xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 2127c01b)
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(x) bit(X86_FEATURE_##x)
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 void cpuid_mask(u32 *word, int wordnum)
285 {
286 	*word &= boot_cpu_data.x86_capability[wordnum];
287 }
288 
289 static void do_host_cpuid(struct kvm_cpuid_entry2 *entry, u32 function,
290 			   u32 index)
291 {
292 	entry->function = function;
293 	entry->index = index;
294 	entry->flags = 0;
295 
296 	cpuid_count(entry->function, entry->index,
297 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
298 
299 	switch (function) {
300 	case 2:
301 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
302 		break;
303 	case 4:
304 	case 7:
305 	case 0xb:
306 	case 0xd:
307 	case 0x14:
308 	case 0x8000001d:
309 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
310 		break;
311 	}
312 }
313 
314 static int __do_cpuid_func_emulated(struct kvm_cpuid_entry2 *entry,
315 				    u32 func, int *nent, int maxnent)
316 {
317 	entry->function = func;
318 	entry->index = 0;
319 	entry->flags = 0;
320 
321 	switch (func) {
322 	case 0:
323 		entry->eax = 7;
324 		++*nent;
325 		break;
326 	case 1:
327 		entry->ecx = F(MOVBE);
328 		++*nent;
329 		break;
330 	case 7:
331 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
332 		entry->eax = 0;
333 		entry->ecx = F(RDPID);
334 		++*nent;
335 	default:
336 		break;
337 	}
338 
339 	return 0;
340 }
341 
342 static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index)
343 {
344 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
345 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
346 	unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
347 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
348 	unsigned f_la57;
349 
350 	/* cpuid 7.0.ebx */
351 	const u32 kvm_cpuid_7_0_ebx_x86_features =
352 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
353 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
354 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
355 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
356 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
357 
358 	/* cpuid 7.0.ecx*/
359 	const u32 kvm_cpuid_7_0_ecx_x86_features =
360 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
361 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
362 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
363 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B);
364 
365 	/* cpuid 7.0.edx*/
366 	const u32 kvm_cpuid_7_0_edx_x86_features =
367 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
368 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
369 		F(MD_CLEAR);
370 
371 	/* cpuid 7.1.eax */
372 	const u32 kvm_cpuid_7_1_eax_x86_features =
373 		F(AVX512_BF16);
374 
375 	switch (index) {
376 	case 0:
377 		entry->eax = min(entry->eax, 1u);
378 		entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
379 		cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
380 		/* TSC_ADJUST is emulated */
381 		entry->ebx |= F(TSC_ADJUST);
382 
383 		entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
384 		f_la57 = entry->ecx & F(LA57);
385 		cpuid_mask(&entry->ecx, CPUID_7_ECX);
386 		/* Set LA57 based on hardware capability. */
387 		entry->ecx |= f_la57;
388 		entry->ecx |= f_umip;
389 		/* PKU is not yet implemented for shadow paging. */
390 		if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
391 			entry->ecx &= ~F(PKU);
392 
393 		entry->edx &= kvm_cpuid_7_0_edx_x86_features;
394 		cpuid_mask(&entry->edx, CPUID_7_EDX);
395 		/*
396 		 * We emulate ARCH_CAPABILITIES in software even
397 		 * if the host doesn't support it.
398 		 */
399 		entry->edx |= F(ARCH_CAPABILITIES);
400 		break;
401 	case 1:
402 		entry->eax &= kvm_cpuid_7_1_eax_x86_features;
403 		entry->ebx = 0;
404 		entry->ecx = 0;
405 		entry->edx = 0;
406 		break;
407 	default:
408 		WARN_ON_ONCE(1);
409 		entry->eax = 0;
410 		entry->ebx = 0;
411 		entry->ecx = 0;
412 		entry->edx = 0;
413 		break;
414 	}
415 }
416 
417 static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,
418 				  int *nent, int maxnent)
419 {
420 	int r;
421 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
422 #ifdef CONFIG_X86_64
423 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
424 				? F(GBPAGES) : 0;
425 	unsigned f_lm = F(LM);
426 #else
427 	unsigned f_gbpages = 0;
428 	unsigned f_lm = 0;
429 #endif
430 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
431 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
432 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
433 
434 	/* cpuid 1.edx */
435 	const u32 kvm_cpuid_1_edx_x86_features =
436 		F(FPU) | F(VME) | F(DE) | F(PSE) |
437 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
438 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
439 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
440 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
441 		0 /* Reserved, DS, ACPI */ | F(MMX) |
442 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
443 		0 /* HTT, TM, Reserved, PBE */;
444 	/* cpuid 0x80000001.edx */
445 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
446 		F(FPU) | F(VME) | F(DE) | F(PSE) |
447 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
448 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
449 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
450 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
451 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
452 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
453 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
454 	/* cpuid 1.ecx */
455 	const u32 kvm_cpuid_1_ecx_x86_features =
456 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
457 		 * but *not* advertised to guests via CPUID ! */
458 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
459 		0 /* DS-CPL, VMX, SMX, EST */ |
460 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
461 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
462 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
463 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
464 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
465 		F(F16C) | F(RDRAND);
466 	/* cpuid 0x80000001.ecx */
467 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
468 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
469 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
470 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
471 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
472 		F(TOPOEXT) | F(PERFCTR_CORE);
473 
474 	/* cpuid 0x80000008.ebx */
475 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
476 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
477 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON);
478 
479 	/* cpuid 0xC0000001.edx */
480 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
481 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
482 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
483 		F(PMM) | F(PMM_EN);
484 
485 	/* cpuid 0xD.1.eax */
486 	const u32 kvm_cpuid_D_1_eax_x86_features =
487 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
488 
489 	/* all calls to cpuid_count() should be made on the same cpu */
490 	get_cpu();
491 
492 	r = -E2BIG;
493 
494 	if (*nent >= maxnent)
495 		goto out;
496 
497 	do_host_cpuid(entry, function, 0);
498 	++*nent;
499 
500 	switch (function) {
501 	case 0:
502 		/* Limited to the highest leaf implemented in KVM. */
503 		entry->eax = min(entry->eax, 0x1fU);
504 		break;
505 	case 1:
506 		entry->edx &= kvm_cpuid_1_edx_x86_features;
507 		cpuid_mask(&entry->edx, CPUID_1_EDX);
508 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
509 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
510 		/* we support x2apic emulation even if host does not support
511 		 * it since we emulate x2apic in software */
512 		entry->ecx |= F(X2APIC);
513 		break;
514 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
515 	 * may return different values. This forces us to get_cpu() before
516 	 * issuing the first command, and also to emulate this annoying behavior
517 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
518 	case 2: {
519 		int t, times = entry->eax & 0xff;
520 
521 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
522 		for (t = 1; t < times; ++t) {
523 			if (*nent >= maxnent)
524 				goto out;
525 
526 			do_host_cpuid(&entry[t], function, 0);
527 			++*nent;
528 		}
529 		break;
530 	}
531 	/* functions 4 and 0x8000001d have additional index. */
532 	case 4:
533 	case 0x8000001d: {
534 		int i, cache_type;
535 
536 		/* read more entries until cache_type is zero */
537 		for (i = 1; ; ++i) {
538 			if (*nent >= maxnent)
539 				goto out;
540 
541 			cache_type = entry[i - 1].eax & 0x1f;
542 			if (!cache_type)
543 				break;
544 			do_host_cpuid(&entry[i], function, i);
545 			++*nent;
546 		}
547 		break;
548 	}
549 	case 6: /* Thermal management */
550 		entry->eax = 0x4; /* allow ARAT */
551 		entry->ebx = 0;
552 		entry->ecx = 0;
553 		entry->edx = 0;
554 		break;
555 	/* function 7 has additional index. */
556 	case 7: {
557 		int i;
558 
559 		for (i = 0; ; ) {
560 			do_cpuid_7_mask(&entry[i], i);
561 			if (i == entry->eax)
562 				break;
563 			if (*nent >= maxnent)
564 				goto out;
565 
566 			++i;
567 			do_host_cpuid(&entry[i], function, i);
568 			++*nent;
569 		}
570 		break;
571 	}
572 	case 9:
573 		break;
574 	case 0xa: { /* Architectural Performance Monitoring */
575 		struct x86_pmu_capability cap;
576 		union cpuid10_eax eax;
577 		union cpuid10_edx edx;
578 
579 		perf_get_x86_pmu_capability(&cap);
580 
581 		/*
582 		 * Only support guest architectural pmu on a host
583 		 * with architectural pmu.
584 		 */
585 		if (!cap.version)
586 			memset(&cap, 0, sizeof(cap));
587 
588 		eax.split.version_id = min(cap.version, 2);
589 		eax.split.num_counters = cap.num_counters_gp;
590 		eax.split.bit_width = cap.bit_width_gp;
591 		eax.split.mask_length = cap.events_mask_len;
592 
593 		edx.split.num_counters_fixed = cap.num_counters_fixed;
594 		edx.split.bit_width_fixed = cap.bit_width_fixed;
595 		edx.split.reserved = 0;
596 
597 		entry->eax = eax.full;
598 		entry->ebx = cap.events_mask;
599 		entry->ecx = 0;
600 		entry->edx = edx.full;
601 		break;
602 	}
603 	/*
604 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
605 	 * thus they can be handled by common code.
606 	 */
607 	case 0x1f:
608 	case 0xb: {
609 		int i, level_type;
610 
611 		/* read more entries until level_type is zero */
612 		for (i = 1; ; ++i) {
613 			if (*nent >= maxnent)
614 				goto out;
615 
616 			level_type = entry[i - 1].ecx & 0xff00;
617 			if (!level_type)
618 				break;
619 			do_host_cpuid(&entry[i], function, i);
620 			++*nent;
621 		}
622 		break;
623 	}
624 	case 0xd: {
625 		int idx, i;
626 		u64 supported = kvm_supported_xcr0();
627 
628 		entry->eax &= supported;
629 		entry->ebx = xstate_required_size(supported, false);
630 		entry->ecx = entry->ebx;
631 		entry->edx &= supported >> 32;
632 		if (!supported)
633 			break;
634 
635 		for (idx = 1, i = 1; idx < 64; ++idx) {
636 			u64 mask = ((u64)1 << idx);
637 			if (*nent >= maxnent)
638 				goto out;
639 
640 			do_host_cpuid(&entry[i], function, idx);
641 			if (idx == 1) {
642 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
643 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
644 				entry[i].ebx = 0;
645 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
646 					entry[i].ebx =
647 						xstate_required_size(supported,
648 								     true);
649 			} else {
650 				if (entry[i].eax == 0 || !(supported & mask))
651 					continue;
652 				if (WARN_ON_ONCE(entry[i].ecx & 1))
653 					continue;
654 			}
655 			entry[i].ecx = 0;
656 			entry[i].edx = 0;
657 			++*nent;
658 			++i;
659 		}
660 		break;
661 	}
662 	/* Intel PT */
663 	case 0x14: {
664 		int t, times = entry->eax;
665 
666 		if (!f_intel_pt)
667 			break;
668 
669 		for (t = 1; t <= times; ++t) {
670 			if (*nent >= maxnent)
671 				goto out;
672 			do_host_cpuid(&entry[t], function, t);
673 			++*nent;
674 		}
675 		break;
676 	}
677 	case KVM_CPUID_SIGNATURE: {
678 		static const char signature[12] = "KVMKVMKVM\0\0";
679 		const u32 *sigptr = (const u32 *)signature;
680 		entry->eax = KVM_CPUID_FEATURES;
681 		entry->ebx = sigptr[0];
682 		entry->ecx = sigptr[1];
683 		entry->edx = sigptr[2];
684 		break;
685 	}
686 	case KVM_CPUID_FEATURES:
687 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
688 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
689 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
690 			     (1 << KVM_FEATURE_ASYNC_PF) |
691 			     (1 << KVM_FEATURE_PV_EOI) |
692 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
693 			     (1 << KVM_FEATURE_PV_UNHALT) |
694 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
695 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
696 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
697 			     (1 << KVM_FEATURE_POLL_CONTROL) |
698 			     (1 << KVM_FEATURE_PV_SCHED_YIELD);
699 
700 		if (sched_info_on())
701 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
702 
703 		entry->ebx = 0;
704 		entry->ecx = 0;
705 		entry->edx = 0;
706 		break;
707 	case 0x80000000:
708 		entry->eax = min(entry->eax, 0x8000001f);
709 		break;
710 	case 0x80000001:
711 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
712 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
713 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
714 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
715 		break;
716 	case 0x80000007: /* Advanced power management */
717 		/* invariant TSC is CPUID.80000007H:EDX[8] */
718 		entry->edx &= (1 << 8);
719 		/* mask against host */
720 		entry->edx &= boot_cpu_data.x86_power;
721 		entry->eax = entry->ebx = entry->ecx = 0;
722 		break;
723 	case 0x80000008: {
724 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
725 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
726 		unsigned phys_as = entry->eax & 0xff;
727 
728 		if (!g_phys_as)
729 			g_phys_as = phys_as;
730 		entry->eax = g_phys_as | (virt_as << 8);
731 		entry->edx = 0;
732 		/*
733 		 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
734 		 * hardware cpuid
735 		 */
736 		if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
737 			entry->ebx |= F(AMD_IBPB);
738 		if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
739 			entry->ebx |= F(AMD_IBRS);
740 		if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
741 			entry->ebx |= F(VIRT_SSBD);
742 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
743 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
744 		/*
745 		 * The preference is to use SPEC CTRL MSR instead of the
746 		 * VIRT_SPEC MSR.
747 		 */
748 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
749 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
750 			entry->ebx |= F(VIRT_SSBD);
751 		break;
752 	}
753 	case 0x80000019:
754 		entry->ecx = entry->edx = 0;
755 		break;
756 	case 0x8000001a:
757 	case 0x8000001e:
758 		break;
759 	/*Add support for Centaur's CPUID instruction*/
760 	case 0xC0000000:
761 		/*Just support up to 0xC0000004 now*/
762 		entry->eax = min(entry->eax, 0xC0000004);
763 		break;
764 	case 0xC0000001:
765 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
766 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
767 		break;
768 	case 3: /* Processor serial number */
769 	case 5: /* MONITOR/MWAIT */
770 	case 0xC0000002:
771 	case 0xC0000003:
772 	case 0xC0000004:
773 	default:
774 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
775 		break;
776 	}
777 
778 	kvm_x86_ops->set_supported_cpuid(function, entry);
779 
780 	r = 0;
781 
782 out:
783 	put_cpu();
784 
785 	return r;
786 }
787 
788 static int do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 func,
789 			 int *nent, int maxnent, unsigned int type)
790 {
791 	if (type == KVM_GET_EMULATED_CPUID)
792 		return __do_cpuid_func_emulated(entry, func, nent, maxnent);
793 
794 	return __do_cpuid_func(entry, func, nent, maxnent);
795 }
796 
797 #undef F
798 
799 struct kvm_cpuid_param {
800 	u32 func;
801 	bool (*qualifier)(const struct kvm_cpuid_param *param);
802 };
803 
804 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
805 {
806 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
807 }
808 
809 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
810 				 __u32 num_entries, unsigned int ioctl_type)
811 {
812 	int i;
813 	__u32 pad[3];
814 
815 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
816 		return false;
817 
818 	/*
819 	 * We want to make sure that ->padding is being passed clean from
820 	 * userspace in case we want to use it for something in the future.
821 	 *
822 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
823 	 * have to give ourselves satisfied only with the emulated side. /me
824 	 * sheds a tear.
825 	 */
826 	for (i = 0; i < num_entries; i++) {
827 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
828 			return true;
829 
830 		if (pad[0] || pad[1] || pad[2])
831 			return true;
832 	}
833 	return false;
834 }
835 
836 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
837 			    struct kvm_cpuid_entry2 __user *entries,
838 			    unsigned int type)
839 {
840 	struct kvm_cpuid_entry2 *cpuid_entries;
841 	int limit, nent = 0, r = -E2BIG, i;
842 	u32 func;
843 	static const struct kvm_cpuid_param param[] = {
844 		{ .func = 0 },
845 		{ .func = 0x80000000 },
846 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu },
847 		{ .func = KVM_CPUID_SIGNATURE },
848 	};
849 
850 	if (cpuid->nent < 1)
851 		goto out;
852 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
853 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
854 
855 	if (sanity_check_entries(entries, cpuid->nent, type))
856 		return -EINVAL;
857 
858 	r = -ENOMEM;
859 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
860 					   cpuid->nent));
861 	if (!cpuid_entries)
862 		goto out;
863 
864 	r = 0;
865 	for (i = 0; i < ARRAY_SIZE(param); i++) {
866 		const struct kvm_cpuid_param *ent = &param[i];
867 
868 		if (ent->qualifier && !ent->qualifier(ent))
869 			continue;
870 
871 		r = do_cpuid_func(&cpuid_entries[nent], ent->func,
872 				  &nent, cpuid->nent, type);
873 
874 		if (r)
875 			goto out_free;
876 
877 		limit = cpuid_entries[nent - 1].eax;
878 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
879 			r = do_cpuid_func(&cpuid_entries[nent], func,
880 				          &nent, cpuid->nent, type);
881 
882 		if (r)
883 			goto out_free;
884 	}
885 
886 	r = -EFAULT;
887 	if (copy_to_user(entries, cpuid_entries,
888 			 nent * sizeof(struct kvm_cpuid_entry2)))
889 		goto out_free;
890 	cpuid->nent = nent;
891 	r = 0;
892 
893 out_free:
894 	vfree(cpuid_entries);
895 out:
896 	return r;
897 }
898 
899 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
900 {
901 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
902 	struct kvm_cpuid_entry2 *ej;
903 	int j = i;
904 	int nent = vcpu->arch.cpuid_nent;
905 
906 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
907 	/* when no next entry is found, the current entry[i] is reselected */
908 	do {
909 		j = (j + 1) % nent;
910 		ej = &vcpu->arch.cpuid_entries[j];
911 	} while (ej->function != e->function);
912 
913 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
914 
915 	return j;
916 }
917 
918 /* find an entry with matching function, matching index (if needed), and that
919  * should be read next (if it's stateful) */
920 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
921 	u32 function, u32 index)
922 {
923 	if (e->function != function)
924 		return 0;
925 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
926 		return 0;
927 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
928 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
929 		return 0;
930 	return 1;
931 }
932 
933 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
934 					      u32 function, u32 index)
935 {
936 	int i;
937 	struct kvm_cpuid_entry2 *best = NULL;
938 
939 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
940 		struct kvm_cpuid_entry2 *e;
941 
942 		e = &vcpu->arch.cpuid_entries[i];
943 		if (is_matching_cpuid_entry(e, function, index)) {
944 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
945 				move_to_next_stateful_cpuid_entry(vcpu, i);
946 			best = e;
947 			break;
948 		}
949 	}
950 	return best;
951 }
952 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
953 
954 /*
955  * If no match is found, check whether we exceed the vCPU's limit
956  * and return the content of the highest valid _standard_ leaf instead.
957  * This is to satisfy the CPUID specification.
958  */
959 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
960                                                   u32 function, u32 index)
961 {
962 	struct kvm_cpuid_entry2 *maxlevel;
963 
964 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
965 	if (!maxlevel || maxlevel->eax >= function)
966 		return NULL;
967 	if (function & 0x80000000) {
968 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
969 		if (!maxlevel)
970 			return NULL;
971 	}
972 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
973 }
974 
975 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
976 	       u32 *ecx, u32 *edx, bool check_limit)
977 {
978 	u32 function = *eax, index = *ecx;
979 	struct kvm_cpuid_entry2 *best;
980 	bool entry_found = true;
981 
982 	best = kvm_find_cpuid_entry(vcpu, function, index);
983 
984 	if (!best) {
985 		entry_found = false;
986 		if (!check_limit)
987 			goto out;
988 
989 		best = check_cpuid_limit(vcpu, function, index);
990 	}
991 
992 out:
993 	if (best) {
994 		*eax = best->eax;
995 		*ebx = best->ebx;
996 		*ecx = best->ecx;
997 		*edx = best->edx;
998 	} else
999 		*eax = *ebx = *ecx = *edx = 0;
1000 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
1001 	return entry_found;
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_cpuid);
1004 
1005 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1006 {
1007 	u32 eax, ebx, ecx, edx;
1008 
1009 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1010 		return 1;
1011 
1012 	eax = kvm_rax_read(vcpu);
1013 	ecx = kvm_rcx_read(vcpu);
1014 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
1015 	kvm_rax_write(vcpu, eax);
1016 	kvm_rbx_write(vcpu, ebx);
1017 	kvm_rcx_write(vcpu, ecx);
1018 	kvm_rdx_write(vcpu, edx);
1019 	return kvm_skip_emulated_instruction(vcpu);
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1022