xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision ccb01374)
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 
339 	/* cpuid 1.edx */
340 	const u32 kvm_cpuid_1_edx_x86_features =
341 		F(FPU) | F(VME) | F(DE) | F(PSE) |
342 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
343 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
344 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
345 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
346 		0 /* Reserved, DS, ACPI */ | F(MMX) |
347 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
348 		0 /* HTT, TM, Reserved, PBE */;
349 	/* cpuid 0x80000001.edx */
350 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
351 		F(FPU) | F(VME) | F(DE) | F(PSE) |
352 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
353 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
354 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
355 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
356 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
357 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
358 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
359 	/* cpuid 1.ecx */
360 	const u32 kvm_cpuid_1_ecx_x86_features =
361 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
362 		 * but *not* advertised to guests via CPUID ! */
363 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
364 		0 /* DS-CPL, VMX, SMX, EST */ |
365 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
366 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
367 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
368 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
369 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
370 		F(F16C) | F(RDRAND);
371 	/* cpuid 0x80000001.ecx */
372 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
373 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
374 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
375 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
376 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
377 		F(TOPOEXT) | F(PERFCTR_CORE);
378 
379 	/* cpuid 0x80000008.ebx */
380 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
381 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
382 		F(AMD_SSB_NO) | F(AMD_STIBP);
383 
384 	/* cpuid 0xC0000001.edx */
385 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
386 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
387 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
388 		F(PMM) | F(PMM_EN);
389 
390 	/* cpuid 7.0.ebx */
391 	const u32 kvm_cpuid_7_0_ebx_x86_features =
392 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
393 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
394 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
395 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
396 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
397 
398 	/* cpuid 0xD.1.eax */
399 	const u32 kvm_cpuid_D_1_eax_x86_features =
400 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
401 
402 	/* cpuid 7.0.ecx*/
403 	const u32 kvm_cpuid_7_0_ecx_x86_features =
404 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
405 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
406 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
407 		F(CLDEMOTE);
408 
409 	/* cpuid 7.0.edx*/
410 	const u32 kvm_cpuid_7_0_edx_x86_features =
411 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
412 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP);
413 
414 	/* all calls to cpuid_count() should be made on the same cpu */
415 	get_cpu();
416 
417 	r = -E2BIG;
418 
419 	if (*nent >= maxnent)
420 		goto out;
421 
422 	do_cpuid_1_ent(entry, function, index);
423 	++*nent;
424 
425 	switch (function) {
426 	case 0:
427 		entry->eax = min(entry->eax, (u32)(f_intel_pt ? 0x14 : 0xd));
428 		break;
429 	case 1:
430 		entry->edx &= kvm_cpuid_1_edx_x86_features;
431 		cpuid_mask(&entry->edx, CPUID_1_EDX);
432 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
433 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
434 		/* we support x2apic emulation even if host does not support
435 		 * it since we emulate x2apic in software */
436 		entry->ecx |= F(X2APIC);
437 		break;
438 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
439 	 * may return different values. This forces us to get_cpu() before
440 	 * issuing the first command, and also to emulate this annoying behavior
441 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
442 	case 2: {
443 		int t, times = entry->eax & 0xff;
444 
445 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
446 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
447 		for (t = 1; t < times; ++t) {
448 			if (*nent >= maxnent)
449 				goto out;
450 
451 			do_cpuid_1_ent(&entry[t], function, 0);
452 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
453 			++*nent;
454 		}
455 		break;
456 	}
457 	/* function 4 has additional index. */
458 	case 4: {
459 		int i, cache_type;
460 
461 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
462 		/* read more entries until cache_type is zero */
463 		for (i = 1; ; ++i) {
464 			if (*nent >= maxnent)
465 				goto out;
466 
467 			cache_type = entry[i - 1].eax & 0x1f;
468 			if (!cache_type)
469 				break;
470 			do_cpuid_1_ent(&entry[i], function, i);
471 			entry[i].flags |=
472 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
473 			++*nent;
474 		}
475 		break;
476 	}
477 	case 6: /* Thermal management */
478 		entry->eax = 0x4; /* allow ARAT */
479 		entry->ebx = 0;
480 		entry->ecx = 0;
481 		entry->edx = 0;
482 		break;
483 	case 7: {
484 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
485 		/* Mask ebx against host capability word 9 */
486 		if (index == 0) {
487 			entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
488 			cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
489 			// TSC_ADJUST is emulated
490 			entry->ebx |= F(TSC_ADJUST);
491 			entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
492 			cpuid_mask(&entry->ecx, CPUID_7_ECX);
493 			entry->ecx |= f_umip;
494 			/* PKU is not yet implemented for shadow paging. */
495 			if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
496 				entry->ecx &= ~F(PKU);
497 			entry->edx &= kvm_cpuid_7_0_edx_x86_features;
498 			cpuid_mask(&entry->edx, CPUID_7_EDX);
499 			/*
500 			 * We emulate ARCH_CAPABILITIES in software even
501 			 * if the host doesn't support it.
502 			 */
503 			entry->edx |= F(ARCH_CAPABILITIES);
504 		} else {
505 			entry->ebx = 0;
506 			entry->ecx = 0;
507 			entry->edx = 0;
508 		}
509 		entry->eax = 0;
510 		break;
511 	}
512 	case 9:
513 		break;
514 	case 0xa: { /* Architectural Performance Monitoring */
515 		struct x86_pmu_capability cap;
516 		union cpuid10_eax eax;
517 		union cpuid10_edx edx;
518 
519 		perf_get_x86_pmu_capability(&cap);
520 
521 		/*
522 		 * Only support guest architectural pmu on a host
523 		 * with architectural pmu.
524 		 */
525 		if (!cap.version)
526 			memset(&cap, 0, sizeof(cap));
527 
528 		eax.split.version_id = min(cap.version, 2);
529 		eax.split.num_counters = cap.num_counters_gp;
530 		eax.split.bit_width = cap.bit_width_gp;
531 		eax.split.mask_length = cap.events_mask_len;
532 
533 		edx.split.num_counters_fixed = cap.num_counters_fixed;
534 		edx.split.bit_width_fixed = cap.bit_width_fixed;
535 		edx.split.reserved = 0;
536 
537 		entry->eax = eax.full;
538 		entry->ebx = cap.events_mask;
539 		entry->ecx = 0;
540 		entry->edx = edx.full;
541 		break;
542 	}
543 	/* function 0xb has additional index. */
544 	case 0xb: {
545 		int i, level_type;
546 
547 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
548 		/* read more entries until level_type is zero */
549 		for (i = 1; ; ++i) {
550 			if (*nent >= maxnent)
551 				goto out;
552 
553 			level_type = entry[i - 1].ecx & 0xff00;
554 			if (!level_type)
555 				break;
556 			do_cpuid_1_ent(&entry[i], function, i);
557 			entry[i].flags |=
558 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
559 			++*nent;
560 		}
561 		break;
562 	}
563 	case 0xd: {
564 		int idx, i;
565 		u64 supported = kvm_supported_xcr0();
566 
567 		entry->eax &= supported;
568 		entry->ebx = xstate_required_size(supported, false);
569 		entry->ecx = entry->ebx;
570 		entry->edx &= supported >> 32;
571 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
572 		if (!supported)
573 			break;
574 
575 		for (idx = 1, i = 1; idx < 64; ++idx) {
576 			u64 mask = ((u64)1 << idx);
577 			if (*nent >= maxnent)
578 				goto out;
579 
580 			do_cpuid_1_ent(&entry[i], function, idx);
581 			if (idx == 1) {
582 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
583 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
584 				entry[i].ebx = 0;
585 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
586 					entry[i].ebx =
587 						xstate_required_size(supported,
588 								     true);
589 			} else {
590 				if (entry[i].eax == 0 || !(supported & mask))
591 					continue;
592 				if (WARN_ON_ONCE(entry[i].ecx & 1))
593 					continue;
594 			}
595 			entry[i].ecx = 0;
596 			entry[i].edx = 0;
597 			entry[i].flags |=
598 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
599 			++*nent;
600 			++i;
601 		}
602 		break;
603 	}
604 	/* Intel PT */
605 	case 0x14: {
606 		int t, times = entry->eax;
607 
608 		if (!f_intel_pt)
609 			break;
610 
611 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
612 		for (t = 1; t <= times; ++t) {
613 			if (*nent >= maxnent)
614 				goto out;
615 			do_cpuid_1_ent(&entry[t], function, t);
616 			entry[t].flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
617 			++*nent;
618 		}
619 		break;
620 	}
621 	case KVM_CPUID_SIGNATURE: {
622 		static const char signature[12] = "KVMKVMKVM\0\0";
623 		const u32 *sigptr = (const u32 *)signature;
624 		entry->eax = KVM_CPUID_FEATURES;
625 		entry->ebx = sigptr[0];
626 		entry->ecx = sigptr[1];
627 		entry->edx = sigptr[2];
628 		break;
629 	}
630 	case KVM_CPUID_FEATURES:
631 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
632 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
633 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
634 			     (1 << KVM_FEATURE_ASYNC_PF) |
635 			     (1 << KVM_FEATURE_PV_EOI) |
636 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
637 			     (1 << KVM_FEATURE_PV_UNHALT) |
638 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
639 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
640 			     (1 << KVM_FEATURE_PV_SEND_IPI);
641 
642 		if (sched_info_on())
643 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
644 
645 		entry->ebx = 0;
646 		entry->ecx = 0;
647 		entry->edx = 0;
648 		break;
649 	case 0x80000000:
650 		entry->eax = min(entry->eax, 0x8000001f);
651 		break;
652 	case 0x80000001:
653 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
654 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
655 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
656 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
657 		break;
658 	case 0x80000007: /* Advanced power management */
659 		/* invariant TSC is CPUID.80000007H:EDX[8] */
660 		entry->edx &= (1 << 8);
661 		/* mask against host */
662 		entry->edx &= boot_cpu_data.x86_power;
663 		entry->eax = entry->ebx = entry->ecx = 0;
664 		break;
665 	case 0x80000008: {
666 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
667 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
668 		unsigned phys_as = entry->eax & 0xff;
669 
670 		if (!g_phys_as)
671 			g_phys_as = phys_as;
672 		entry->eax = g_phys_as | (virt_as << 8);
673 		entry->edx = 0;
674 		/*
675 		 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
676 		 * hardware cpuid
677 		 */
678 		if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
679 			entry->ebx |= F(AMD_IBPB);
680 		if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
681 			entry->ebx |= F(AMD_IBRS);
682 		if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
683 			entry->ebx |= F(VIRT_SSBD);
684 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
685 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
686 		/*
687 		 * The preference is to use SPEC CTRL MSR instead of the
688 		 * VIRT_SPEC MSR.
689 		 */
690 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
691 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
692 			entry->ebx |= F(VIRT_SSBD);
693 		break;
694 	}
695 	case 0x80000019:
696 		entry->ecx = entry->edx = 0;
697 		break;
698 	case 0x8000001a:
699 		break;
700 	case 0x8000001d:
701 		break;
702 	/*Add support for Centaur's CPUID instruction*/
703 	case 0xC0000000:
704 		/*Just support up to 0xC0000004 now*/
705 		entry->eax = min(entry->eax, 0xC0000004);
706 		break;
707 	case 0xC0000001:
708 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
709 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
710 		break;
711 	case 3: /* Processor serial number */
712 	case 5: /* MONITOR/MWAIT */
713 	case 0xC0000002:
714 	case 0xC0000003:
715 	case 0xC0000004:
716 	default:
717 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
718 		break;
719 	}
720 
721 	kvm_x86_ops->set_supported_cpuid(function, entry);
722 
723 	r = 0;
724 
725 out:
726 	put_cpu();
727 
728 	return r;
729 }
730 
731 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
732 			u32 idx, int *nent, int maxnent, unsigned int type)
733 {
734 	if (type == KVM_GET_EMULATED_CPUID)
735 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
736 
737 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
738 }
739 
740 #undef F
741 
742 struct kvm_cpuid_param {
743 	u32 func;
744 	u32 idx;
745 	bool has_leaf_count;
746 	bool (*qualifier)(const struct kvm_cpuid_param *param);
747 };
748 
749 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
750 {
751 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
752 }
753 
754 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
755 				 __u32 num_entries, unsigned int ioctl_type)
756 {
757 	int i;
758 	__u32 pad[3];
759 
760 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
761 		return false;
762 
763 	/*
764 	 * We want to make sure that ->padding is being passed clean from
765 	 * userspace in case we want to use it for something in the future.
766 	 *
767 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
768 	 * have to give ourselves satisfied only with the emulated side. /me
769 	 * sheds a tear.
770 	 */
771 	for (i = 0; i < num_entries; i++) {
772 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
773 			return true;
774 
775 		if (pad[0] || pad[1] || pad[2])
776 			return true;
777 	}
778 	return false;
779 }
780 
781 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
782 			    struct kvm_cpuid_entry2 __user *entries,
783 			    unsigned int type)
784 {
785 	struct kvm_cpuid_entry2 *cpuid_entries;
786 	int limit, nent = 0, r = -E2BIG, i;
787 	u32 func;
788 	static const struct kvm_cpuid_param param[] = {
789 		{ .func = 0, .has_leaf_count = true },
790 		{ .func = 0x80000000, .has_leaf_count = true },
791 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
792 		{ .func = KVM_CPUID_SIGNATURE },
793 		{ .func = KVM_CPUID_FEATURES },
794 	};
795 
796 	if (cpuid->nent < 1)
797 		goto out;
798 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
799 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
800 
801 	if (sanity_check_entries(entries, cpuid->nent, type))
802 		return -EINVAL;
803 
804 	r = -ENOMEM;
805 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
806 					   cpuid->nent));
807 	if (!cpuid_entries)
808 		goto out;
809 
810 	r = 0;
811 	for (i = 0; i < ARRAY_SIZE(param); i++) {
812 		const struct kvm_cpuid_param *ent = &param[i];
813 
814 		if (ent->qualifier && !ent->qualifier(ent))
815 			continue;
816 
817 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
818 				&nent, cpuid->nent, type);
819 
820 		if (r)
821 			goto out_free;
822 
823 		if (!ent->has_leaf_count)
824 			continue;
825 
826 		limit = cpuid_entries[nent - 1].eax;
827 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
828 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
829 				     &nent, cpuid->nent, type);
830 
831 		if (r)
832 			goto out_free;
833 	}
834 
835 	r = -EFAULT;
836 	if (copy_to_user(entries, cpuid_entries,
837 			 nent * sizeof(struct kvm_cpuid_entry2)))
838 		goto out_free;
839 	cpuid->nent = nent;
840 	r = 0;
841 
842 out_free:
843 	vfree(cpuid_entries);
844 out:
845 	return r;
846 }
847 
848 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
849 {
850 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
851 	struct kvm_cpuid_entry2 *ej;
852 	int j = i;
853 	int nent = vcpu->arch.cpuid_nent;
854 
855 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
856 	/* when no next entry is found, the current entry[i] is reselected */
857 	do {
858 		j = (j + 1) % nent;
859 		ej = &vcpu->arch.cpuid_entries[j];
860 	} while (ej->function != e->function);
861 
862 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
863 
864 	return j;
865 }
866 
867 /* find an entry with matching function, matching index (if needed), and that
868  * should be read next (if it's stateful) */
869 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
870 	u32 function, u32 index)
871 {
872 	if (e->function != function)
873 		return 0;
874 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
875 		return 0;
876 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
877 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
878 		return 0;
879 	return 1;
880 }
881 
882 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
883 					      u32 function, u32 index)
884 {
885 	int i;
886 	struct kvm_cpuid_entry2 *best = NULL;
887 
888 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
889 		struct kvm_cpuid_entry2 *e;
890 
891 		e = &vcpu->arch.cpuid_entries[i];
892 		if (is_matching_cpuid_entry(e, function, index)) {
893 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
894 				move_to_next_stateful_cpuid_entry(vcpu, i);
895 			best = e;
896 			break;
897 		}
898 	}
899 	return best;
900 }
901 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
902 
903 /*
904  * If no match is found, check whether we exceed the vCPU's limit
905  * and return the content of the highest valid _standard_ leaf instead.
906  * This is to satisfy the CPUID specification.
907  */
908 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
909                                                   u32 function, u32 index)
910 {
911 	struct kvm_cpuid_entry2 *maxlevel;
912 
913 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
914 	if (!maxlevel || maxlevel->eax >= function)
915 		return NULL;
916 	if (function & 0x80000000) {
917 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
918 		if (!maxlevel)
919 			return NULL;
920 	}
921 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
922 }
923 
924 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
925 	       u32 *ecx, u32 *edx, bool check_limit)
926 {
927 	u32 function = *eax, index = *ecx;
928 	struct kvm_cpuid_entry2 *best;
929 	bool entry_found = true;
930 
931 	best = kvm_find_cpuid_entry(vcpu, function, index);
932 
933 	if (!best) {
934 		entry_found = false;
935 		if (!check_limit)
936 			goto out;
937 
938 		best = check_cpuid_limit(vcpu, function, index);
939 	}
940 
941 out:
942 	if (best) {
943 		*eax = best->eax;
944 		*ebx = best->ebx;
945 		*ecx = best->ecx;
946 		*edx = best->edx;
947 	} else
948 		*eax = *ebx = *ecx = *edx = 0;
949 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
950 	return entry_found;
951 }
952 EXPORT_SYMBOL_GPL(kvm_cpuid);
953 
954 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
955 {
956 	u32 eax, ebx, ecx, edx;
957 
958 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
959 		return 1;
960 
961 	eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
962 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
963 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
964 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
965 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
966 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
967 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
968 	return kvm_skip_emulated_instruction(vcpu);
969 }
970 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
971