xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 3213486f)
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 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
71 {
72 	struct kvm_cpuid_entry2 *best;
73 	struct kvm_lapic *apic = vcpu->arch.apic;
74 
75 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
76 	if (!best)
77 		return 0;
78 
79 	/* Update OSXSAVE bit */
80 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
81 		best->ecx &= ~F(OSXSAVE);
82 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
83 			best->ecx |= F(OSXSAVE);
84 	}
85 
86 	best->edx &= ~F(APIC);
87 	if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
88 		best->edx |= F(APIC);
89 
90 	if (apic) {
91 		if (best->ecx & F(TSC_DEADLINE_TIMER))
92 			apic->lapic_timer.timer_mode_mask = 3 << 17;
93 		else
94 			apic->lapic_timer.timer_mode_mask = 1 << 17;
95 	}
96 
97 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
98 	if (best) {
99 		/* Update OSPKE bit */
100 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
101 			best->ecx &= ~F(OSPKE);
102 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
103 				best->ecx |= F(OSPKE);
104 		}
105 	}
106 
107 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
108 	if (!best) {
109 		vcpu->arch.guest_supported_xcr0 = 0;
110 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
111 	} else {
112 		vcpu->arch.guest_supported_xcr0 =
113 			(best->eax | ((u64)best->edx << 32)) &
114 			kvm_supported_xcr0();
115 		vcpu->arch.guest_xstate_size = best->ebx =
116 			xstate_required_size(vcpu->arch.xcr0, false);
117 	}
118 
119 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
120 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
121 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
122 
123 	/*
124 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
125 	 * canonical address checks; exit if it is ever changed.
126 	 */
127 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
128 	if (best) {
129 		int vaddr_bits = (best->eax & 0xff00) >> 8;
130 
131 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
132 			return -EINVAL;
133 	}
134 
135 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
136 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
137 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
138 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
139 
140 	/* Update physical-address width */
141 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
142 	kvm_mmu_reset_context(vcpu);
143 
144 	kvm_pmu_refresh(vcpu);
145 	return 0;
146 }
147 
148 static int is_efer_nx(void)
149 {
150 	unsigned long long efer = 0;
151 
152 	rdmsrl_safe(MSR_EFER, &efer);
153 	return efer & EFER_NX;
154 }
155 
156 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
157 {
158 	int i;
159 	struct kvm_cpuid_entry2 *e, *entry;
160 
161 	entry = NULL;
162 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
163 		e = &vcpu->arch.cpuid_entries[i];
164 		if (e->function == 0x80000001) {
165 			entry = e;
166 			break;
167 		}
168 	}
169 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
170 		entry->edx &= ~F(NX);
171 		printk(KERN_INFO "kvm: guest NX capability removed\n");
172 	}
173 }
174 
175 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
176 {
177 	struct kvm_cpuid_entry2 *best;
178 
179 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
180 	if (!best || best->eax < 0x80000008)
181 		goto not_found;
182 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
183 	if (best)
184 		return best->eax & 0xff;
185 not_found:
186 	return 36;
187 }
188 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
189 
190 /* when an old userspace process fills a new kernel module */
191 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
192 			     struct kvm_cpuid *cpuid,
193 			     struct kvm_cpuid_entry __user *entries)
194 {
195 	int r, i;
196 	struct kvm_cpuid_entry *cpuid_entries = NULL;
197 
198 	r = -E2BIG;
199 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
200 		goto out;
201 	r = -ENOMEM;
202 	if (cpuid->nent) {
203 		cpuid_entries =
204 			vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
205 					   cpuid->nent));
206 		if (!cpuid_entries)
207 			goto out;
208 		r = -EFAULT;
209 		if (copy_from_user(cpuid_entries, entries,
210 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
211 			goto out;
212 	}
213 	for (i = 0; i < cpuid->nent; i++) {
214 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
215 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
216 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
217 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
218 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
219 		vcpu->arch.cpuid_entries[i].index = 0;
220 		vcpu->arch.cpuid_entries[i].flags = 0;
221 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
222 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
223 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
224 	}
225 	vcpu->arch.cpuid_nent = cpuid->nent;
226 	cpuid_fix_nx_cap(vcpu);
227 	kvm_apic_set_version(vcpu);
228 	kvm_x86_ops->cpuid_update(vcpu);
229 	r = kvm_update_cpuid(vcpu);
230 
231 out:
232 	vfree(cpuid_entries);
233 	return r;
234 }
235 
236 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
237 			      struct kvm_cpuid2 *cpuid,
238 			      struct kvm_cpuid_entry2 __user *entries)
239 {
240 	int r;
241 
242 	r = -E2BIG;
243 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
244 		goto out;
245 	r = -EFAULT;
246 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
247 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
248 		goto out;
249 	vcpu->arch.cpuid_nent = cpuid->nent;
250 	kvm_apic_set_version(vcpu);
251 	kvm_x86_ops->cpuid_update(vcpu);
252 	r = kvm_update_cpuid(vcpu);
253 out:
254 	return r;
255 }
256 
257 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
258 			      struct kvm_cpuid2 *cpuid,
259 			      struct kvm_cpuid_entry2 __user *entries)
260 {
261 	int r;
262 
263 	r = -E2BIG;
264 	if (cpuid->nent < vcpu->arch.cpuid_nent)
265 		goto out;
266 	r = -EFAULT;
267 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
268 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
269 		goto out;
270 	return 0;
271 
272 out:
273 	cpuid->nent = vcpu->arch.cpuid_nent;
274 	return r;
275 }
276 
277 static void cpuid_mask(u32 *word, int wordnum)
278 {
279 	*word &= boot_cpu_data.x86_capability[wordnum];
280 }
281 
282 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
283 			   u32 index)
284 {
285 	entry->function = function;
286 	entry->index = index;
287 	cpuid_count(entry->function, entry->index,
288 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
289 	entry->flags = 0;
290 }
291 
292 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
293 				   u32 func, u32 index, int *nent, int maxnent)
294 {
295 	switch (func) {
296 	case 0:
297 		entry->eax = 7;
298 		++*nent;
299 		break;
300 	case 1:
301 		entry->ecx = F(MOVBE);
302 		++*nent;
303 		break;
304 	case 7:
305 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
306 		if (index == 0)
307 			entry->ecx = F(RDPID);
308 		++*nent;
309 	default:
310 		break;
311 	}
312 
313 	entry->function = func;
314 	entry->index = index;
315 
316 	return 0;
317 }
318 
319 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
320 				 u32 index, int *nent, int maxnent)
321 {
322 	int r;
323 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
324 #ifdef CONFIG_X86_64
325 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
326 				? F(GBPAGES) : 0;
327 	unsigned f_lm = F(LM);
328 #else
329 	unsigned f_gbpages = 0;
330 	unsigned f_lm = 0;
331 #endif
332 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
333 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
334 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
335 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
336 	unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
337 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
338 	unsigned f_la57 = 0;
339 
340 	/* cpuid 1.edx */
341 	const u32 kvm_cpuid_1_edx_x86_features =
342 		F(FPU) | F(VME) | F(DE) | F(PSE) |
343 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
344 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
345 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
346 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
347 		0 /* Reserved, DS, ACPI */ | F(MMX) |
348 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
349 		0 /* HTT, TM, Reserved, PBE */;
350 	/* cpuid 0x80000001.edx */
351 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
352 		F(FPU) | F(VME) | F(DE) | F(PSE) |
353 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
354 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
355 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
356 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
357 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
358 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
359 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
360 	/* cpuid 1.ecx */
361 	const u32 kvm_cpuid_1_ecx_x86_features =
362 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
363 		 * but *not* advertised to guests via CPUID ! */
364 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
365 		0 /* DS-CPL, VMX, SMX, EST */ |
366 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
367 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
368 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
369 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
370 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
371 		F(F16C) | F(RDRAND);
372 	/* cpuid 0x80000001.ecx */
373 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
374 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
375 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
376 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
377 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
378 		F(TOPOEXT) | F(PERFCTR_CORE);
379 
380 	/* cpuid 0x80000008.ebx */
381 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
382 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
383 		F(AMD_SSB_NO) | F(AMD_STIBP);
384 
385 	/* cpuid 0xC0000001.edx */
386 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
387 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
388 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
389 		F(PMM) | F(PMM_EN);
390 
391 	/* cpuid 7.0.ebx */
392 	const u32 kvm_cpuid_7_0_ebx_x86_features =
393 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
394 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
395 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
396 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
397 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
398 
399 	/* cpuid 0xD.1.eax */
400 	const u32 kvm_cpuid_D_1_eax_x86_features =
401 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
402 
403 	/* cpuid 7.0.ecx*/
404 	const u32 kvm_cpuid_7_0_ecx_x86_features =
405 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
406 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
407 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
408 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B);
409 
410 	/* cpuid 7.0.edx*/
411 	const u32 kvm_cpuid_7_0_edx_x86_features =
412 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
413 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP);
414 
415 	/* all calls to cpuid_count() should be made on the same cpu */
416 	get_cpu();
417 
418 	r = -E2BIG;
419 
420 	if (*nent >= maxnent)
421 		goto out;
422 
423 	do_cpuid_1_ent(entry, function, index);
424 	++*nent;
425 
426 	switch (function) {
427 	case 0:
428 		entry->eax = min(entry->eax, (u32)(f_intel_pt ? 0x14 : 0xd));
429 		break;
430 	case 1:
431 		entry->edx &= kvm_cpuid_1_edx_x86_features;
432 		cpuid_mask(&entry->edx, CPUID_1_EDX);
433 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
434 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
435 		/* we support x2apic emulation even if host does not support
436 		 * it since we emulate x2apic in software */
437 		entry->ecx |= F(X2APIC);
438 		break;
439 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
440 	 * may return different values. This forces us to get_cpu() before
441 	 * issuing the first command, and also to emulate this annoying behavior
442 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
443 	case 2: {
444 		int t, times = entry->eax & 0xff;
445 
446 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
447 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
448 		for (t = 1; t < times; ++t) {
449 			if (*nent >= maxnent)
450 				goto out;
451 
452 			do_cpuid_1_ent(&entry[t], function, 0);
453 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
454 			++*nent;
455 		}
456 		break;
457 	}
458 	/* function 4 has additional index. */
459 	case 4: {
460 		int i, cache_type;
461 
462 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
463 		/* read more entries until cache_type is zero */
464 		for (i = 1; ; ++i) {
465 			if (*nent >= maxnent)
466 				goto out;
467 
468 			cache_type = entry[i - 1].eax & 0x1f;
469 			if (!cache_type)
470 				break;
471 			do_cpuid_1_ent(&entry[i], function, i);
472 			entry[i].flags |=
473 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
474 			++*nent;
475 		}
476 		break;
477 	}
478 	case 6: /* Thermal management */
479 		entry->eax = 0x4; /* allow ARAT */
480 		entry->ebx = 0;
481 		entry->ecx = 0;
482 		entry->edx = 0;
483 		break;
484 	case 7: {
485 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
486 		/* Mask ebx against host capability word 9 */
487 		if (index == 0) {
488 			entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
489 			cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
490 			// TSC_ADJUST is emulated
491 			entry->ebx |= F(TSC_ADJUST);
492 			entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
493 			f_la57 = entry->ecx & F(LA57);
494 			cpuid_mask(&entry->ecx, CPUID_7_ECX);
495 			/* Set LA57 based on hardware capability. */
496 			entry->ecx |= f_la57;
497 			entry->ecx |= f_umip;
498 			/* PKU is not yet implemented for shadow paging. */
499 			if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
500 				entry->ecx &= ~F(PKU);
501 			entry->edx &= kvm_cpuid_7_0_edx_x86_features;
502 			cpuid_mask(&entry->edx, CPUID_7_EDX);
503 			/*
504 			 * We emulate ARCH_CAPABILITIES in software even
505 			 * if the host doesn't support it.
506 			 */
507 			entry->edx |= F(ARCH_CAPABILITIES);
508 		} else {
509 			entry->ebx = 0;
510 			entry->ecx = 0;
511 			entry->edx = 0;
512 		}
513 		entry->eax = 0;
514 		break;
515 	}
516 	case 9:
517 		break;
518 	case 0xa: { /* Architectural Performance Monitoring */
519 		struct x86_pmu_capability cap;
520 		union cpuid10_eax eax;
521 		union cpuid10_edx edx;
522 
523 		perf_get_x86_pmu_capability(&cap);
524 
525 		/*
526 		 * Only support guest architectural pmu on a host
527 		 * with architectural pmu.
528 		 */
529 		if (!cap.version)
530 			memset(&cap, 0, sizeof(cap));
531 
532 		eax.split.version_id = min(cap.version, 2);
533 		eax.split.num_counters = cap.num_counters_gp;
534 		eax.split.bit_width = cap.bit_width_gp;
535 		eax.split.mask_length = cap.events_mask_len;
536 
537 		edx.split.num_counters_fixed = cap.num_counters_fixed;
538 		edx.split.bit_width_fixed = cap.bit_width_fixed;
539 		edx.split.reserved = 0;
540 
541 		entry->eax = eax.full;
542 		entry->ebx = cap.events_mask;
543 		entry->ecx = 0;
544 		entry->edx = edx.full;
545 		break;
546 	}
547 	/* function 0xb has additional index. */
548 	case 0xb: {
549 		int i, level_type;
550 
551 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
552 		/* read more entries until level_type is zero */
553 		for (i = 1; ; ++i) {
554 			if (*nent >= maxnent)
555 				goto out;
556 
557 			level_type = entry[i - 1].ecx & 0xff00;
558 			if (!level_type)
559 				break;
560 			do_cpuid_1_ent(&entry[i], function, i);
561 			entry[i].flags |=
562 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
563 			++*nent;
564 		}
565 		break;
566 	}
567 	case 0xd: {
568 		int idx, i;
569 		u64 supported = kvm_supported_xcr0();
570 
571 		entry->eax &= supported;
572 		entry->ebx = xstate_required_size(supported, false);
573 		entry->ecx = entry->ebx;
574 		entry->edx &= supported >> 32;
575 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
576 		if (!supported)
577 			break;
578 
579 		for (idx = 1, i = 1; idx < 64; ++idx) {
580 			u64 mask = ((u64)1 << idx);
581 			if (*nent >= maxnent)
582 				goto out;
583 
584 			do_cpuid_1_ent(&entry[i], function, idx);
585 			if (idx == 1) {
586 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
587 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
588 				entry[i].ebx = 0;
589 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
590 					entry[i].ebx =
591 						xstate_required_size(supported,
592 								     true);
593 			} else {
594 				if (entry[i].eax == 0 || !(supported & mask))
595 					continue;
596 				if (WARN_ON_ONCE(entry[i].ecx & 1))
597 					continue;
598 			}
599 			entry[i].ecx = 0;
600 			entry[i].edx = 0;
601 			entry[i].flags |=
602 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
603 			++*nent;
604 			++i;
605 		}
606 		break;
607 	}
608 	/* Intel PT */
609 	case 0x14: {
610 		int t, times = entry->eax;
611 
612 		if (!f_intel_pt)
613 			break;
614 
615 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
616 		for (t = 1; t <= times; ++t) {
617 			if (*nent >= maxnent)
618 				goto out;
619 			do_cpuid_1_ent(&entry[t], function, t);
620 			entry[t].flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
621 			++*nent;
622 		}
623 		break;
624 	}
625 	case KVM_CPUID_SIGNATURE: {
626 		static const char signature[12] = "KVMKVMKVM\0\0";
627 		const u32 *sigptr = (const u32 *)signature;
628 		entry->eax = KVM_CPUID_FEATURES;
629 		entry->ebx = sigptr[0];
630 		entry->ecx = sigptr[1];
631 		entry->edx = sigptr[2];
632 		break;
633 	}
634 	case KVM_CPUID_FEATURES:
635 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
636 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
637 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
638 			     (1 << KVM_FEATURE_ASYNC_PF) |
639 			     (1 << KVM_FEATURE_PV_EOI) |
640 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
641 			     (1 << KVM_FEATURE_PV_UNHALT) |
642 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
643 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
644 			     (1 << KVM_FEATURE_PV_SEND_IPI);
645 
646 		if (sched_info_on())
647 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
648 
649 		entry->ebx = 0;
650 		entry->ecx = 0;
651 		entry->edx = 0;
652 		break;
653 	case 0x80000000:
654 		entry->eax = min(entry->eax, 0x8000001f);
655 		break;
656 	case 0x80000001:
657 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
658 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
659 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
660 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
661 		break;
662 	case 0x80000007: /* Advanced power management */
663 		/* invariant TSC is CPUID.80000007H:EDX[8] */
664 		entry->edx &= (1 << 8);
665 		/* mask against host */
666 		entry->edx &= boot_cpu_data.x86_power;
667 		entry->eax = entry->ebx = entry->ecx = 0;
668 		break;
669 	case 0x80000008: {
670 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
671 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
672 		unsigned phys_as = entry->eax & 0xff;
673 
674 		if (!g_phys_as)
675 			g_phys_as = phys_as;
676 		entry->eax = g_phys_as | (virt_as << 8);
677 		entry->edx = 0;
678 		/*
679 		 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
680 		 * hardware cpuid
681 		 */
682 		if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
683 			entry->ebx |= F(AMD_IBPB);
684 		if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
685 			entry->ebx |= F(AMD_IBRS);
686 		if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
687 			entry->ebx |= F(VIRT_SSBD);
688 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
689 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
690 		/*
691 		 * The preference is to use SPEC CTRL MSR instead of the
692 		 * VIRT_SPEC MSR.
693 		 */
694 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
695 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
696 			entry->ebx |= F(VIRT_SSBD);
697 		break;
698 	}
699 	case 0x80000019:
700 		entry->ecx = entry->edx = 0;
701 		break;
702 	case 0x8000001a:
703 		break;
704 	case 0x8000001d:
705 		break;
706 	/*Add support for Centaur's CPUID instruction*/
707 	case 0xC0000000:
708 		/*Just support up to 0xC0000004 now*/
709 		entry->eax = min(entry->eax, 0xC0000004);
710 		break;
711 	case 0xC0000001:
712 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
713 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
714 		break;
715 	case 3: /* Processor serial number */
716 	case 5: /* MONITOR/MWAIT */
717 	case 0xC0000002:
718 	case 0xC0000003:
719 	case 0xC0000004:
720 	default:
721 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
722 		break;
723 	}
724 
725 	kvm_x86_ops->set_supported_cpuid(function, entry);
726 
727 	r = 0;
728 
729 out:
730 	put_cpu();
731 
732 	return r;
733 }
734 
735 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
736 			u32 idx, int *nent, int maxnent, unsigned int type)
737 {
738 	if (type == KVM_GET_EMULATED_CPUID)
739 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
740 
741 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
742 }
743 
744 #undef F
745 
746 struct kvm_cpuid_param {
747 	u32 func;
748 	u32 idx;
749 	bool has_leaf_count;
750 	bool (*qualifier)(const struct kvm_cpuid_param *param);
751 };
752 
753 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
754 {
755 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
756 }
757 
758 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
759 				 __u32 num_entries, unsigned int ioctl_type)
760 {
761 	int i;
762 	__u32 pad[3];
763 
764 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
765 		return false;
766 
767 	/*
768 	 * We want to make sure that ->padding is being passed clean from
769 	 * userspace in case we want to use it for something in the future.
770 	 *
771 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
772 	 * have to give ourselves satisfied only with the emulated side. /me
773 	 * sheds a tear.
774 	 */
775 	for (i = 0; i < num_entries; i++) {
776 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
777 			return true;
778 
779 		if (pad[0] || pad[1] || pad[2])
780 			return true;
781 	}
782 	return false;
783 }
784 
785 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
786 			    struct kvm_cpuid_entry2 __user *entries,
787 			    unsigned int type)
788 {
789 	struct kvm_cpuid_entry2 *cpuid_entries;
790 	int limit, nent = 0, r = -E2BIG, i;
791 	u32 func;
792 	static const struct kvm_cpuid_param param[] = {
793 		{ .func = 0, .has_leaf_count = true },
794 		{ .func = 0x80000000, .has_leaf_count = true },
795 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
796 		{ .func = KVM_CPUID_SIGNATURE },
797 		{ .func = KVM_CPUID_FEATURES },
798 	};
799 
800 	if (cpuid->nent < 1)
801 		goto out;
802 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
803 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
804 
805 	if (sanity_check_entries(entries, cpuid->nent, type))
806 		return -EINVAL;
807 
808 	r = -ENOMEM;
809 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
810 					   cpuid->nent));
811 	if (!cpuid_entries)
812 		goto out;
813 
814 	r = 0;
815 	for (i = 0; i < ARRAY_SIZE(param); i++) {
816 		const struct kvm_cpuid_param *ent = &param[i];
817 
818 		if (ent->qualifier && !ent->qualifier(ent))
819 			continue;
820 
821 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
822 				&nent, cpuid->nent, type);
823 
824 		if (r)
825 			goto out_free;
826 
827 		if (!ent->has_leaf_count)
828 			continue;
829 
830 		limit = cpuid_entries[nent - 1].eax;
831 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
832 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
833 				     &nent, cpuid->nent, type);
834 
835 		if (r)
836 			goto out_free;
837 	}
838 
839 	r = -EFAULT;
840 	if (copy_to_user(entries, cpuid_entries,
841 			 nent * sizeof(struct kvm_cpuid_entry2)))
842 		goto out_free;
843 	cpuid->nent = nent;
844 	r = 0;
845 
846 out_free:
847 	vfree(cpuid_entries);
848 out:
849 	return r;
850 }
851 
852 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
853 {
854 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
855 	struct kvm_cpuid_entry2 *ej;
856 	int j = i;
857 	int nent = vcpu->arch.cpuid_nent;
858 
859 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
860 	/* when no next entry is found, the current entry[i] is reselected */
861 	do {
862 		j = (j + 1) % nent;
863 		ej = &vcpu->arch.cpuid_entries[j];
864 	} while (ej->function != e->function);
865 
866 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
867 
868 	return j;
869 }
870 
871 /* find an entry with matching function, matching index (if needed), and that
872  * should be read next (if it's stateful) */
873 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
874 	u32 function, u32 index)
875 {
876 	if (e->function != function)
877 		return 0;
878 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
879 		return 0;
880 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
881 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
882 		return 0;
883 	return 1;
884 }
885 
886 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
887 					      u32 function, u32 index)
888 {
889 	int i;
890 	struct kvm_cpuid_entry2 *best = NULL;
891 
892 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
893 		struct kvm_cpuid_entry2 *e;
894 
895 		e = &vcpu->arch.cpuid_entries[i];
896 		if (is_matching_cpuid_entry(e, function, index)) {
897 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
898 				move_to_next_stateful_cpuid_entry(vcpu, i);
899 			best = e;
900 			break;
901 		}
902 	}
903 	return best;
904 }
905 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
906 
907 /*
908  * If no match is found, check whether we exceed the vCPU's limit
909  * and return the content of the highest valid _standard_ leaf instead.
910  * This is to satisfy the CPUID specification.
911  */
912 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
913                                                   u32 function, u32 index)
914 {
915 	struct kvm_cpuid_entry2 *maxlevel;
916 
917 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
918 	if (!maxlevel || maxlevel->eax >= function)
919 		return NULL;
920 	if (function & 0x80000000) {
921 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
922 		if (!maxlevel)
923 			return NULL;
924 	}
925 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
926 }
927 
928 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
929 	       u32 *ecx, u32 *edx, bool check_limit)
930 {
931 	u32 function = *eax, index = *ecx;
932 	struct kvm_cpuid_entry2 *best;
933 	bool entry_found = true;
934 
935 	best = kvm_find_cpuid_entry(vcpu, function, index);
936 
937 	if (!best) {
938 		entry_found = false;
939 		if (!check_limit)
940 			goto out;
941 
942 		best = check_cpuid_limit(vcpu, function, index);
943 	}
944 
945 out:
946 	if (best) {
947 		*eax = best->eax;
948 		*ebx = best->ebx;
949 		*ecx = best->ecx;
950 		*edx = best->edx;
951 	} else
952 		*eax = *ebx = *ecx = *edx = 0;
953 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
954 	return entry_found;
955 }
956 EXPORT_SYMBOL_GPL(kvm_cpuid);
957 
958 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
959 {
960 	u32 eax, ebx, ecx, edx;
961 
962 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
963 		return 1;
964 
965 	eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
966 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
967 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
968 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
969 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
970 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
971 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
972 	return kvm_skip_emulated_instruction(vcpu);
973 }
974 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
975