xref: /linux/arch/x86/kvm/cpuid.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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 /* These are scattered features in cpufeatures.h. */
71 #define KVM_CPUID_BIT_AVX512_4VNNIW     2
72 #define KVM_CPUID_BIT_AVX512_4FMAPS     3
73 #define KF(x) bit(KVM_CPUID_BIT_##x)
74 
75 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
76 {
77 	struct kvm_cpuid_entry2 *best;
78 	struct kvm_lapic *apic = vcpu->arch.apic;
79 
80 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
81 	if (!best)
82 		return 0;
83 
84 	/* Update OSXSAVE bit */
85 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
86 		best->ecx &= ~F(OSXSAVE);
87 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
88 			best->ecx |= F(OSXSAVE);
89 	}
90 
91 	best->edx &= ~F(APIC);
92 	if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
93 		best->edx |= F(APIC);
94 
95 	if (apic) {
96 		if (best->ecx & F(TSC_DEADLINE_TIMER))
97 			apic->lapic_timer.timer_mode_mask = 3 << 17;
98 		else
99 			apic->lapic_timer.timer_mode_mask = 1 << 17;
100 	}
101 
102 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
103 	if (best) {
104 		/* Update OSPKE bit */
105 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
106 			best->ecx &= ~F(OSPKE);
107 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
108 				best->ecx |= F(OSPKE);
109 		}
110 	}
111 
112 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
113 	if (!best) {
114 		vcpu->arch.guest_supported_xcr0 = 0;
115 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
116 	} else {
117 		vcpu->arch.guest_supported_xcr0 =
118 			(best->eax | ((u64)best->edx << 32)) &
119 			kvm_supported_xcr0();
120 		vcpu->arch.guest_xstate_size = best->ebx =
121 			xstate_required_size(vcpu->arch.xcr0, false);
122 	}
123 
124 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
125 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
126 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
127 
128 	/*
129 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
130 	 * canonical address checks; exit if it is ever changed.
131 	 */
132 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
133 	if (best) {
134 		int vaddr_bits = (best->eax & 0xff00) >> 8;
135 
136 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
137 			return -EINVAL;
138 	}
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 = vmalloc(sizeof(struct kvm_cpuid_entry) *
204 					cpuid->nent);
205 		if (!cpuid_entries)
206 			goto out;
207 		r = -EFAULT;
208 		if (copy_from_user(cpuid_entries, entries,
209 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
210 			goto out;
211 	}
212 	for (i = 0; i < cpuid->nent; i++) {
213 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
214 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
215 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
216 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
217 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
218 		vcpu->arch.cpuid_entries[i].index = 0;
219 		vcpu->arch.cpuid_entries[i].flags = 0;
220 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
221 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
222 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
223 	}
224 	vcpu->arch.cpuid_nent = cpuid->nent;
225 	cpuid_fix_nx_cap(vcpu);
226 	kvm_apic_set_version(vcpu);
227 	kvm_x86_ops->cpuid_update(vcpu);
228 	r = kvm_update_cpuid(vcpu);
229 
230 out:
231 	vfree(cpuid_entries);
232 	return r;
233 }
234 
235 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
236 			      struct kvm_cpuid2 *cpuid,
237 			      struct kvm_cpuid_entry2 __user *entries)
238 {
239 	int r;
240 
241 	r = -E2BIG;
242 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
243 		goto out;
244 	r = -EFAULT;
245 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
246 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
247 		goto out;
248 	vcpu->arch.cpuid_nent = cpuid->nent;
249 	kvm_apic_set_version(vcpu);
250 	kvm_x86_ops->cpuid_update(vcpu);
251 	r = kvm_update_cpuid(vcpu);
252 out:
253 	return r;
254 }
255 
256 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
257 			      struct kvm_cpuid2 *cpuid,
258 			      struct kvm_cpuid_entry2 __user *entries)
259 {
260 	int r;
261 
262 	r = -E2BIG;
263 	if (cpuid->nent < vcpu->arch.cpuid_nent)
264 		goto out;
265 	r = -EFAULT;
266 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
267 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
268 		goto out;
269 	return 0;
270 
271 out:
272 	cpuid->nent = vcpu->arch.cpuid_nent;
273 	return r;
274 }
275 
276 static void cpuid_mask(u32 *word, int wordnum)
277 {
278 	*word &= boot_cpu_data.x86_capability[wordnum];
279 }
280 
281 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
282 			   u32 index)
283 {
284 	entry->function = function;
285 	entry->index = index;
286 	cpuid_count(entry->function, entry->index,
287 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
288 	entry->flags = 0;
289 }
290 
291 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
292 				   u32 func, u32 index, int *nent, int maxnent)
293 {
294 	switch (func) {
295 	case 0:
296 		entry->eax = 1;		/* only one leaf currently */
297 		++*nent;
298 		break;
299 	case 1:
300 		entry->ecx = F(MOVBE);
301 		++*nent;
302 		break;
303 	default:
304 		break;
305 	}
306 
307 	entry->function = func;
308 	entry->index = index;
309 
310 	return 0;
311 }
312 
313 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
314 				 u32 index, int *nent, int maxnent)
315 {
316 	int r;
317 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
318 #ifdef CONFIG_X86_64
319 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
320 				? F(GBPAGES) : 0;
321 	unsigned f_lm = F(LM);
322 #else
323 	unsigned f_gbpages = 0;
324 	unsigned f_lm = 0;
325 #endif
326 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
327 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
328 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
329 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
330 
331 	/* cpuid 1.edx */
332 	const u32 kvm_cpuid_1_edx_x86_features =
333 		F(FPU) | F(VME) | F(DE) | F(PSE) |
334 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
335 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
336 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
337 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
338 		0 /* Reserved, DS, ACPI */ | F(MMX) |
339 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
340 		0 /* HTT, TM, Reserved, PBE */;
341 	/* cpuid 0x80000001.edx */
342 	const u32 kvm_cpuid_8000_0001_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(SYSCALL) |
346 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
347 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
348 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
349 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
350 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
351 	/* cpuid 1.ecx */
352 	const u32 kvm_cpuid_1_ecx_x86_features =
353 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
354 		 * but *not* advertised to guests via CPUID ! */
355 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
356 		0 /* DS-CPL, VMX, SMX, EST */ |
357 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
358 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
359 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
360 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
361 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
362 		F(F16C) | F(RDRAND);
363 	/* cpuid 0x80000001.ecx */
364 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
365 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
366 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
367 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
368 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
369 
370 	/* cpuid 0xC0000001.edx */
371 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
372 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
373 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
374 		F(PMM) | F(PMM_EN);
375 
376 	/* cpuid 7.0.ebx */
377 	const u32 kvm_cpuid_7_0_ebx_x86_features =
378 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
379 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
380 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
381 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
382 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
383 
384 	/* cpuid 0xD.1.eax */
385 	const u32 kvm_cpuid_D_1_eax_x86_features =
386 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
387 
388 	/* cpuid 7.0.ecx*/
389 	const u32 kvm_cpuid_7_0_ecx_x86_features =
390 		F(AVX512VBMI) | F(LA57) | F(PKU) |
391 		0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ);
392 
393 	/* cpuid 7.0.edx*/
394 	const u32 kvm_cpuid_7_0_edx_x86_features =
395 		KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
396 
397 	/* all calls to cpuid_count() should be made on the same cpu */
398 	get_cpu();
399 
400 	r = -E2BIG;
401 
402 	if (*nent >= maxnent)
403 		goto out;
404 
405 	do_cpuid_1_ent(entry, function, index);
406 	++*nent;
407 
408 	switch (function) {
409 	case 0:
410 		entry->eax = min(entry->eax, (u32)0xd);
411 		break;
412 	case 1:
413 		entry->edx &= kvm_cpuid_1_edx_x86_features;
414 		cpuid_mask(&entry->edx, CPUID_1_EDX);
415 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
416 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
417 		/* we support x2apic emulation even if host does not support
418 		 * it since we emulate x2apic in software */
419 		entry->ecx |= F(X2APIC);
420 		break;
421 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
422 	 * may return different values. This forces us to get_cpu() before
423 	 * issuing the first command, and also to emulate this annoying behavior
424 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
425 	case 2: {
426 		int t, times = entry->eax & 0xff;
427 
428 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
429 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
430 		for (t = 1; t < times; ++t) {
431 			if (*nent >= maxnent)
432 				goto out;
433 
434 			do_cpuid_1_ent(&entry[t], function, 0);
435 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
436 			++*nent;
437 		}
438 		break;
439 	}
440 	/* function 4 has additional index. */
441 	case 4: {
442 		int i, cache_type;
443 
444 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
445 		/* read more entries until cache_type is zero */
446 		for (i = 1; ; ++i) {
447 			if (*nent >= maxnent)
448 				goto out;
449 
450 			cache_type = entry[i - 1].eax & 0x1f;
451 			if (!cache_type)
452 				break;
453 			do_cpuid_1_ent(&entry[i], function, i);
454 			entry[i].flags |=
455 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
456 			++*nent;
457 		}
458 		break;
459 	}
460 	case 6: /* Thermal management */
461 		entry->eax = 0x4; /* allow ARAT */
462 		entry->ebx = 0;
463 		entry->ecx = 0;
464 		entry->edx = 0;
465 		break;
466 	case 7: {
467 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
468 		/* Mask ebx against host capability word 9 */
469 		if (index == 0) {
470 			entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
471 			cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
472 			// TSC_ADJUST is emulated
473 			entry->ebx |= F(TSC_ADJUST);
474 			entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
475 			cpuid_mask(&entry->ecx, CPUID_7_ECX);
476 			/* PKU is not yet implemented for shadow paging. */
477 			if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
478 				entry->ecx &= ~F(PKU);
479 			entry->edx &= kvm_cpuid_7_0_edx_x86_features;
480 			entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
481 		} else {
482 			entry->ebx = 0;
483 			entry->ecx = 0;
484 			entry->edx = 0;
485 		}
486 		entry->eax = 0;
487 		break;
488 	}
489 	case 9:
490 		break;
491 	case 0xa: { /* Architectural Performance Monitoring */
492 		struct x86_pmu_capability cap;
493 		union cpuid10_eax eax;
494 		union cpuid10_edx edx;
495 
496 		perf_get_x86_pmu_capability(&cap);
497 
498 		/*
499 		 * Only support guest architectural pmu on a host
500 		 * with architectural pmu.
501 		 */
502 		if (!cap.version)
503 			memset(&cap, 0, sizeof(cap));
504 
505 		eax.split.version_id = min(cap.version, 2);
506 		eax.split.num_counters = cap.num_counters_gp;
507 		eax.split.bit_width = cap.bit_width_gp;
508 		eax.split.mask_length = cap.events_mask_len;
509 
510 		edx.split.num_counters_fixed = cap.num_counters_fixed;
511 		edx.split.bit_width_fixed = cap.bit_width_fixed;
512 		edx.split.reserved = 0;
513 
514 		entry->eax = eax.full;
515 		entry->ebx = cap.events_mask;
516 		entry->ecx = 0;
517 		entry->edx = edx.full;
518 		break;
519 	}
520 	/* function 0xb has additional index. */
521 	case 0xb: {
522 		int i, level_type;
523 
524 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
525 		/* read more entries until level_type is zero */
526 		for (i = 1; ; ++i) {
527 			if (*nent >= maxnent)
528 				goto out;
529 
530 			level_type = entry[i - 1].ecx & 0xff00;
531 			if (!level_type)
532 				break;
533 			do_cpuid_1_ent(&entry[i], function, i);
534 			entry[i].flags |=
535 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
536 			++*nent;
537 		}
538 		break;
539 	}
540 	case 0xd: {
541 		int idx, i;
542 		u64 supported = kvm_supported_xcr0();
543 
544 		entry->eax &= supported;
545 		entry->ebx = xstate_required_size(supported, false);
546 		entry->ecx = entry->ebx;
547 		entry->edx &= supported >> 32;
548 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
549 		if (!supported)
550 			break;
551 
552 		for (idx = 1, i = 1; idx < 64; ++idx) {
553 			u64 mask = ((u64)1 << idx);
554 			if (*nent >= maxnent)
555 				goto out;
556 
557 			do_cpuid_1_ent(&entry[i], function, idx);
558 			if (idx == 1) {
559 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
560 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
561 				entry[i].ebx = 0;
562 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
563 					entry[i].ebx =
564 						xstate_required_size(supported,
565 								     true);
566 			} else {
567 				if (entry[i].eax == 0 || !(supported & mask))
568 					continue;
569 				if (WARN_ON_ONCE(entry[i].ecx & 1))
570 					continue;
571 			}
572 			entry[i].ecx = 0;
573 			entry[i].edx = 0;
574 			entry[i].flags |=
575 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
576 			++*nent;
577 			++i;
578 		}
579 		break;
580 	}
581 	case KVM_CPUID_SIGNATURE: {
582 		static const char signature[12] = "KVMKVMKVM\0\0";
583 		const u32 *sigptr = (const u32 *)signature;
584 		entry->eax = KVM_CPUID_FEATURES;
585 		entry->ebx = sigptr[0];
586 		entry->ecx = sigptr[1];
587 		entry->edx = sigptr[2];
588 		break;
589 	}
590 	case KVM_CPUID_FEATURES:
591 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
592 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
593 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
594 			     (1 << KVM_FEATURE_ASYNC_PF) |
595 			     (1 << KVM_FEATURE_PV_EOI) |
596 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
597 			     (1 << KVM_FEATURE_PV_UNHALT);
598 
599 		if (sched_info_on())
600 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
601 
602 		entry->ebx = 0;
603 		entry->ecx = 0;
604 		entry->edx = 0;
605 		break;
606 	case 0x80000000:
607 		entry->eax = min(entry->eax, 0x8000001a);
608 		break;
609 	case 0x80000001:
610 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
611 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
612 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
613 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
614 		break;
615 	case 0x80000007: /* Advanced power management */
616 		/* invariant TSC is CPUID.80000007H:EDX[8] */
617 		entry->edx &= (1 << 8);
618 		/* mask against host */
619 		entry->edx &= boot_cpu_data.x86_power;
620 		entry->eax = entry->ebx = entry->ecx = 0;
621 		break;
622 	case 0x80000008: {
623 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
624 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
625 		unsigned phys_as = entry->eax & 0xff;
626 
627 		if (!g_phys_as)
628 			g_phys_as = phys_as;
629 		entry->eax = g_phys_as | (virt_as << 8);
630 		entry->ebx = entry->edx = 0;
631 		break;
632 	}
633 	case 0x80000019:
634 		entry->ecx = entry->edx = 0;
635 		break;
636 	case 0x8000001a:
637 		break;
638 	case 0x8000001d:
639 		break;
640 	/*Add support for Centaur's CPUID instruction*/
641 	case 0xC0000000:
642 		/*Just support up to 0xC0000004 now*/
643 		entry->eax = min(entry->eax, 0xC0000004);
644 		break;
645 	case 0xC0000001:
646 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
647 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
648 		break;
649 	case 3: /* Processor serial number */
650 	case 5: /* MONITOR/MWAIT */
651 	case 0xC0000002:
652 	case 0xC0000003:
653 	case 0xC0000004:
654 	default:
655 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
656 		break;
657 	}
658 
659 	kvm_x86_ops->set_supported_cpuid(function, entry);
660 
661 	r = 0;
662 
663 out:
664 	put_cpu();
665 
666 	return r;
667 }
668 
669 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
670 			u32 idx, int *nent, int maxnent, unsigned int type)
671 {
672 	if (type == KVM_GET_EMULATED_CPUID)
673 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
674 
675 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
676 }
677 
678 #undef F
679 
680 struct kvm_cpuid_param {
681 	u32 func;
682 	u32 idx;
683 	bool has_leaf_count;
684 	bool (*qualifier)(const struct kvm_cpuid_param *param);
685 };
686 
687 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
688 {
689 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
690 }
691 
692 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
693 				 __u32 num_entries, unsigned int ioctl_type)
694 {
695 	int i;
696 	__u32 pad[3];
697 
698 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
699 		return false;
700 
701 	/*
702 	 * We want to make sure that ->padding is being passed clean from
703 	 * userspace in case we want to use it for something in the future.
704 	 *
705 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
706 	 * have to give ourselves satisfied only with the emulated side. /me
707 	 * sheds a tear.
708 	 */
709 	for (i = 0; i < num_entries; i++) {
710 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
711 			return true;
712 
713 		if (pad[0] || pad[1] || pad[2])
714 			return true;
715 	}
716 	return false;
717 }
718 
719 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
720 			    struct kvm_cpuid_entry2 __user *entries,
721 			    unsigned int type)
722 {
723 	struct kvm_cpuid_entry2 *cpuid_entries;
724 	int limit, nent = 0, r = -E2BIG, i;
725 	u32 func;
726 	static const struct kvm_cpuid_param param[] = {
727 		{ .func = 0, .has_leaf_count = true },
728 		{ .func = 0x80000000, .has_leaf_count = true },
729 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
730 		{ .func = KVM_CPUID_SIGNATURE },
731 		{ .func = KVM_CPUID_FEATURES },
732 	};
733 
734 	if (cpuid->nent < 1)
735 		goto out;
736 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
737 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
738 
739 	if (sanity_check_entries(entries, cpuid->nent, type))
740 		return -EINVAL;
741 
742 	r = -ENOMEM;
743 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
744 	if (!cpuid_entries)
745 		goto out;
746 
747 	r = 0;
748 	for (i = 0; i < ARRAY_SIZE(param); i++) {
749 		const struct kvm_cpuid_param *ent = &param[i];
750 
751 		if (ent->qualifier && !ent->qualifier(ent))
752 			continue;
753 
754 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
755 				&nent, cpuid->nent, type);
756 
757 		if (r)
758 			goto out_free;
759 
760 		if (!ent->has_leaf_count)
761 			continue;
762 
763 		limit = cpuid_entries[nent - 1].eax;
764 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
765 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
766 				     &nent, cpuid->nent, type);
767 
768 		if (r)
769 			goto out_free;
770 	}
771 
772 	r = -EFAULT;
773 	if (copy_to_user(entries, cpuid_entries,
774 			 nent * sizeof(struct kvm_cpuid_entry2)))
775 		goto out_free;
776 	cpuid->nent = nent;
777 	r = 0;
778 
779 out_free:
780 	vfree(cpuid_entries);
781 out:
782 	return r;
783 }
784 
785 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
786 {
787 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
788 	struct kvm_cpuid_entry2 *ej;
789 	int j = i;
790 	int nent = vcpu->arch.cpuid_nent;
791 
792 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
793 	/* when no next entry is found, the current entry[i] is reselected */
794 	do {
795 		j = (j + 1) % nent;
796 		ej = &vcpu->arch.cpuid_entries[j];
797 	} while (ej->function != e->function);
798 
799 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
800 
801 	return j;
802 }
803 
804 /* find an entry with matching function, matching index (if needed), and that
805  * should be read next (if it's stateful) */
806 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
807 	u32 function, u32 index)
808 {
809 	if (e->function != function)
810 		return 0;
811 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
812 		return 0;
813 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
814 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
815 		return 0;
816 	return 1;
817 }
818 
819 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
820 					      u32 function, u32 index)
821 {
822 	int i;
823 	struct kvm_cpuid_entry2 *best = NULL;
824 
825 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
826 		struct kvm_cpuid_entry2 *e;
827 
828 		e = &vcpu->arch.cpuid_entries[i];
829 		if (is_matching_cpuid_entry(e, function, index)) {
830 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
831 				move_to_next_stateful_cpuid_entry(vcpu, i);
832 			best = e;
833 			break;
834 		}
835 	}
836 	return best;
837 }
838 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
839 
840 /*
841  * If no match is found, check whether we exceed the vCPU's limit
842  * and return the content of the highest valid _standard_ leaf instead.
843  * This is to satisfy the CPUID specification.
844  */
845 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
846                                                   u32 function, u32 index)
847 {
848 	struct kvm_cpuid_entry2 *maxlevel;
849 
850 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
851 	if (!maxlevel || maxlevel->eax >= function)
852 		return NULL;
853 	if (function & 0x80000000) {
854 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
855 		if (!maxlevel)
856 			return NULL;
857 	}
858 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
859 }
860 
861 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
862 	       u32 *ecx, u32 *edx, bool check_limit)
863 {
864 	u32 function = *eax, index = *ecx;
865 	struct kvm_cpuid_entry2 *best;
866 	bool entry_found = true;
867 
868 	best = kvm_find_cpuid_entry(vcpu, function, index);
869 
870 	if (!best) {
871 		entry_found = false;
872 		if (!check_limit)
873 			goto out;
874 
875 		best = check_cpuid_limit(vcpu, function, index);
876 	}
877 
878 out:
879 	if (best) {
880 		*eax = best->eax;
881 		*ebx = best->ebx;
882 		*ecx = best->ecx;
883 		*edx = best->edx;
884 	} else
885 		*eax = *ebx = *ecx = *edx = 0;
886 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
887 	return entry_found;
888 }
889 EXPORT_SYMBOL_GPL(kvm_cpuid);
890 
891 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
892 {
893 	u32 eax, ebx, ecx, edx;
894 
895 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
896 		return 1;
897 
898 	eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
899 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
900 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
901 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
902 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
903 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
904 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
905 	return kvm_skip_emulated_instruction(vcpu);
906 }
907 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
908