f_pku;
398 		/* PKU is not yet implemented for shadow paging. */
399 		if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
400 			entry->ecx &= ~F(PKU);
401 
402 		entry->edx &= kvm_cpuid_7_0_edx_x86_features;
403 		cpuid_mask(&entry->edx, CPUID_7_EDX);
404 		if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
405 			entry->edx |= F(SPEC_CTRL);
406 		if (boot_cpu_has(X86_FEATURE_STIBP))
407 			entry->edx |= F(INTEL_STIBP);
408 		if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
409 		    boot_cpu_has(X86_FEATURE_AMD_SSBD))
410 			entry->edx |= F(SPEC_CTRL_SSBD);
411 		/*
412 		 * We emulate ARCH_CAPABILITIES in software even
413 		 * if the host doesn't support it.
414 		 */
415 		entry->edx |= F(ARCH_CAPABILITIES);
416 		break;
417 	case 1:
418 		entry->eax &= kvm_cpuid_7_1_eax_x86_features;
419 		entry->ebx = 0;
420 		entry->ecx = 0;
421 		entry->edx = 0;
422 		break;
423 	default:
424 		WARN_ON_ONCE(1);
425 		entry->eax = 0;
426 		entry->ebx = 0;
427 		entry->ecx = 0;
428 		entry->edx = 0;
429 		break;
430 	}
431 }
432 
433 static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,
434 				  int *nent, int maxnent)
435 {
436 	int r;
437 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
438 #ifdef CONFIG_X86_64
439 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
440 				? F(GBPAGES) : 0;
441 	unsigned f_lm = F(LM);
442 #else
443 	unsigned f_gbpages = 0;
444 	unsigned f_lm = 0;
445 #endif
446 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
447 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
448 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
449 
450 	/* cpuid 1.edx */
451 	const u32 kvm_cpuid_1_edx_x86_features =
452 		F(FPU) | F(VME) | F(DE) | F(PSE) |
453 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
454 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
455 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
456 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
457 		0 /* Reserved, DS, ACPI */ | F(MMX) |
458 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
459 		0 /* HTT, TM, Reserved, PBE */;
460 	/* cpuid 0x80000001.edx */
461 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
462 		F(FPU) | F(VME) | F(DE) | F(PSE) |
463 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
464 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
465 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
466 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
467 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
468 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
469 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
470 	/* cpuid 1.ecx */
471 	const u32 kvm_cpuid_1_ecx_x86_features =
472 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
473 		 * but *not* advertised to guests via CPUID ! */
474 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
475 		0 /* DS-CPL, VMX, SMX, EST */ |
476 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
477 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
478 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
479 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
480 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
481 		F(F16C) | F(RDRAND);
482 	/* cpuid 0x80000001.ecx */
483 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
484 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
485 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
486 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
487 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
488 		F(TOPOEXT) | F(PERFCTR_CORE);
489 
490 	/* cpuid 0x80000008.ebx */
491 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
492 		F(CLZERO) | F(XSAVEERPTR) |
493 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
494 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON);
495 
496 	/* cpuid 0xC0000001.edx */
497 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
498 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
499 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
500 		F(PMM) | F(PMM_EN);
501 
502 	/* cpuid 0xD.1.eax */
503 	const u32 kvm_cpuid_D_1_eax_x86_features =
504 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
505 
506 	/* all calls to cpuid_count() should be made on the same cpu */
507 	get_cpu();
508 
509 	r = -E2BIG;
510 
511 	if (WARN_ON(*nent >= maxnent))
512 		goto out;
513 
514 	do_host_cpuid(entry, function, 0);
515 	++*nent;
516 
517 	switch (function) {
518 	case 0:
519 		/* Limited to the highest leaf implemented in KVM. */
520 		entry->eax = min(entry->eax, 0x1fU);
521 		break;
522 	case 1:
523 		entry->edx &= kvm_cpuid_1_edx_x86_features;
524 		cpuid_mask(&entry->edx, CPUID_1_EDX);
525 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
526 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
527 		/* we support x2apic emulation even if host does not support
528 		 * it since we emulate x2apic in software */
529 		entry->ecx |= F(X2APIC);
530 		break;
531 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
532 	 * may return different values. This forces us to get_cpu() before
533 	 * issuing the first command, and also to emulate this annoying behavior
534 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
535 	case 2: {
536 		int t, times = entry->eax & 0xff;
537 
538 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
539 		for (t = 1; t < times; ++t) {
540 			if (*nent >= maxnent)
541 				goto out;
542 
543 			do_host_cpuid(&entry[t], function, 0);
544 			++*nent;
545 		}
546 		break;
547 	}
548 	/* functions 4 and 0x8000001d have additional index. */
549 	case 4:
550 	case 0x8000001d: {
551 		int i, cache_type;
552 
553 		/* read more entries until cache_type is zero */
554 		for (i = 1; ; ++i) {
555 			if (*nent >= maxnent)
556 				goto out;
557 
558 			cache_type = entry[i - 1].eax & 0x1f;
559 			if (!cache_type)
560 				break;
561 			do_host_cpuid(&entry[i], function, i);
562 			++*nent;
563 		}
564 		break;
565 	}
566 	case 6: /* Thermal management */
567 		entry->eax = 0x4; /* allow ARAT */
568 		entry->ebx = 0;
569 		entry->ecx = 0;
570 		entry->edx = 0;
571 		break;
572 	/* function 7 has additional index. */
573 	case 7: {
574 		int i;
575 
576 		for (i = 0; ; ) {
577 			do_cpuid_7_mask(&entry[i], i);
578 			if (i == entry->eax)
579 				break;
580 			if (*nent >= maxnent)
581 				goto out;
582 
583 			++i;
584 			do_host_cpuid(&entry[i], function, i);
585 			++*nent;
586 		}
587 		break;
588 	}
589 	case 9:
590 		break;
591 	case 0xa: { /* Architectural Performance Monitoring */
592 		struct x86_pmu_capability cap;
593 		union cpuid10_eax eax;
594 		union cpuid10_edx edx;
595 
596 		perf_get_x86_pmu_capability(&cap);
597 
598 		/*
599 		 * Only support guest architectural pmu on a host
600 		 * with architectural pmu.
601 		 */
602 		if (!cap.version)
603 			memset(&cap, 0, sizeof(cap));
604 
605 		eax.split.version_id = min(cap.version, 2);
606 		eax.split.num_counters = cap.num_counters_gp;
607 		eax.split.bit_width = cap.bit_width_gp;
608 		eax.split.mask_length = cap.events_mask_len;
609 
610 		edx.split.num_counters_fixed = cap.num_counters_fixed;
611 		edx.split.bit_width_fixed = cap.bit_width_fixed;
612 		edx.split.reserved = 0;
613 
614 		entry->eax = eax.full;
615 		entry->ebx = cap.events_mask;
616 		entry->ecx = 0;
617 		entry->edx = edx.full;
618 		break;
619 	}
620 	/*
621 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
622 	 * thus they can be handled by common code.
623 	 */
624 	case 0x1f:
625 	case 0xb: {
626 		int i;
627 
628 		/*
629 		 * We filled in entry[0] for CPUID(EAX=<function>,
630 		 * ECX=00H) above.  If its level type (ECX[15:8]) is
631 		 * zero, then the leaf is unimplemented, and we're
632 		 * done.  Otherwise, continue to populate entries
633 		 * until the level type (ECX[15:8]) of the previously
634 		 * added entry is zero.
635 		 */
636 		for (i = 1; entry[i - 1].ecx & 0xff00; ++i) {
637 			if (*nent >= maxnent)
638 				goto out;
639 
640 			do_host_cpuid(&entry[i], function, i);
641 			++*nent;
642 		}
643 		break;
644 	}
645 	case 0xd: {
646 		int idx, i;
647 		u64 supported = kvm_supported_xcr0();
648 
649 		entry->eax &= supported;
650 		entry->ebx = xstate_required_size(supported, false);
651 		entry->ecx = entry->ebx;
652 		entry->edx &= supported >> 32;
653 		if (!supported)
654 			break;
655 
656 		for (idx = 1, i = 1; idx < 64; ++idx) {
657 			u64 mask = ((u64)1 << idx);
658 			if (*nent >= maxnent)
659 				goto out;
660 
661 			do_host_cpuid(&entry[i], function, idx);
662 			if (idx == 1) {
663 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
664 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
665 				entry[i].ebx = 0;
666 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
667 					entry[i].ebx =
668 						xstate_required_size(supported,
669 								     true);
670 			} else {
671 				if (entry[i].eax == 0 || !(supported & mask))
672 					continue;
673 				if (WARN_ON_ONCE(entry[i].ecx & 1))
674 					continue;
675 			}
676 			entry[i].ecx = 0;
677 			entry[i].edx = 0;
678 			++*nent;
679 			++i;
680 		}
681 		break;
682 	}
683 	/* Intel PT */
684 	case 0x14: {
685 		int t, times = entry->eax;
686 
687 		if (!f_intel_pt)
688 			break;
689 
690 		for (t = 1; t <= times; ++t) {
691 			if (*nent >= maxnent)
692 				goto out;
693 			do_host_cpuid(&entry[t], function, t);
694 			++*nent;
695 		}
696 		break;
697 	}
698 	case KVM_CPUID_SIGNATURE: {
699 		static const char signature[12] = "KVMKVMKVM\0\0";
700 		const u32 *sigptr = (const u32 *)signature;
701 		entry->eax = KVM_CPUID_FEATURES;
702 		entry->ebx = sigptr[0];
703 		entry->ecx = sigptr[1];
704 		entry->edx = sigptr[2];
705 		break;
706 	}
707 	case KVM_CPUID_FEATURES:
708 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
709 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
710 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
711 			     (1 << KVM_FEATURE_ASYNC_PF) |
712 			     (1 << KVM_FEATURE_PV_EOI) |
713 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
714 			     (1 << KVM_FEATURE_PV_UNHALT) |
715 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
716 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
717 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
718 			     (1 << KVM_FEATURE_POLL_CONTROL) |
719 			     (1 << KVM_FEATURE_PV_SCHED_YIELD);
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 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
733 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
734 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
735 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
736 		break;
737 	case 0x80000007: /* Advanced power management */
738 		/* invariant TSC is CPUID.80000007H:EDX[8] */
739 		entry->edx &= (1 << 8);
740 		/* mask against host */
741 		entry->edx &= boot_cpu_data.x86_power;
742 		entry->eax = entry->ebx = entry->ecx = 0;
743 		break;
744 	case 0x80000008: {
745 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
746 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
747 		unsigned phys_as = entry->eax & 0xff;
748 
749 		if (!g_phys_as)
750 			g_phys_as = phys_as;
751 		entry->eax = g_phys_as | (virt_as << 8);
752 		entry->edx = 0;
753 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
754 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
755 		/*
756 		 * AMD has separate bits for each SPEC_CTRL bit.
757 		 * arch/x86/kernel/cpu/bugs.c is kind enough to
758 		 * record that in cpufeatures so use them.
759 		 */
760 		if (boot_cpu_has(X86_FEATURE_IBPB))
761 			entry->ebx |= F(AMD_IBPB);
762 		if (boot_cpu_has(X86_FEATURE_IBRS))
763 			entry->ebx |= F(AMD_IBRS);
764 		if (boot_cpu_has(X86_FEATURE_STIBP))
765 			entry->ebx |= F(AMD_STIBP);
766 		if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
767 		    boot_cpu_has(X86_FEATURE_AMD_SSBD))
768 			entry->ebx |= F(AMD_SSBD);
769 		if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
770 			entry->ebx |= F(AMD_SSB_NO);
771 		/*
772 		 * The preference is to use SPEC CTRL MSR instead of the
773 		 * VIRT_SPEC MSR.
774 		 */
775 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
776 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
777 			entry->ebx |= F(VIRT_SSBD);
778 		break;
779 	}
780 	case 0x80000019:
781 		entry->ecx = entry->edx = 0;
782 		break;
783 	case 0x8000001a:
784 	case 0x8000001e:
785 		break;
786 	/* Support memory encryption cpuid if host supports it */
787 	case 0x8000001F:
788 		if (!boot_cpu_has(X86_FEATURE_SEV))
789 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
790 		break;
791 	/*Add support for Centaur's CPUID instruction*/
792 	case 0xC0000000:
793 		/*Just support up to 0xC0000004 now*/
794 		entry->eax = min(entry->eax, 0xC0000004);
795 		break;
796 	case 0xC0000001:
797 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
798 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
799 		break;
800 	case 3: /* Processor serial number */
801 	case 5: /* MONITOR/MWAIT */
802 	case 0xC0000002:
803 	case 0xC0000003:
804 	case 0xC0000004:
805 	default:
806 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
807 		break;
808 	}
809 
810 	kvm_x86_ops->set_supported_cpuid(function, entry);
811 
812 	r = 0;
813 
814 out:
815 	put_cpu();
816 
817 	return r;
818 }
819 
820 static int do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 func,
821 			 int *nent, int maxnent, unsigned int type)
822 {
823 	if (*nent >= maxnent)
824 		return -E2BIG;
825 
826 	if (type == KVM_GET_EMULATED_CPUID)
827 		return __do_cpuid_func_emulated(entry, func, nent, maxnent);
828 
829 	return __do_cpuid_func(entry, func, nent, maxnent);
830 }
831 
832 struct kvm_cpuid_param {
833 	u32 func;
834 	bool (*qualifier)(const struct kvm_cpuid_param *param);
835 };
836 
837 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
838 {
839 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
840 }
841 
842 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
843 				 __u32 num_entries, unsigned int ioctl_type)
844 {
845 	int i;
846 	__u32 pad[3];
847 
848 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
849 		return false;
850 
851 	/*
852 	 * We want to make sure that ->padding is being passed clean from
853 	 * userspace in case we want to use it for something in the future.
854 	 *
855 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
856 	 * have to give ourselves satisfied only with the emulated side. /me
857 	 * sheds a tear.
858 	 */
859 	for (i = 0; i < num_entries; i++) {
860 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
861 			return true;
862 
863 		if (pad[0] || pad[1] || pad[2])
864 			return true;
865 	}
866 	return false;
867 }
868 
869 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
870 			    struct kvm_cpuid_entry2 __user *entries,
871 			    unsigned int type)
872 {
873 	struct kvm_cpuid_entry2 *cpuid_entries;
874 	int limit, nent = 0, r = -E2BIG, i;
875 	u32 func;
876 	static const struct kvm_cpuid_param param[] = {
877 		{ .func = 0 },
878 		{ .func = 0x80000000 },
879 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu },
880 		{ .func = KVM_CPUID_SIGNATURE },
881 	};
882 
883 	if (cpuid->nent < 1)
884 		goto out;
885 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
886 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
887 
888 	if (sanity_check_entries(entries, cpuid->nent, type))
889 		return -EINVAL;
890 
891 	r = -ENOMEM;
892 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
893 					   cpuid->nent));
894 	if (!cpuid_entries)
895 		goto out;
896 
897 	r = 0;
898 	for (i = 0; i < ARRAY_SIZE(param); i++) {
899 		const struct kvm_cpuid_param *ent = &param[i];
900 
901 		if (ent->qualifier && !ent->qualifier(ent))
902 			continue;
903 
904 		r = do_cpuid_func(&cpuid_entries[nent], ent->func,
905 				  &nent, cpuid->nent, type);
906 
907 		if (r)
908 			goto out_free;
909 
910 		limit = cpuid_entries[nent - 1].eax;
911 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
912 			r = do_cpuid_func(&cpuid_entries[nent], func,
913 				          &nent, cpuid->nent, type);
914 
915 		if (r)
916 			goto out_free;
917 	}
918 
919 	r = -EFAULT;
920 	if (copy_to_user(entries, cpuid_entries,
921 			 nent * sizeof(struct kvm_cpuid_entry2)))
922 		goto out_free;
923 	cpuid->nent = nent;
924 	r = 0;
925 
926 out_free:
927 	vfree(cpuid_entries);
928 out:
929 	return r;
930 }
931 
932 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
933 {
934 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
935 	struct kvm_cpuid_entry2 *ej;
936 	int j = i;
937 	int nent = vcpu->arch.cpuid_nent;
938 
939 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
940 	/* when no next entry is found, the current entry[i] is reselected */
941 	do {
942 		j = (j + 1) % nent;
943 		ej = &vcpu->arch.cpuid_entries[j];
944 	} while (ej->function != e->function);
945 
946 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
947 
948 	return j;
949 }
950 
951 /* find an entry with matching function, matching index (if needed), and that
952  * should be read next (if it's stateful) */
953 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
954 	u32 function, u32 index)
955 {
956 	if (e->function != function)
957 		return 0;
958 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
959 		return 0;
960 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
961 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
962 		return 0;
963 	return 1;
964 }
965 
966 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
967 					      u32 function, u32 index)
968 {
969 	int i;
970 	struct kvm_cpuid_entry2 *best = NULL;
971 
972 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
973 		struct kvm_cpuid_entry2 *e;
974 
975 		e = &vcpu->arch.cpuid_entries[i];
976 		if (is_matching_cpuid_entry(e, function, index)) {
977 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
978 				move_to_next_stateful_cpuid_entry(vcpu, i);
979 			best = e;
980 			break;
981 		}
982 	}
983 	return best;
984 }
985 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
986 
987 /*
988  * If the basic or extended CPUID leaf requested is higher than the
989  * maximum supported basic or extended leaf, respectively, then it is
990  * out of range.
991  */
992 static bool cpuid_function_in_range(struct kvm_vcpu *vcpu, u32 function)
993 {
994 	struct kvm_cpuid_entry2 *max;
995 
996 	max = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
997 	return max && function <= max->eax;
998 }
999 
1000 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1001 	       u32 *ecx, u32 *edx, bool check_limit)
1002 {
1003 	u32 function = *eax, index = *ecx;
1004 	struct kvm_cpuid_entry2 *entry;
1005 	struct kvm_cpuid_entry2 *max;
1006 	bool found;
1007 
1008 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1009 	found = entry;
1010 	/*
1011 	 * Intel CPUID semantics treats any query for an out-of-range
1012 	 * leaf as if the highest basic leaf (i.e. CPUID.0H:EAX) were
1013 	 * requested. AMD CPUID semantics returns all zeroes for any
1014 	 * undefined leaf, whether or not the leaf is in range.
1015 	 */
1016 	if (!entry && check_limit && !guest_cpuid_is_amd(vcpu) &&
1017 	    !cpuid_function_in_range(vcpu, function)) {
1018 		max = kvm_find_cpuid_entry(vcpu, 0, 0);
1019 		if (max) {
1020 			function = max->eax;
1021 			entry = kvm_find_cpuid_entry(vcpu, function, index);
1022 		}
1023 	}
1024 	if (entry) {
1025 		*eax = entry->eax;
1026 		*ebx = entry->ebx;
1027 		*ecx = entry->ecx;
1028 		*edx = entry->edx;
1029 		if (function == 7 && index == 0) {
1030 			u64 data;
1031 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1032 			    (data & TSX_CTRL_CPUID_CLEAR))
1033 				*ebx &= ~(F(RTM) | F(HLE));
1034 		}
1035 	} else {
1036 		*eax = *ebx = *ecx = *edx = 0;
1037 		/*
1038 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1039 		 * and EDX is always the x2APIC ID, even for undefined
1040 		 * subleaves. Index 1 will exist iff the leaf is
1041 		 * implemented, so we pass through CL iff leaf 1
1042 		 * exists. EDX can be copied from any existing index.
1043 		 */
1044 		if (function == 0xb || function == 0x1f) {
1045 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1046 			if (entry) {
1047 				*ecx = index & 0xff;
1048 				*edx = entry->edx;
1049 			}
1050 		}
1051 	}
1052 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, found);
1053 	return found;
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_cpuid);
1056 
1057 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1058 {
1059 	u32 eax, ebx, ecx, edx;
1060 
1061 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1062 		return 1;
1063 
1064 	eax = kvm_rax_read(vcpu);
1065 	ecx = kvm_rcx_read(vcpu);
1066 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
1067 	kvm_rax_write(vcpu, eax);
1068 	kvm_rbx_write(vcpu, ebx);
1069 	kvm_rcx_write(vcpu, ecx);
1070 	kvm_rdx_write(vcpu, edx);
1071 	return kvm_skip_emulated_instruction(vcpu);
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1074