xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 9cfc5c90)
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 u64 kvm_supported_xcr0(void)
50 {
51 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
52 
53 	if (!kvm_x86_ops->mpx_supported())
54 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
55 
56 	return xcr0;
57 }
58 
59 #define F(x) bit(X86_FEATURE_##x)
60 
61 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
62 {
63 	struct kvm_cpuid_entry2 *best;
64 	struct kvm_lapic *apic = vcpu->arch.apic;
65 
66 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
67 	if (!best)
68 		return 0;
69 
70 	/* Update OSXSAVE bit */
71 	if (cpu_has_xsave && best->function == 0x1) {
72 		best->ecx &= ~F(OSXSAVE);
73 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
74 			best->ecx |= F(OSXSAVE);
75 	}
76 
77 	if (apic) {
78 		if (best->ecx & F(TSC_DEADLINE_TIMER))
79 			apic->lapic_timer.timer_mode_mask = 3 << 17;
80 		else
81 			apic->lapic_timer.timer_mode_mask = 1 << 17;
82 	}
83 
84 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
85 	if (!best) {
86 		vcpu->arch.guest_supported_xcr0 = 0;
87 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
88 	} else {
89 		vcpu->arch.guest_supported_xcr0 =
90 			(best->eax | ((u64)best->edx << 32)) &
91 			kvm_supported_xcr0();
92 		vcpu->arch.guest_xstate_size = best->ebx =
93 			xstate_required_size(vcpu->arch.xcr0, false);
94 	}
95 
96 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
97 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
98 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
99 
100 	vcpu->arch.eager_fpu = use_eager_fpu() || guest_cpuid_has_mpx(vcpu);
101 	if (vcpu->arch.eager_fpu)
102 		kvm_x86_ops->fpu_activate(vcpu);
103 
104 	/*
105 	 * The existing code assumes virtual address is 48-bit in the canonical
106 	 * address checks; exit if it is ever changed.
107 	 */
108 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
109 	if (best && ((best->eax & 0xff00) >> 8) != 48 &&
110 		((best->eax & 0xff00) >> 8) != 0)
111 		return -EINVAL;
112 
113 	/* Update physical-address width */
114 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
115 
116 	kvm_pmu_refresh(vcpu);
117 	return 0;
118 }
119 
120 static int is_efer_nx(void)
121 {
122 	unsigned long long efer = 0;
123 
124 	rdmsrl_safe(MSR_EFER, &efer);
125 	return efer & EFER_NX;
126 }
127 
128 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
129 {
130 	int i;
131 	struct kvm_cpuid_entry2 *e, *entry;
132 
133 	entry = NULL;
134 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
135 		e = &vcpu->arch.cpuid_entries[i];
136 		if (e->function == 0x80000001) {
137 			entry = e;
138 			break;
139 		}
140 	}
141 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
142 		entry->edx &= ~F(NX);
143 		printk(KERN_INFO "kvm: guest NX capability removed\n");
144 	}
145 }
146 
147 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
148 {
149 	struct kvm_cpuid_entry2 *best;
150 
151 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
152 	if (!best || best->eax < 0x80000008)
153 		goto not_found;
154 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
155 	if (best)
156 		return best->eax & 0xff;
157 not_found:
158 	return 36;
159 }
160 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
161 
162 /* when an old userspace process fills a new kernel module */
163 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
164 			     struct kvm_cpuid *cpuid,
165 			     struct kvm_cpuid_entry __user *entries)
166 {
167 	int r, i;
168 	struct kvm_cpuid_entry *cpuid_entries;
169 
170 	r = -E2BIG;
171 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
172 		goto out;
173 	r = -ENOMEM;
174 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
175 	if (!cpuid_entries)
176 		goto out;
177 	r = -EFAULT;
178 	if (copy_from_user(cpuid_entries, entries,
179 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
180 		goto out_free;
181 	for (i = 0; i < cpuid->nent; i++) {
182 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
183 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
184 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
185 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
186 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
187 		vcpu->arch.cpuid_entries[i].index = 0;
188 		vcpu->arch.cpuid_entries[i].flags = 0;
189 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
190 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
191 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
192 	}
193 	vcpu->arch.cpuid_nent = cpuid->nent;
194 	cpuid_fix_nx_cap(vcpu);
195 	kvm_apic_set_version(vcpu);
196 	kvm_x86_ops->cpuid_update(vcpu);
197 	r = kvm_update_cpuid(vcpu);
198 
199 out_free:
200 	vfree(cpuid_entries);
201 out:
202 	return r;
203 }
204 
205 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
206 			      struct kvm_cpuid2 *cpuid,
207 			      struct kvm_cpuid_entry2 __user *entries)
208 {
209 	int r;
210 
211 	r = -E2BIG;
212 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
213 		goto out;
214 	r = -EFAULT;
215 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
216 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
217 		goto out;
218 	vcpu->arch.cpuid_nent = cpuid->nent;
219 	kvm_apic_set_version(vcpu);
220 	kvm_x86_ops->cpuid_update(vcpu);
221 	r = kvm_update_cpuid(vcpu);
222 out:
223 	return r;
224 }
225 
226 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
227 			      struct kvm_cpuid2 *cpuid,
228 			      struct kvm_cpuid_entry2 __user *entries)
229 {
230 	int r;
231 
232 	r = -E2BIG;
233 	if (cpuid->nent < vcpu->arch.cpuid_nent)
234 		goto out;
235 	r = -EFAULT;
236 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
237 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
238 		goto out;
239 	return 0;
240 
241 out:
242 	cpuid->nent = vcpu->arch.cpuid_nent;
243 	return r;
244 }
245 
246 static void cpuid_mask(u32 *word, int wordnum)
247 {
248 	*word &= boot_cpu_data.x86_capability[wordnum];
249 }
250 
251 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
252 			   u32 index)
253 {
254 	entry->function = function;
255 	entry->index = index;
256 	cpuid_count(entry->function, entry->index,
257 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
258 	entry->flags = 0;
259 }
260 
261 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
262 				   u32 func, u32 index, int *nent, int maxnent)
263 {
264 	switch (func) {
265 	case 0:
266 		entry->eax = 1;		/* only one leaf currently */
267 		++*nent;
268 		break;
269 	case 1:
270 		entry->ecx = F(MOVBE);
271 		++*nent;
272 		break;
273 	default:
274 		break;
275 	}
276 
277 	entry->function = func;
278 	entry->index = index;
279 
280 	return 0;
281 }
282 
283 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
284 				 u32 index, int *nent, int maxnent)
285 {
286 	int r;
287 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
288 #ifdef CONFIG_X86_64
289 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
290 				? F(GBPAGES) : 0;
291 	unsigned f_lm = F(LM);
292 #else
293 	unsigned f_gbpages = 0;
294 	unsigned f_lm = 0;
295 #endif
296 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
297 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
298 	unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
299 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
300 
301 	/* cpuid 1.edx */
302 	const u32 kvm_supported_word0_x86_features =
303 		F(FPU) | F(VME) | F(DE) | F(PSE) |
304 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
305 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
306 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
307 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
308 		0 /* Reserved, DS, ACPI */ | F(MMX) |
309 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
310 		0 /* HTT, TM, Reserved, PBE */;
311 	/* cpuid 0x80000001.edx */
312 	const u32 kvm_supported_word1_x86_features =
313 		F(FPU) | F(VME) | F(DE) | F(PSE) |
314 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
315 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
316 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
317 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
318 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
319 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
320 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
321 	/* cpuid 1.ecx */
322 	const u32 kvm_supported_word4_x86_features =
323 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
324 		 * but *not* advertised to guests via CPUID ! */
325 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
326 		0 /* DS-CPL, VMX, SMX, EST */ |
327 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
328 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
329 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
330 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
331 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
332 		F(F16C) | F(RDRAND);
333 	/* cpuid 0x80000001.ecx */
334 	const u32 kvm_supported_word6_x86_features =
335 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
336 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
337 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
338 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
339 
340 	/* cpuid 0xC0000001.edx */
341 	const u32 kvm_supported_word5_x86_features =
342 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
343 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
344 		F(PMM) | F(PMM_EN);
345 
346 	/* cpuid 7.0.ebx */
347 	const u32 kvm_supported_word9_x86_features =
348 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
349 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
350 		F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
351 		F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
352 
353 	/* cpuid 0xD.1.eax */
354 	const u32 kvm_supported_word10_x86_features =
355 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
356 
357 	/* all calls to cpuid_count() should be made on the same cpu */
358 	get_cpu();
359 
360 	r = -E2BIG;
361 
362 	if (*nent >= maxnent)
363 		goto out;
364 
365 	do_cpuid_1_ent(entry, function, index);
366 	++*nent;
367 
368 	switch (function) {
369 	case 0:
370 		entry->eax = min(entry->eax, (u32)0xd);
371 		break;
372 	case 1:
373 		entry->edx &= kvm_supported_word0_x86_features;
374 		cpuid_mask(&entry->edx, 0);
375 		entry->ecx &= kvm_supported_word4_x86_features;
376 		cpuid_mask(&entry->ecx, 4);
377 		/* we support x2apic emulation even if host does not support
378 		 * it since we emulate x2apic in software */
379 		entry->ecx |= F(X2APIC);
380 		break;
381 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
382 	 * may return different values. This forces us to get_cpu() before
383 	 * issuing the first command, and also to emulate this annoying behavior
384 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
385 	case 2: {
386 		int t, times = entry->eax & 0xff;
387 
388 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
389 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
390 		for (t = 1; t < times; ++t) {
391 			if (*nent >= maxnent)
392 				goto out;
393 
394 			do_cpuid_1_ent(&entry[t], function, 0);
395 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
396 			++*nent;
397 		}
398 		break;
399 	}
400 	/* function 4 has additional index. */
401 	case 4: {
402 		int i, cache_type;
403 
404 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
405 		/* read more entries until cache_type is zero */
406 		for (i = 1; ; ++i) {
407 			if (*nent >= maxnent)
408 				goto out;
409 
410 			cache_type = entry[i - 1].eax & 0x1f;
411 			if (!cache_type)
412 				break;
413 			do_cpuid_1_ent(&entry[i], function, i);
414 			entry[i].flags |=
415 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
416 			++*nent;
417 		}
418 		break;
419 	}
420 	case 6: /* Thermal management */
421 		entry->eax = 0x4; /* allow ARAT */
422 		entry->ebx = 0;
423 		entry->ecx = 0;
424 		entry->edx = 0;
425 		break;
426 	case 7: {
427 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
428 		/* Mask ebx against host capability word 9 */
429 		if (index == 0) {
430 			entry->ebx &= kvm_supported_word9_x86_features;
431 			cpuid_mask(&entry->ebx, 9);
432 			// TSC_ADJUST is emulated
433 			entry->ebx |= F(TSC_ADJUST);
434 		} else
435 			entry->ebx = 0;
436 		entry->eax = 0;
437 		entry->ecx = 0;
438 		entry->edx = 0;
439 		break;
440 	}
441 	case 9:
442 		break;
443 	case 0xa: { /* Architectural Performance Monitoring */
444 		struct x86_pmu_capability cap;
445 		union cpuid10_eax eax;
446 		union cpuid10_edx edx;
447 
448 		perf_get_x86_pmu_capability(&cap);
449 
450 		/*
451 		 * Only support guest architectural pmu on a host
452 		 * with architectural pmu.
453 		 */
454 		if (!cap.version)
455 			memset(&cap, 0, sizeof(cap));
456 
457 		eax.split.version_id = min(cap.version, 2);
458 		eax.split.num_counters = cap.num_counters_gp;
459 		eax.split.bit_width = cap.bit_width_gp;
460 		eax.split.mask_length = cap.events_mask_len;
461 
462 		edx.split.num_counters_fixed = cap.num_counters_fixed;
463 		edx.split.bit_width_fixed = cap.bit_width_fixed;
464 		edx.split.reserved = 0;
465 
466 		entry->eax = eax.full;
467 		entry->ebx = cap.events_mask;
468 		entry->ecx = 0;
469 		entry->edx = edx.full;
470 		break;
471 	}
472 	/* function 0xb has additional index. */
473 	case 0xb: {
474 		int i, level_type;
475 
476 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
477 		/* read more entries until level_type is zero */
478 		for (i = 1; ; ++i) {
479 			if (*nent >= maxnent)
480 				goto out;
481 
482 			level_type = entry[i - 1].ecx & 0xff00;
483 			if (!level_type)
484 				break;
485 			do_cpuid_1_ent(&entry[i], function, i);
486 			entry[i].flags |=
487 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
488 			++*nent;
489 		}
490 		break;
491 	}
492 	case 0xd: {
493 		int idx, i;
494 		u64 supported = kvm_supported_xcr0();
495 
496 		entry->eax &= supported;
497 		entry->ebx = xstate_required_size(supported, false);
498 		entry->ecx = entry->ebx;
499 		entry->edx &= supported >> 32;
500 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
501 		if (!supported)
502 			break;
503 
504 		for (idx = 1, i = 1; idx < 64; ++idx) {
505 			u64 mask = ((u64)1 << idx);
506 			if (*nent >= maxnent)
507 				goto out;
508 
509 			do_cpuid_1_ent(&entry[i], function, idx);
510 			if (idx == 1) {
511 				entry[i].eax &= kvm_supported_word10_x86_features;
512 				entry[i].ebx = 0;
513 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
514 					entry[i].ebx =
515 						xstate_required_size(supported,
516 								     true);
517 			} else {
518 				if (entry[i].eax == 0 || !(supported & mask))
519 					continue;
520 				if (WARN_ON_ONCE(entry[i].ecx & 1))
521 					continue;
522 			}
523 			entry[i].ecx = 0;
524 			entry[i].edx = 0;
525 			entry[i].flags |=
526 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
527 			++*nent;
528 			++i;
529 		}
530 		break;
531 	}
532 	case KVM_CPUID_SIGNATURE: {
533 		static const char signature[12] = "KVMKVMKVM\0\0";
534 		const u32 *sigptr = (const u32 *)signature;
535 		entry->eax = KVM_CPUID_FEATURES;
536 		entry->ebx = sigptr[0];
537 		entry->ecx = sigptr[1];
538 		entry->edx = sigptr[2];
539 		break;
540 	}
541 	case KVM_CPUID_FEATURES:
542 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
543 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
544 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
545 			     (1 << KVM_FEATURE_ASYNC_PF) |
546 			     (1 << KVM_FEATURE_PV_EOI) |
547 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
548 			     (1 << KVM_FEATURE_PV_UNHALT);
549 
550 		if (sched_info_on())
551 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
552 
553 		entry->ebx = 0;
554 		entry->ecx = 0;
555 		entry->edx = 0;
556 		break;
557 	case 0x80000000:
558 		entry->eax = min(entry->eax, 0x8000001a);
559 		break;
560 	case 0x80000001:
561 		entry->edx &= kvm_supported_word1_x86_features;
562 		cpuid_mask(&entry->edx, 1);
563 		entry->ecx &= kvm_supported_word6_x86_features;
564 		cpuid_mask(&entry->ecx, 6);
565 		break;
566 	case 0x80000007: /* Advanced power management */
567 		/* invariant TSC is CPUID.80000007H:EDX[8] */
568 		entry->edx &= (1 << 8);
569 		/* mask against host */
570 		entry->edx &= boot_cpu_data.x86_power;
571 		entry->eax = entry->ebx = entry->ecx = 0;
572 		break;
573 	case 0x80000008: {
574 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
575 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
576 		unsigned phys_as = entry->eax & 0xff;
577 
578 		if (!g_phys_as)
579 			g_phys_as = phys_as;
580 		entry->eax = g_phys_as | (virt_as << 8);
581 		entry->ebx = entry->edx = 0;
582 		break;
583 	}
584 	case 0x80000019:
585 		entry->ecx = entry->edx = 0;
586 		break;
587 	case 0x8000001a:
588 		break;
589 	case 0x8000001d:
590 		break;
591 	/*Add support for Centaur's CPUID instruction*/
592 	case 0xC0000000:
593 		/*Just support up to 0xC0000004 now*/
594 		entry->eax = min(entry->eax, 0xC0000004);
595 		break;
596 	case 0xC0000001:
597 		entry->edx &= kvm_supported_word5_x86_features;
598 		cpuid_mask(&entry->edx, 5);
599 		break;
600 	case 3: /* Processor serial number */
601 	case 5: /* MONITOR/MWAIT */
602 	case 0xC0000002:
603 	case 0xC0000003:
604 	case 0xC0000004:
605 	default:
606 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
607 		break;
608 	}
609 
610 	kvm_x86_ops->set_supported_cpuid(function, entry);
611 
612 	r = 0;
613 
614 out:
615 	put_cpu();
616 
617 	return r;
618 }
619 
620 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
621 			u32 idx, int *nent, int maxnent, unsigned int type)
622 {
623 	if (type == KVM_GET_EMULATED_CPUID)
624 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
625 
626 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
627 }
628 
629 #undef F
630 
631 struct kvm_cpuid_param {
632 	u32 func;
633 	u32 idx;
634 	bool has_leaf_count;
635 	bool (*qualifier)(const struct kvm_cpuid_param *param);
636 };
637 
638 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
639 {
640 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
641 }
642 
643 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
644 				 __u32 num_entries, unsigned int ioctl_type)
645 {
646 	int i;
647 	__u32 pad[3];
648 
649 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
650 		return false;
651 
652 	/*
653 	 * We want to make sure that ->padding is being passed clean from
654 	 * userspace in case we want to use it for something in the future.
655 	 *
656 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
657 	 * have to give ourselves satisfied only with the emulated side. /me
658 	 * sheds a tear.
659 	 */
660 	for (i = 0; i < num_entries; i++) {
661 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
662 			return true;
663 
664 		if (pad[0] || pad[1] || pad[2])
665 			return true;
666 	}
667 	return false;
668 }
669 
670 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
671 			    struct kvm_cpuid_entry2 __user *entries,
672 			    unsigned int type)
673 {
674 	struct kvm_cpuid_entry2 *cpuid_entries;
675 	int limit, nent = 0, r = -E2BIG, i;
676 	u32 func;
677 	static const struct kvm_cpuid_param param[] = {
678 		{ .func = 0, .has_leaf_count = true },
679 		{ .func = 0x80000000, .has_leaf_count = true },
680 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
681 		{ .func = KVM_CPUID_SIGNATURE },
682 		{ .func = KVM_CPUID_FEATURES },
683 	};
684 
685 	if (cpuid->nent < 1)
686 		goto out;
687 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
688 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
689 
690 	if (sanity_check_entries(entries, cpuid->nent, type))
691 		return -EINVAL;
692 
693 	r = -ENOMEM;
694 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
695 	if (!cpuid_entries)
696 		goto out;
697 
698 	r = 0;
699 	for (i = 0; i < ARRAY_SIZE(param); i++) {
700 		const struct kvm_cpuid_param *ent = &param[i];
701 
702 		if (ent->qualifier && !ent->qualifier(ent))
703 			continue;
704 
705 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
706 				&nent, cpuid->nent, type);
707 
708 		if (r)
709 			goto out_free;
710 
711 		if (!ent->has_leaf_count)
712 			continue;
713 
714 		limit = cpuid_entries[nent - 1].eax;
715 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
716 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
717 				     &nent, cpuid->nent, type);
718 
719 		if (r)
720 			goto out_free;
721 	}
722 
723 	r = -EFAULT;
724 	if (copy_to_user(entries, cpuid_entries,
725 			 nent * sizeof(struct kvm_cpuid_entry2)))
726 		goto out_free;
727 	cpuid->nent = nent;
728 	r = 0;
729 
730 out_free:
731 	vfree(cpuid_entries);
732 out:
733 	return r;
734 }
735 
736 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
737 {
738 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
739 	int j, nent = vcpu->arch.cpuid_nent;
740 
741 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
742 	/* when no next entry is found, the current entry[i] is reselected */
743 	for (j = i + 1; ; j = (j + 1) % nent) {
744 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
745 		if (ej->function == e->function) {
746 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
747 			return j;
748 		}
749 	}
750 	return 0; /* silence gcc, even though control never reaches here */
751 }
752 
753 /* find an entry with matching function, matching index (if needed), and that
754  * should be read next (if it's stateful) */
755 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
756 	u32 function, u32 index)
757 {
758 	if (e->function != function)
759 		return 0;
760 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
761 		return 0;
762 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
763 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
764 		return 0;
765 	return 1;
766 }
767 
768 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
769 					      u32 function, u32 index)
770 {
771 	int i;
772 	struct kvm_cpuid_entry2 *best = NULL;
773 
774 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
775 		struct kvm_cpuid_entry2 *e;
776 
777 		e = &vcpu->arch.cpuid_entries[i];
778 		if (is_matching_cpuid_entry(e, function, index)) {
779 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
780 				move_to_next_stateful_cpuid_entry(vcpu, i);
781 			best = e;
782 			break;
783 		}
784 	}
785 	return best;
786 }
787 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
788 
789 /*
790  * If no match is found, check whether we exceed the vCPU's limit
791  * and return the content of the highest valid _standard_ leaf instead.
792  * This is to satisfy the CPUID specification.
793  */
794 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
795                                                   u32 function, u32 index)
796 {
797 	struct kvm_cpuid_entry2 *maxlevel;
798 
799 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
800 	if (!maxlevel || maxlevel->eax >= function)
801 		return NULL;
802 	if (function & 0x80000000) {
803 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
804 		if (!maxlevel)
805 			return NULL;
806 	}
807 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
808 }
809 
810 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
811 {
812 	u32 function = *eax, index = *ecx;
813 	struct kvm_cpuid_entry2 *best;
814 
815 	best = kvm_find_cpuid_entry(vcpu, function, index);
816 
817 	if (!best)
818 		best = check_cpuid_limit(vcpu, function, index);
819 
820 	/*
821 	 * Perfmon not yet supported for L2 guest.
822 	 */
823 	if (is_guest_mode(vcpu) && function == 0xa)
824 		best = NULL;
825 
826 	if (best) {
827 		*eax = best->eax;
828 		*ebx = best->ebx;
829 		*ecx = best->ecx;
830 		*edx = best->edx;
831 	} else
832 		*eax = *ebx = *ecx = *edx = 0;
833 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
834 }
835 EXPORT_SYMBOL_GPL(kvm_cpuid);
836 
837 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
838 {
839 	u32 function, eax, ebx, ecx, edx;
840 
841 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
842 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
843 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
844 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
845 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
846 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
847 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
848 	kvm_x86_ops->skip_emulated_instruction(vcpu);
849 }
850 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
851