xref: /linux/arch/x86/kvm/cpuid.c (revision 3b812ecce736432e6b55e77028ea387eb1517d24)
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/module.h>
17 #include <linux/vmalloc.h>
18 #include <linux/uaccess.h>
19 #include <asm/fpu/internal.h> /* For use_eager_fpu.  Ugh! */
20 #include <asm/user.h>
21 #include <asm/fpu/xstate.h>
22 #include "cpuid.h"
23 #include "lapic.h"
24 #include "mmu.h"
25 #include "trace.h"
26 #include "pmu.h"
27 
28 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
29 {
30 	int feature_bit = 0;
31 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
32 
33 	xstate_bv &= XFEATURE_MASK_EXTEND;
34 	while (xstate_bv) {
35 		if (xstate_bv & 0x1) {
36 		        u32 eax, ebx, ecx, edx, offset;
37 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
38 			offset = compacted ? ret : ebx;
39 			ret = max(ret, offset + eax);
40 		}
41 
42 		xstate_bv >>= 1;
43 		feature_bit++;
44 	}
45 
46 	return ret;
47 }
48 
49 bool kvm_mpx_supported(void)
50 {
51 	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
52 		 && kvm_x86_ops->mpx_supported());
53 }
54 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
55 
56 u64 kvm_supported_xcr0(void)
57 {
58 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
59 
60 	if (!kvm_mpx_supported())
61 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
62 
63 	return xcr0;
64 }
65 
66 #define F(x) bit(X86_FEATURE_##x)
67 
68 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
69 {
70 	struct kvm_cpuid_entry2 *best;
71 	struct kvm_lapic *apic = vcpu->arch.apic;
72 
73 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
74 	if (!best)
75 		return 0;
76 
77 	/* Update OSXSAVE bit */
78 	if (cpu_has_xsave && best->function == 0x1) {
79 		best->ecx &= ~F(OSXSAVE);
80 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
81 			best->ecx |= F(OSXSAVE);
82 	}
83 
84 	if (apic) {
85 		if (best->ecx & F(TSC_DEADLINE_TIMER))
86 			apic->lapic_timer.timer_mode_mask = 3 << 17;
87 		else
88 			apic->lapic_timer.timer_mode_mask = 1 << 17;
89 	}
90 
91 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
92 	if (!best) {
93 		vcpu->arch.guest_supported_xcr0 = 0;
94 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
95 	} else {
96 		vcpu->arch.guest_supported_xcr0 =
97 			(best->eax | ((u64)best->edx << 32)) &
98 			kvm_supported_xcr0();
99 		vcpu->arch.guest_xstate_size = best->ebx =
100 			xstate_required_size(vcpu->arch.xcr0, false);
101 	}
102 
103 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
104 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
105 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
106 
107 	if (use_eager_fpu())
108 		kvm_x86_ops->fpu_activate(vcpu);
109 
110 	/*
111 	 * The existing code assumes virtual address is 48-bit in the canonical
112 	 * address checks; exit if it is ever changed.
113 	 */
114 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
115 	if (best && ((best->eax & 0xff00) >> 8) != 48 &&
116 		((best->eax & 0xff00) >> 8) != 0)
117 		return -EINVAL;
118 
119 	/* Update physical-address width */
120 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
121 
122 	kvm_pmu_refresh(vcpu);
123 	return 0;
124 }
125 
126 static int is_efer_nx(void)
127 {
128 	unsigned long long efer = 0;
129 
130 	rdmsrl_safe(MSR_EFER, &efer);
131 	return efer & EFER_NX;
132 }
133 
134 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
135 {
136 	int i;
137 	struct kvm_cpuid_entry2 *e, *entry;
138 
139 	entry = NULL;
140 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
141 		e = &vcpu->arch.cpuid_entries[i];
142 		if (e->function == 0x80000001) {
143 			entry = e;
144 			break;
145 		}
146 	}
147 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
148 		entry->edx &= ~F(NX);
149 		printk(KERN_INFO "kvm: guest NX capability removed\n");
150 	}
151 }
152 
153 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
154 {
155 	struct kvm_cpuid_entry2 *best;
156 
157 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
158 	if (!best || best->eax < 0x80000008)
159 		goto not_found;
160 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
161 	if (best)
162 		return best->eax & 0xff;
163 not_found:
164 	return 36;
165 }
166 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
167 
168 /* when an old userspace process fills a new kernel module */
169 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
170 			     struct kvm_cpuid *cpuid,
171 			     struct kvm_cpuid_entry __user *entries)
172 {
173 	int r, i;
174 	struct kvm_cpuid_entry *cpuid_entries;
175 
176 	r = -E2BIG;
177 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
178 		goto out;
179 	r = -ENOMEM;
180 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
181 	if (!cpuid_entries)
182 		goto out;
183 	r = -EFAULT;
184 	if (copy_from_user(cpuid_entries, entries,
185 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
186 		goto out_free;
187 	for (i = 0; i < cpuid->nent; i++) {
188 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
189 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
190 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
191 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
192 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
193 		vcpu->arch.cpuid_entries[i].index = 0;
194 		vcpu->arch.cpuid_entries[i].flags = 0;
195 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
196 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
197 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
198 	}
199 	vcpu->arch.cpuid_nent = cpuid->nent;
200 	cpuid_fix_nx_cap(vcpu);
201 	kvm_apic_set_version(vcpu);
202 	kvm_x86_ops->cpuid_update(vcpu);
203 	r = kvm_update_cpuid(vcpu);
204 
205 out_free:
206 	vfree(cpuid_entries);
207 out:
208 	return r;
209 }
210 
211 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
212 			      struct kvm_cpuid2 *cpuid,
213 			      struct kvm_cpuid_entry2 __user *entries)
214 {
215 	int r;
216 
217 	r = -E2BIG;
218 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
219 		goto out;
220 	r = -EFAULT;
221 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
222 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
223 		goto out;
224 	vcpu->arch.cpuid_nent = cpuid->nent;
225 	kvm_apic_set_version(vcpu);
226 	kvm_x86_ops->cpuid_update(vcpu);
227 	r = kvm_update_cpuid(vcpu);
228 out:
229 	return r;
230 }
231 
232 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
233 			      struct kvm_cpuid2 *cpuid,
234 			      struct kvm_cpuid_entry2 __user *entries)
235 {
236 	int r;
237 
238 	r = -E2BIG;
239 	if (cpuid->nent < vcpu->arch.cpuid_nent)
240 		goto out;
241 	r = -EFAULT;
242 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
243 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
244 		goto out;
245 	return 0;
246 
247 out:
248 	cpuid->nent = vcpu->arch.cpuid_nent;
249 	return r;
250 }
251 
252 static void cpuid_mask(u32 *word, int wordnum)
253 {
254 	*word &= boot_cpu_data.x86_capability[wordnum];
255 }
256 
257 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
258 			   u32 index)
259 {
260 	entry->function = function;
261 	entry->index = index;
262 	cpuid_count(entry->function, entry->index,
263 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
264 	entry->flags = 0;
265 }
266 
267 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
268 				   u32 func, u32 index, int *nent, int maxnent)
269 {
270 	switch (func) {
271 	case 0:
272 		entry->eax = 1;		/* only one leaf currently */
273 		++*nent;
274 		break;
275 	case 1:
276 		entry->ecx = F(MOVBE);
277 		++*nent;
278 		break;
279 	default:
280 		break;
281 	}
282 
283 	entry->function = func;
284 	entry->index = index;
285 
286 	return 0;
287 }
288 
289 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
290 				 u32 index, int *nent, int maxnent)
291 {
292 	int r;
293 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
294 #ifdef CONFIG_X86_64
295 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
296 				? F(GBPAGES) : 0;
297 	unsigned f_lm = F(LM);
298 #else
299 	unsigned f_gbpages = 0;
300 	unsigned f_lm = 0;
301 #endif
302 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
303 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
304 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
305 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
306 
307 	/* cpuid 1.edx */
308 	const u32 kvm_supported_word0_x86_features =
309 		F(FPU) | F(VME) | F(DE) | F(PSE) |
310 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
311 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
312 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
313 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
314 		0 /* Reserved, DS, ACPI */ | F(MMX) |
315 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
316 		0 /* HTT, TM, Reserved, PBE */;
317 	/* cpuid 0x80000001.edx */
318 	const u32 kvm_supported_word1_x86_features =
319 		F(FPU) | F(VME) | F(DE) | F(PSE) |
320 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
321 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
322 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
323 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
324 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
325 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
326 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
327 	/* cpuid 1.ecx */
328 	const u32 kvm_supported_word4_x86_features =
329 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
330 		 * but *not* advertised to guests via CPUID ! */
331 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
332 		0 /* DS-CPL, VMX, SMX, EST */ |
333 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
334 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
335 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
336 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
337 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
338 		F(F16C) | F(RDRAND);
339 	/* cpuid 0x80000001.ecx */
340 	const u32 kvm_supported_word6_x86_features =
341 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
342 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
343 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
344 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
345 
346 	/* cpuid 0xC0000001.edx */
347 	const u32 kvm_supported_word5_x86_features =
348 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
349 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
350 		F(PMM) | F(PMM_EN);
351 
352 	/* cpuid 7.0.ebx */
353 	const u32 kvm_supported_word9_x86_features =
354 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
355 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
356 		F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
357 		F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
358 
359 	/* cpuid 0xD.1.eax */
360 	const u32 kvm_supported_word10_x86_features =
361 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
362 
363 	/* all calls to cpuid_count() should be made on the same cpu */
364 	get_cpu();
365 
366 	r = -E2BIG;
367 
368 	if (*nent >= maxnent)
369 		goto out;
370 
371 	do_cpuid_1_ent(entry, function, index);
372 	++*nent;
373 
374 	switch (function) {
375 	case 0:
376 		entry->eax = min(entry->eax, (u32)0xd);
377 		break;
378 	case 1:
379 		entry->edx &= kvm_supported_word0_x86_features;
380 		cpuid_mask(&entry->edx, 0);
381 		entry->ecx &= kvm_supported_word4_x86_features;
382 		cpuid_mask(&entry->ecx, 4);
383 		/* we support x2apic emulation even if host does not support
384 		 * it since we emulate x2apic in software */
385 		entry->ecx |= F(X2APIC);
386 		break;
387 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
388 	 * may return different values. This forces us to get_cpu() before
389 	 * issuing the first command, and also to emulate this annoying behavior
390 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
391 	case 2: {
392 		int t, times = entry->eax & 0xff;
393 
394 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
395 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
396 		for (t = 1; t < times; ++t) {
397 			if (*nent >= maxnent)
398 				goto out;
399 
400 			do_cpuid_1_ent(&entry[t], function, 0);
401 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
402 			++*nent;
403 		}
404 		break;
405 	}
406 	/* function 4 has additional index. */
407 	case 4: {
408 		int i, cache_type;
409 
410 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
411 		/* read more entries until cache_type is zero */
412 		for (i = 1; ; ++i) {
413 			if (*nent >= maxnent)
414 				goto out;
415 
416 			cache_type = entry[i - 1].eax & 0x1f;
417 			if (!cache_type)
418 				break;
419 			do_cpuid_1_ent(&entry[i], function, i);
420 			entry[i].flags |=
421 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
422 			++*nent;
423 		}
424 		break;
425 	}
426 	case 6: /* Thermal management */
427 		entry->eax = 0x4; /* allow ARAT */
428 		entry->ebx = 0;
429 		entry->ecx = 0;
430 		entry->edx = 0;
431 		break;
432 	case 7: {
433 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
434 		/* Mask ebx against host capability word 9 */
435 		if (index == 0) {
436 			entry->ebx &= kvm_supported_word9_x86_features;
437 			cpuid_mask(&entry->ebx, 9);
438 			// TSC_ADJUST is emulated
439 			entry->ebx |= F(TSC_ADJUST);
440 		} else
441 			entry->ebx = 0;
442 		entry->eax = 0;
443 		entry->ecx = 0;
444 		entry->edx = 0;
445 		break;
446 	}
447 	case 9:
448 		break;
449 	case 0xa: { /* Architectural Performance Monitoring */
450 		struct x86_pmu_capability cap;
451 		union cpuid10_eax eax;
452 		union cpuid10_edx edx;
453 
454 		perf_get_x86_pmu_capability(&cap);
455 
456 		/*
457 		 * Only support guest architectural pmu on a host
458 		 * with architectural pmu.
459 		 */
460 		if (!cap.version)
461 			memset(&cap, 0, sizeof(cap));
462 
463 		eax.split.version_id = min(cap.version, 2);
464 		eax.split.num_counters = cap.num_counters_gp;
465 		eax.split.bit_width = cap.bit_width_gp;
466 		eax.split.mask_length = cap.events_mask_len;
467 
468 		edx.split.num_counters_fixed = cap.num_counters_fixed;
469 		edx.split.bit_width_fixed = cap.bit_width_fixed;
470 		edx.split.reserved = 0;
471 
472 		entry->eax = eax.full;
473 		entry->ebx = cap.events_mask;
474 		entry->ecx = 0;
475 		entry->edx = edx.full;
476 		break;
477 	}
478 	/* function 0xb has additional index. */
479 	case 0xb: {
480 		int i, level_type;
481 
482 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
483 		/* read more entries until level_type is zero */
484 		for (i = 1; ; ++i) {
485 			if (*nent >= maxnent)
486 				goto out;
487 
488 			level_type = entry[i - 1].ecx & 0xff00;
489 			if (!level_type)
490 				break;
491 			do_cpuid_1_ent(&entry[i], function, i);
492 			entry[i].flags |=
493 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
494 			++*nent;
495 		}
496 		break;
497 	}
498 	case 0xd: {
499 		int idx, i;
500 		u64 supported = kvm_supported_xcr0();
501 
502 		entry->eax &= supported;
503 		entry->ebx = xstate_required_size(supported, false);
504 		entry->ecx = entry->ebx;
505 		entry->edx &= supported >> 32;
506 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
507 		if (!supported)
508 			break;
509 
510 		for (idx = 1, i = 1; idx < 64; ++idx) {
511 			u64 mask = ((u64)1 << idx);
512 			if (*nent >= maxnent)
513 				goto out;
514 
515 			do_cpuid_1_ent(&entry[i], function, idx);
516 			if (idx == 1) {
517 				entry[i].eax &= kvm_supported_word10_x86_features;
518 				entry[i].ebx = 0;
519 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
520 					entry[i].ebx =
521 						xstate_required_size(supported,
522 								     true);
523 			} else {
524 				if (entry[i].eax == 0 || !(supported & mask))
525 					continue;
526 				if (WARN_ON_ONCE(entry[i].ecx & 1))
527 					continue;
528 			}
529 			entry[i].ecx = 0;
530 			entry[i].edx = 0;
531 			entry[i].flags |=
532 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
533 			++*nent;
534 			++i;
535 		}
536 		break;
537 	}
538 	case KVM_CPUID_SIGNATURE: {
539 		static const char signature[12] = "KVMKVMKVM\0\0";
540 		const u32 *sigptr = (const u32 *)signature;
541 		entry->eax = KVM_CPUID_FEATURES;
542 		entry->ebx = sigptr[0];
543 		entry->ecx = sigptr[1];
544 		entry->edx = sigptr[2];
545 		break;
546 	}
547 	case KVM_CPUID_FEATURES:
548 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
549 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
550 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
551 			     (1 << KVM_FEATURE_ASYNC_PF) |
552 			     (1 << KVM_FEATURE_PV_EOI) |
553 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
554 			     (1 << KVM_FEATURE_PV_UNHALT);
555 
556 		if (sched_info_on())
557 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
558 
559 		entry->ebx = 0;
560 		entry->ecx = 0;
561 		entry->edx = 0;
562 		break;
563 	case 0x80000000:
564 		entry->eax = min(entry->eax, 0x8000001a);
565 		break;
566 	case 0x80000001:
567 		entry->edx &= kvm_supported_word1_x86_features;
568 		cpuid_mask(&entry->edx, 1);
569 		entry->ecx &= kvm_supported_word6_x86_features;
570 		cpuid_mask(&entry->ecx, 6);
571 		break;
572 	case 0x80000007: /* Advanced power management */
573 		/* invariant TSC is CPUID.80000007H:EDX[8] */
574 		entry->edx &= (1 << 8);
575 		/* mask against host */
576 		entry->edx &= boot_cpu_data.x86_power;
577 		entry->eax = entry->ebx = entry->ecx = 0;
578 		break;
579 	case 0x80000008: {
580 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
581 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
582 		unsigned phys_as = entry->eax & 0xff;
583 
584 		if (!g_phys_as)
585 			g_phys_as = phys_as;
586 		entry->eax = g_phys_as | (virt_as << 8);
587 		entry->ebx = entry->edx = 0;
588 		break;
589 	}
590 	case 0x80000019:
591 		entry->ecx = entry->edx = 0;
592 		break;
593 	case 0x8000001a:
594 		break;
595 	case 0x8000001d:
596 		break;
597 	/*Add support for Centaur's CPUID instruction*/
598 	case 0xC0000000:
599 		/*Just support up to 0xC0000004 now*/
600 		entry->eax = min(entry->eax, 0xC0000004);
601 		break;
602 	case 0xC0000001:
603 		entry->edx &= kvm_supported_word5_x86_features;
604 		cpuid_mask(&entry->edx, 5);
605 		break;
606 	case 3: /* Processor serial number */
607 	case 5: /* MONITOR/MWAIT */
608 	case 0xC0000002:
609 	case 0xC0000003:
610 	case 0xC0000004:
611 	default:
612 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
613 		break;
614 	}
615 
616 	kvm_x86_ops->set_supported_cpuid(function, entry);
617 
618 	r = 0;
619 
620 out:
621 	put_cpu();
622 
623 	return r;
624 }
625 
626 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
627 			u32 idx, int *nent, int maxnent, unsigned int type)
628 {
629 	if (type == KVM_GET_EMULATED_CPUID)
630 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
631 
632 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
633 }
634 
635 #undef F
636 
637 struct kvm_cpuid_param {
638 	u32 func;
639 	u32 idx;
640 	bool has_leaf_count;
641 	bool (*qualifier)(const struct kvm_cpuid_param *param);
642 };
643 
644 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
645 {
646 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
647 }
648 
649 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
650 				 __u32 num_entries, unsigned int ioctl_type)
651 {
652 	int i;
653 	__u32 pad[3];
654 
655 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
656 		return false;
657 
658 	/*
659 	 * We want to make sure that ->padding is being passed clean from
660 	 * userspace in case we want to use it for something in the future.
661 	 *
662 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
663 	 * have to give ourselves satisfied only with the emulated side. /me
664 	 * sheds a tear.
665 	 */
666 	for (i = 0; i < num_entries; i++) {
667 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
668 			return true;
669 
670 		if (pad[0] || pad[1] || pad[2])
671 			return true;
672 	}
673 	return false;
674 }
675 
676 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
677 			    struct kvm_cpuid_entry2 __user *entries,
678 			    unsigned int type)
679 {
680 	struct kvm_cpuid_entry2 *cpuid_entries;
681 	int limit, nent = 0, r = -E2BIG, i;
682 	u32 func;
683 	static const struct kvm_cpuid_param param[] = {
684 		{ .func = 0, .has_leaf_count = true },
685 		{ .func = 0x80000000, .has_leaf_count = true },
686 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
687 		{ .func = KVM_CPUID_SIGNATURE },
688 		{ .func = KVM_CPUID_FEATURES },
689 	};
690 
691 	if (cpuid->nent < 1)
692 		goto out;
693 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
694 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
695 
696 	if (sanity_check_entries(entries, cpuid->nent, type))
697 		return -EINVAL;
698 
699 	r = -ENOMEM;
700 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
701 	if (!cpuid_entries)
702 		goto out;
703 
704 	r = 0;
705 	for (i = 0; i < ARRAY_SIZE(param); i++) {
706 		const struct kvm_cpuid_param *ent = &param[i];
707 
708 		if (ent->qualifier && !ent->qualifier(ent))
709 			continue;
710 
711 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
712 				&nent, cpuid->nent, type);
713 
714 		if (r)
715 			goto out_free;
716 
717 		if (!ent->has_leaf_count)
718 			continue;
719 
720 		limit = cpuid_entries[nent - 1].eax;
721 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
722 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
723 				     &nent, cpuid->nent, type);
724 
725 		if (r)
726 			goto out_free;
727 	}
728 
729 	r = -EFAULT;
730 	if (copy_to_user(entries, cpuid_entries,
731 			 nent * sizeof(struct kvm_cpuid_entry2)))
732 		goto out_free;
733 	cpuid->nent = nent;
734 	r = 0;
735 
736 out_free:
737 	vfree(cpuid_entries);
738 out:
739 	return r;
740 }
741 
742 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
743 {
744 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
745 	int j, nent = vcpu->arch.cpuid_nent;
746 
747 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
748 	/* when no next entry is found, the current entry[i] is reselected */
749 	for (j = i + 1; ; j = (j + 1) % nent) {
750 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
751 		if (ej->function == e->function) {
752 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
753 			return j;
754 		}
755 	}
756 	return 0; /* silence gcc, even though control never reaches here */
757 }
758 
759 /* find an entry with matching function, matching index (if needed), and that
760  * should be read next (if it's stateful) */
761 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
762 	u32 function, u32 index)
763 {
764 	if (e->function != function)
765 		return 0;
766 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
767 		return 0;
768 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
769 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
770 		return 0;
771 	return 1;
772 }
773 
774 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
775 					      u32 function, u32 index)
776 {
777 	int i;
778 	struct kvm_cpuid_entry2 *best = NULL;
779 
780 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
781 		struct kvm_cpuid_entry2 *e;
782 
783 		e = &vcpu->arch.cpuid_entries[i];
784 		if (is_matching_cpuid_entry(e, function, index)) {
785 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
786 				move_to_next_stateful_cpuid_entry(vcpu, i);
787 			best = e;
788 			break;
789 		}
790 	}
791 	return best;
792 }
793 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
794 
795 /*
796  * If no match is found, check whether we exceed the vCPU's limit
797  * and return the content of the highest valid _standard_ leaf instead.
798  * This is to satisfy the CPUID specification.
799  */
800 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
801                                                   u32 function, u32 index)
802 {
803 	struct kvm_cpuid_entry2 *maxlevel;
804 
805 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
806 	if (!maxlevel || maxlevel->eax >= function)
807 		return NULL;
808 	if (function & 0x80000000) {
809 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
810 		if (!maxlevel)
811 			return NULL;
812 	}
813 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
814 }
815 
816 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
817 {
818 	u32 function = *eax, index = *ecx;
819 	struct kvm_cpuid_entry2 *best;
820 
821 	best = kvm_find_cpuid_entry(vcpu, function, index);
822 
823 	if (!best)
824 		best = check_cpuid_limit(vcpu, function, index);
825 
826 	/*
827 	 * Perfmon not yet supported for L2 guest.
828 	 */
829 	if (is_guest_mode(vcpu) && function == 0xa)
830 		best = NULL;
831 
832 	if (best) {
833 		*eax = best->eax;
834 		*ebx = best->ebx;
835 		*ecx = best->ecx;
836 		*edx = best->edx;
837 	} else
838 		*eax = *ebx = *ecx = *edx = 0;
839 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
840 }
841 EXPORT_SYMBOL_GPL(kvm_cpuid);
842 
843 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
844 {
845 	u32 function, eax, ebx, ecx, edx;
846 
847 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
848 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
849 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
850 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
851 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
852 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
853 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
854 	kvm_x86_ops->skip_emulated_instruction(vcpu);
855 }
856 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
857