xref: /linux/arch/x86/kvm/cpuid.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  * cpuid support routines
4  *
5  * derived from arch/x86/kvm/x86.c
6  *
7  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
8  * Copyright IBM Corporation, 2008
9  *
10  * This work is licensed under the terms of the GNU GPL, version 2.  See
11  * the COPYING file in the top-level directory.
12  *
13  */
14 
15 #include <linux/kvm_host.h>
16 #include <linux/export.h>
17 #include <linux/vmalloc.h>
18 #include <linux/uaccess.h>
19 #include <linux/sched/stat.h>
20 
21 #include <asm/processor.h>
22 #include <asm/user.h>
23 #include <asm/fpu/xstate.h>
24 #include "cpuid.h"
25 #include "lapic.h"
26 #include "mmu.h"
27 #include "trace.h"
28 #include "pmu.h"
29 
30 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
31 {
32 	int feature_bit = 0;
33 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
34 
35 	xstate_bv &= XFEATURE_MASK_EXTEND;
36 	while (xstate_bv) {
37 		if (xstate_bv & 0x1) {
38 		        u32 eax, ebx, ecx, edx, offset;
39 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
40 			offset = compacted ? ret : ebx;
41 			ret = max(ret, offset + eax);
42 		}
43 
44 		xstate_bv >>= 1;
45 		feature_bit++;
46 	}
47 
48 	return ret;
49 }
50 
51 bool kvm_mpx_supported(void)
52 {
53 	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
54 		 && kvm_x86_ops->mpx_supported());
55 }
56 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
57 
58 u64 kvm_supported_xcr0(void)
59 {
60 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
61 
62 	if (!kvm_mpx_supported())
63 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
64 
65 	return xcr0;
66 }
67 
68 #define F(x) bit(X86_FEATURE_##x)
69 
70 /* For scattered features from cpufeatures.h; we currently expose none */
71 #define KF(x) bit(KVM_CPUID_BIT_##x)
72 
73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
74 {
75 	struct kvm_cpuid_entry2 *best;
76 	struct kvm_lapic *apic = vcpu->arch.apic;
77 
78 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
79 	if (!best)
80 		return 0;
81 
82 	/* Update OSXSAVE bit */
83 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
84 		best->ecx &= ~F(OSXSAVE);
85 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
86 			best->ecx |= F(OSXSAVE);
87 	}
88 
89 	best->edx &= ~F(APIC);
90 	if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
91 		best->edx |= F(APIC);
92 
93 	if (apic) {
94 		if (best->ecx & F(TSC_DEADLINE_TIMER))
95 			apic->lapic_timer.timer_mode_mask = 3 << 17;
96 		else
97 			apic->lapic_timer.timer_mode_mask = 1 << 17;
98 	}
99 
100 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
101 	if (best) {
102 		/* Update OSPKE bit */
103 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
104 			best->ecx &= ~F(OSPKE);
105 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
106 				best->ecx |= F(OSPKE);
107 		}
108 	}
109 
110 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
111 	if (!best) {
112 		vcpu->arch.guest_supported_xcr0 = 0;
113 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
114 	} else {
115 		vcpu->arch.guest_supported_xcr0 =
116 			(best->eax | ((u64)best->edx << 32)) &
117 			kvm_supported_xcr0();
118 		vcpu->arch.guest_xstate_size = best->ebx =
119 			xstate_required_size(vcpu->arch.xcr0, false);
120 	}
121 
122 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
123 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
124 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
125 
126 	/*
127 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
128 	 * canonical address checks; exit if it is ever changed.
129 	 */
130 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
131 	if (best) {
132 		int vaddr_bits = (best->eax & 0xff00) >> 8;
133 
134 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
135 			return -EINVAL;
136 	}
137 
138 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
139 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
140 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
141 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
142 
143 	/* Update physical-address width */
144 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
145 	kvm_mmu_reset_context(vcpu);
146 
147 	kvm_pmu_refresh(vcpu);
148 	return 0;
149 }
150 
151 static int is_efer_nx(void)
152 {
153 	unsigned long long efer = 0;
154 
155 	rdmsrl_safe(MSR_EFER, &efer);
156 	return efer & EFER_NX;
157 }
158 
159 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
160 {
161 	int i;
162 	struct kvm_cpuid_entry2 *e, *entry;
163 
164 	entry = NULL;
165 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
166 		e = &vcpu->arch.cpuid_entries[i];
167 		if (e->function == 0x80000001) {
168 			entry = e;
169 			break;
170 		}
171 	}
172 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
173 		entry->edx &= ~F(NX);
174 		printk(KERN_INFO "kvm: guest NX capability removed\n");
175 	}
176 }
177 
178 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
179 {
180 	struct kvm_cpuid_entry2 *best;
181 
182 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
183 	if (!best || best->eax < 0x80000008)
184 		goto not_found;
185 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
186 	if (best)
187 		return best->eax & 0xff;
188 not_found:
189 	return 36;
190 }
191 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
192 
193 /* when an old userspace process fills a new kernel module */
194 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
195 			     struct kvm_cpuid *cpuid,
196 			     struct kvm_cpuid_entry __user *entries)
197 {
198 	int r, i;
199 	struct kvm_cpuid_entry *cpuid_entries = NULL;
200 
201 	r = -E2BIG;
202 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
203 		goto out;
204 	r = -ENOMEM;
205 	if (cpuid->nent) {
206 		cpuid_entries =
207 			vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
208 					   cpuid->nent));
209 		if (!cpuid_entries)
210 			goto out;
211 		r = -EFAULT;
212 		if (copy_from_user(cpuid_entries, entries,
213 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
214 			goto out;
215 	}
216 	for (i = 0; i < cpuid->nent; i++) {
217 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
218 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
219 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
220 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
221 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
222 		vcpu->arch.cpuid_entries[i].index = 0;
223 		vcpu->arch.cpuid_entries[i].flags = 0;
224 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
225 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
226 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
227 	}
228 	vcpu->arch.cpuid_nent = cpuid->nent;
229 	cpuid_fix_nx_cap(vcpu);
230 	kvm_apic_set_version(vcpu);
231 	kvm_x86_ops->cpuid_update(vcpu);
232 	r = kvm_update_cpuid(vcpu);
233 
234 out:
235 	vfree(cpuid_entries);
236 	return r;
237 }
238 
239 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
240 			      struct kvm_cpuid2 *cpuid,
241 			      struct kvm_cpuid_entry2 __user *entries)
242 {
243 	int r;
244 
245 	r = -E2BIG;
246 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
247 		goto out;
248 	r = -EFAULT;
249 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
250 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
251 		goto out;
252 	vcpu->arch.cpuid_nent = cpuid->nent;
253 	kvm_apic_set_version(vcpu);
254 	kvm_x86_ops->cpuid_update(vcpu);
255 	r = kvm_update_cpuid(vcpu);
256 out:
257 	return r;
258 }
259 
260 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
261 			      struct kvm_cpuid2 *cpuid,
262 			      struct kvm_cpuid_entry2 __user *entries)
263 {
264 	int r;
265 
266 	r = -E2BIG;
267 	if (cpuid->nent < vcpu->arch.cpuid_nent)
268 		goto out;
269 	r = -EFAULT;
270 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
271 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
272 		goto out;
273 	return 0;
274 
275 out:
276 	cpuid->nent = vcpu->arch.cpuid_nent;
277 	return r;
278 }
279 
280 static void cpuid_mask(u32 *word, int wordnum)
281 {
282 	*word &= boot_cpu_data.x86_capability[wordnum];
283 }
284 
285 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
286 			   u32 index)
287 {
288 	entry->function = function;
289 	entry->index = index;
290 	cpuid_count(entry->function, entry->index,
291 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
292 	entry->flags = 0;
293 }
294 
295 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
296 				   u32 func, u32 index, int *nent, int maxnent)
297 {
298 	switch (func) {
299 	case 0:
300 		entry->eax = 7;
301 		++*nent;
302 		break;
303 	case 1:
304 		entry->ecx = F(MOVBE);
305 		++*nent;
306 		break;
307 	case 7:
308 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
309 		if (index == 0)
310 			entry->ecx = F(RDPID);
311 		++*nent;
312 	default:
313 		break;
314 	}
315 
316 	entry->function = func;
317 	entry->index = index;
318 
319 	return 0;
320 }
321 
322 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
323 				 u32 index, int *nent, int maxnent)
324 {
325 	int r;
326 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
327 #ifdef CONFIG_X86_64
328 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
329 				? F(GBPAGES) : 0;
330 	unsigned f_lm = F(LM);
331 #else
332 	unsigned f_gbpages = 0;
333 	unsigned f_lm = 0;
334 #endif
335 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
336 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
337 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
338 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
339 	unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
340 
341 	/* cpuid 1.edx */
342 	const u32 kvm_cpuid_1_edx_x86_features =
343 		F(FPU) | F(VME) | F(DE) | F(PSE) |
344 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
345 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
346 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
347 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
348 		0 /* Reserved, DS, ACPI */ | F(MMX) |
349 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
350 		0 /* HTT, TM, Reserved, PBE */;
351 	/* cpuid 0x80000001.edx */
352 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
353 		F(FPU) | F(VME) | F(DE) | F(PSE) |
354 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
355 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
356 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
357 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
358 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
359 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
360 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
361 	/* cpuid 1.ecx */
362 	const u32 kvm_cpuid_1_ecx_x86_features =
363 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
364 		 * but *not* advertised to guests via CPUID ! */
365 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
366 		0 /* DS-CPL, VMX, SMX, EST */ |
367 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
368 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
369 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
370 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
371 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
372 		F(F16C) | F(RDRAND);
373 	/* cpuid 0x80000001.ecx */
374 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
375 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
376 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
377 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
378 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
379 		F(TOPOEXT) | F(PERFCTR_CORE);
380 
381 	/* cpuid 0x80000008.ebx */
382 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
383 		F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
384 		F(AMD_SSB_NO);
385 
386 	/* cpuid 0xC0000001.edx */
387 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
388 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
389 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
390 		F(PMM) | F(PMM_EN);
391 
392 	/* cpuid 7.0.ebx */
393 	const u32 kvm_cpuid_7_0_ebx_x86_features =
394 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
395 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
396 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
397 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
398 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
399 
400 	/* cpuid 0xD.1.eax */
401 	const u32 kvm_cpuid_D_1_eax_x86_features =
402 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
403 
404 	/* cpuid 7.0.ecx*/
405 	const u32 kvm_cpuid_7_0_ecx_x86_features =
406 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
407 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
408 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
409 		F(CLDEMOTE);
410 
411 	/* cpuid 7.0.edx*/
412 	const u32 kvm_cpuid_7_0_edx_x86_features =
413 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
414 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES);
415 
416 	/* all calls to cpuid_count() should be made on the same cpu */
417 	get_cpu();
418 
419 	r = -E2BIG;
420 
421 	if (*nent >= maxnent)
422 		goto out;
423 
424 	do_cpuid_1_ent(entry, function, index);
425 	++*nent;
426 
427 	switch (function) {
428 	case 0:
429 		entry->eax = min(entry->eax, (u32)0xd);
430 		break;
431 	case 1:
432 		entry->edx &= kvm_cpuid_1_edx_x86_features;
433 		cpuid_mask(&entry->edx, CPUID_1_EDX);
434 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
435 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
436 		/* we support x2apic emulation even if host does not support
437 		 * it since we emulate x2apic in software */
438 		entry->ecx |= F(X2APIC);
439 		break;
440 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
441 	 * may return different values. This forces us to get_cpu() before
442 	 * issuing the first command, and also to emulate this annoying behavior
443 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
444 	case 2: {
445 		int t, times = entry->eax & 0xff;
446 
447 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
448 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
449 		for (t = 1; t < times; ++t) {
450 			if (*nent >= maxnent)
451 				goto out;
452 
453 			do_cpuid_1_ent(&entry[t], function, 0);
454 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
455 			++*nent;
456 		}
457 		break;
458 	}
459 	/* function 4 has additional index. */
460 	case 4: {
461 		int i, cache_type;
462 
463 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
464 		/* read more entries until cache_type is zero */
465 		for (i = 1; ; ++i) {
466 			if (*nent >= maxnent)
467 				goto out;
468 
469 			cache_type = entry[i - 1].eax & 0x1f;
470 			if (!cache_type)
471 				break;
472 			do_cpuid_1_ent(&entry[i], function, i);
473 			entry[i].flags |=
474 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
475 			++*nent;
476 		}
477 		break;
478 	}
479 	case 6: /* Thermal management */
480 		entry->eax = 0x4; /* allow ARAT */
481 		entry->ebx = 0;
482 		entry->ecx = 0;
483 		entry->edx = 0;
484 		break;
485 	case 7: {
486 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
487 		/* Mask ebx against host capability word 9 */
488 		if (index == 0) {
489 			entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
490 			cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
491 			// TSC_ADJUST is emulated
492 			entry->ebx |= F(TSC_ADJUST);
493 			entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
494 			cpuid_mask(&entry->ecx, CPUID_7_ECX);
495 			entry->ecx |= f_umip;
496 			/* PKU is not yet implemented for shadow paging. */
497 			if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
498 				entry->ecx &= ~F(PKU);
499 			entry->edx &= kvm_cpuid_7_0_edx_x86_features;
500 			cpuid_mask(&entry->edx, CPUID_7_EDX);
501 			/*
502 			 * We emulate ARCH_CAPABILITIES in software even
503 			 * if the host doesn't support it.
504 			 */
505 			entry->edx |= F(ARCH_CAPABILITIES);
506 		} else {
507 			entry->ebx = 0;
508 			entry->ecx = 0;
509 			entry->edx = 0;
510 		}
511 		entry->eax = 0;
512 		break;
513 	}
514 	case 9:
515 		break;
516 	case 0xa: { /* Architectural Performance Monitoring */
517 		struct x86_pmu_capability cap;
518 		union cpuid10_eax eax;
519 		union cpuid10_edx edx;
520 
521 		perf_get_x86_pmu_capability(&cap);
522 
523 		/*
524 		 * Only support guest architectural pmu on a host
525 		 * with architectural pmu.
526 		 */
527 		if (!cap.version)
528 			memset(&cap, 0, sizeof(cap));
529 
530 		eax.split.version_id = min(cap.version, 2);
531 		eax.split.num_counters = cap.num_counters_gp;
532 		eax.split.bit_width = cap.bit_width_gp;
533 		eax.split.mask_length = cap.events_mask_len;
534 
535 		edx.split.num_counters_fixed = cap.num_counters_fixed;
536 		edx.split.bit_width_fixed = cap.bit_width_fixed;
537 		edx.split.reserved = 0;
538 
539 		entry->eax = eax.full;
540 		entry->ebx = cap.events_mask;
541 		entry->ecx = 0;
542 		entry->edx = edx.full;
543 		break;
544 	}
545 	/* function 0xb has additional index. */
546 	case 0xb: {
547 		int i, level_type;
548 
549 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
550 		/* read more entries until level_type is zero */
551 		for (i = 1; ; ++i) {
552 			if (*nent >= maxnent)
553 				goto out;
554 
555 			level_type = entry[i - 1].ecx & 0xff00;
556 			if (!level_type)
557 				break;
558 			do_cpuid_1_ent(&entry[i], function, i);
559 			entry[i].flags |=
560 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
561 			++*nent;
562 		}
563 		break;
564 	}
565 	case 0xd: {
566 		int idx, i;
567 		u64 supported = kvm_supported_xcr0();
568 
569 		entry->eax &= supported;
570 		entry->ebx = xstate_required_size(supported, false);
571 		entry->ecx = entry->ebx;
572 		entry->edx &= supported >> 32;
573 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
574 		if (!supported)
575 			break;
576 
577 		for (idx = 1, i = 1; idx < 64; ++idx) {
578 			u64 mask = ((u64)1 << idx);
579 			if (*nent >= maxnent)
580 				goto out;
581 
582 			do_cpuid_1_ent(&entry[i], function, idx);
583 			if (idx == 1) {
584 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
585 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
586 				entry[i].ebx = 0;
587 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
588 					entry[i].ebx =
589 						xstate_required_size(supported,
590 								     true);
591 			} else {
592 				if (entry[i].eax == 0 || !(supported & mask))
593 					continue;
594 				if (WARN_ON_ONCE(entry[i].ecx & 1))
595 					continue;
596 			}
597 			entry[i].ecx = 0;
598 			entry[i].edx = 0;
599 			entry[i].flags |=
600 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
601 			++*nent;
602 			++i;
603 		}
604 		break;
605 	}
606 	case KVM_CPUID_SIGNATURE: {
607 		static const char signature[12] = "KVMKVMKVM\0\0";
608 		const u32 *sigptr = (const u32 *)signature;
609 		entry->eax = KVM_CPUID_FEATURES;
610 		entry->ebx = sigptr[0];
611 		entry->ecx = sigptr[1];
612 		entry->edx = sigptr[2];
613 		break;
614 	}
615 	case KVM_CPUID_FEATURES:
616 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
617 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
618 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
619 			     (1 << KVM_FEATURE_ASYNC_PF) |
620 			     (1 << KVM_FEATURE_PV_EOI) |
621 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
622 			     (1 << KVM_FEATURE_PV_UNHALT) |
623 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
624 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
625 			     (1 << KVM_FEATURE_PV_SEND_IPI);
626 
627 		if (sched_info_on())
628 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
629 
630 		entry->ebx = 0;
631 		entry->ecx = 0;
632 		entry->edx = 0;
633 		break;
634 	case 0x80000000:
635 		entry->eax = min(entry->eax, 0x8000001f);
636 		break;
637 	case 0x80000001:
638 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
639 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
640 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
641 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
642 		break;
643 	case 0x80000007: /* Advanced power management */
644 		/* invariant TSC is CPUID.80000007H:EDX[8] */
645 		entry->edx &= (1 << 8);
646 		/* mask against host */
647 		entry->edx &= boot_cpu_data.x86_power;
648 		entry->eax = entry->ebx = entry->ecx = 0;
649 		break;
650 	case 0x80000008: {
651 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
652 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
653 		unsigned phys_as = entry->eax & 0xff;
654 
655 		if (!g_phys_as)
656 			g_phys_as = phys_as;
657 		entry->eax = g_phys_as | (virt_as << 8);
658 		entry->edx = 0;
659 		/*
660 		 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
661 		 * hardware cpuid
662 		 */
663 		if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
664 			entry->ebx |= F(AMD_IBPB);
665 		if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
666 			entry->ebx |= F(AMD_IBRS);
667 		if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
668 			entry->ebx |= F(VIRT_SSBD);
669 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
670 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
671 		/*
672 		 * The preference is to use SPEC CTRL MSR instead of the
673 		 * VIRT_SPEC MSR.
674 		 */
675 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
676 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
677 			entry->ebx |= F(VIRT_SSBD);
678 		break;
679 	}
680 	case 0x80000019:
681 		entry->ecx = entry->edx = 0;
682 		break;
683 	case 0x8000001a:
684 		break;
685 	case 0x8000001d:
686 		break;
687 	/*Add support for Centaur's CPUID instruction*/
688 	case 0xC0000000:
689 		/*Just support up to 0xC0000004 now*/
690 		entry->eax = min(entry->eax, 0xC0000004);
691 		break;
692 	case 0xC0000001:
693 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
694 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
695 		break;
696 	case 3: /* Processor serial number */
697 	case 5: /* MONITOR/MWAIT */
698 	case 0xC0000002:
699 	case 0xC0000003:
700 	case 0xC0000004:
701 	default:
702 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
703 		break;
704 	}
705 
706 	kvm_x86_ops->set_supported_cpuid(function, entry);
707 
708 	r = 0;
709 
710 out:
711 	put_cpu();
712 
713 	return r;
714 }
715 
716 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
717 			u32 idx, int *nent, int maxnent, unsigned int type)
718 {
719 	if (type == KVM_GET_EMULATED_CPUID)
720 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
721 
722 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
723 }
724 
725 #undef F
726 
727 struct kvm_cpuid_param {
728 	u32 func;
729 	u32 idx;
730 	bool has_leaf_count;
731 	bool (*qualifier)(const struct kvm_cpuid_param *param);
732 };
733 
734 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
735 {
736 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
737 }
738 
739 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
740 				 __u32 num_entries, unsigned int ioctl_type)
741 {
742 	int i;
743 	__u32 pad[3];
744 
745 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
746 		return false;
747 
748 	/*
749 	 * We want to make sure that ->padding is being passed clean from
750 	 * userspace in case we want to use it for something in the future.
751 	 *
752 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
753 	 * have to give ourselves satisfied only with the emulated side. /me
754 	 * sheds a tear.
755 	 */
756 	for (i = 0; i < num_entries; i++) {
757 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
758 			return true;
759 
760 		if (pad[0] || pad[1] || pad[2])
761 			return true;
762 	}
763 	return false;
764 }
765 
766 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
767 			    struct kvm_cpuid_entry2 __user *entries,
768 			    unsigned int type)
769 {
770 	struct kvm_cpuid_entry2 *cpuid_entries;
771 	int limit, nent = 0, r = -E2BIG, i;
772 	u32 func;
773 	static const struct kvm_cpuid_param param[] = {
774 		{ .func = 0, .has_leaf_count = true },
775 		{ .func = 0x80000000, .has_leaf_count = true },
776 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
777 		{ .func = KVM_CPUID_SIGNATURE },
778 		{ .func = KVM_CPUID_FEATURES },
779 	};
780 
781 	if (cpuid->nent < 1)
782 		goto out;
783 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
784 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
785 
786 	if (sanity_check_entries(entries, cpuid->nent, type))
787 		return -EINVAL;
788 
789 	r = -ENOMEM;
790 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
791 					   cpuid->nent));
792 	if (!cpuid_entries)
793 		goto out;
794 
795 	r = 0;
796 	for (i = 0; i < ARRAY_SIZE(param); i++) {
797 		const struct kvm_cpuid_param *ent = &param[i];
798 
799 		if (ent->qualifier && !ent->qualifier(ent))
800 			continue;
801 
802 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
803 				&nent, cpuid->nent, type);
804 
805 		if (r)
806 			goto out_free;
807 
808 		if (!ent->has_leaf_count)
809 			continue;
810 
811 		limit = cpuid_entries[nent - 1].eax;
812 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
813 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
814 				     &nent, cpuid->nent, type);
815 
816 		if (r)
817 			goto out_free;
818 	}
819 
820 	r = -EFAULT;
821 	if (copy_to_user(entries, cpuid_entries,
822 			 nent * sizeof(struct kvm_cpuid_entry2)))
823 		goto out_free;
824 	cpuid->nent = nent;
825 	r = 0;
826 
827 out_free:
828 	vfree(cpuid_entries);
829 out:
830 	return r;
831 }
832 
833 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
834 {
835 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
836 	struct kvm_cpuid_entry2 *ej;
837 	int j = i;
838 	int nent = vcpu->arch.cpuid_nent;
839 
840 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
841 	/* when no next entry is found, the current entry[i] is reselected */
842 	do {
843 		j = (j + 1) % nent;
844 		ej = &vcpu->arch.cpuid_entries[j];
845 	} while (ej->function != e->function);
846 
847 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
848 
849 	return j;
850 }
851 
852 /* find an entry with matching function, matching index (if needed), and that
853  * should be read next (if it's stateful) */
854 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
855 	u32 function, u32 index)
856 {
857 	if (e->function != function)
858 		return 0;
859 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
860 		return 0;
861 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
862 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
863 		return 0;
864 	return 1;
865 }
866 
867 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
868 					      u32 function, u32 index)
869 {
870 	int i;
871 	struct kvm_cpuid_entry2 *best = NULL;
872 
873 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
874 		struct kvm_cpuid_entry2 *e;
875 
876 		e = &vcpu->arch.cpuid_entries[i];
877 		if (is_matching_cpuid_entry(e, function, index)) {
878 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
879 				move_to_next_stateful_cpuid_entry(vcpu, i);
880 			best = e;
881 			break;
882 		}
883 	}
884 	return best;
885 }
886 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
887 
888 /*
889  * If no match is found, check whether we exceed the vCPU's limit
890  * and return the content of the highest valid _standard_ leaf instead.
891  * This is to satisfy the CPUID specification.
892  */
893 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
894                                                   u32 function, u32 index)
895 {
896 	struct kvm_cpuid_entry2 *maxlevel;
897 
898 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
899 	if (!maxlevel || maxlevel->eax >= function)
900 		return NULL;
901 	if (function & 0x80000000) {
902 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
903 		if (!maxlevel)
904 			return NULL;
905 	}
906 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
907 }
908 
909 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
910 	       u32 *ecx, u32 *edx, bool check_limit)
911 {
912 	u32 function = *eax, index = *ecx;
913 	struct kvm_cpuid_entry2 *best;
914 	bool entry_found = true;
915 
916 	best = kvm_find_cpuid_entry(vcpu, function, index);
917 
918 	if (!best) {
919 		entry_found = false;
920 		if (!check_limit)
921 			goto out;
922 
923 		best = check_cpuid_limit(vcpu, function, index);
924 	}
925 
926 out:
927 	if (best) {
928 		*eax = best->eax;
929 		*ebx = best->ebx;
930 		*ecx = best->ecx;
931 		*edx = best->edx;
932 	} else
933 		*eax = *ebx = *ecx = *edx = 0;
934 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
935 	return entry_found;
936 }
937 EXPORT_SYMBOL_GPL(kvm_cpuid);
938 
939 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
940 {
941 	u32 eax, ebx, ecx, edx;
942 
943 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
944 		return 1;
945 
946 	eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
947 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
948 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
949 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
950 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
951 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
952 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
953 	return kvm_skip_emulated_instruction(vcpu);
954 }
955 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
956