xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 609e478b)
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/user.h>
20 #include <asm/xsave.h>
21 #include "cpuid.h"
22 #include "lapic.h"
23 #include "mmu.h"
24 #include "trace.h"
25 
26 static u32 xstate_required_size(u64 xstate_bv)
27 {
28 	int feature_bit = 0;
29 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
30 
31 	xstate_bv &= XSTATE_EXTEND_MASK;
32 	while (xstate_bv) {
33 		if (xstate_bv & 0x1) {
34 		        u32 eax, ebx, ecx, edx;
35 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
36 			ret = max(ret, eax + ebx);
37 		}
38 
39 		xstate_bv >>= 1;
40 		feature_bit++;
41 	}
42 
43 	return ret;
44 }
45 
46 u64 kvm_supported_xcr0(void)
47 {
48 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
49 
50 	if (!kvm_x86_ops->mpx_supported())
51 		xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR);
52 
53 	return xcr0;
54 }
55 
56 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
57 {
58 	struct kvm_cpuid_entry2 *best;
59 	struct kvm_lapic *apic = vcpu->arch.apic;
60 
61 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
62 	if (!best)
63 		return 0;
64 
65 	/* Update OSXSAVE bit */
66 	if (cpu_has_xsave && best->function == 0x1) {
67 		best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
68 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
69 			best->ecx |= bit(X86_FEATURE_OSXSAVE);
70 	}
71 
72 	if (apic) {
73 		if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
74 			apic->lapic_timer.timer_mode_mask = 3 << 17;
75 		else
76 			apic->lapic_timer.timer_mode_mask = 1 << 17;
77 	}
78 
79 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
80 	if (!best) {
81 		vcpu->arch.guest_supported_xcr0 = 0;
82 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
83 	} else {
84 		vcpu->arch.guest_supported_xcr0 =
85 			(best->eax | ((u64)best->edx << 32)) &
86 			kvm_supported_xcr0();
87 		vcpu->arch.guest_xstate_size = best->ebx =
88 			xstate_required_size(vcpu->arch.xcr0);
89 	}
90 
91 	/*
92 	 * The existing code assumes virtual address is 48-bit in the canonical
93 	 * address checks; exit if it is ever changed.
94 	 */
95 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
96 	if (best && ((best->eax & 0xff00) >> 8) != 48 &&
97 		((best->eax & 0xff00) >> 8) != 0)
98 		return -EINVAL;
99 
100 	kvm_pmu_cpuid_update(vcpu);
101 	return 0;
102 }
103 
104 static int is_efer_nx(void)
105 {
106 	unsigned long long efer = 0;
107 
108 	rdmsrl_safe(MSR_EFER, &efer);
109 	return efer & EFER_NX;
110 }
111 
112 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
113 {
114 	int i;
115 	struct kvm_cpuid_entry2 *e, *entry;
116 
117 	entry = NULL;
118 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
119 		e = &vcpu->arch.cpuid_entries[i];
120 		if (e->function == 0x80000001) {
121 			entry = e;
122 			break;
123 		}
124 	}
125 	if (entry && (entry->edx & bit(X86_FEATURE_NX)) && !is_efer_nx()) {
126 		entry->edx &= ~bit(X86_FEATURE_NX);
127 		printk(KERN_INFO "kvm: guest NX capability removed\n");
128 	}
129 }
130 
131 /* when an old userspace process fills a new kernel module */
132 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
133 			     struct kvm_cpuid *cpuid,
134 			     struct kvm_cpuid_entry __user *entries)
135 {
136 	int r, i;
137 	struct kvm_cpuid_entry *cpuid_entries;
138 
139 	r = -E2BIG;
140 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
141 		goto out;
142 	r = -ENOMEM;
143 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
144 	if (!cpuid_entries)
145 		goto out;
146 	r = -EFAULT;
147 	if (copy_from_user(cpuid_entries, entries,
148 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
149 		goto out_free;
150 	for (i = 0; i < cpuid->nent; i++) {
151 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
152 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
153 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
154 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
155 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
156 		vcpu->arch.cpuid_entries[i].index = 0;
157 		vcpu->arch.cpuid_entries[i].flags = 0;
158 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
159 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
160 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
161 	}
162 	vcpu->arch.cpuid_nent = cpuid->nent;
163 	cpuid_fix_nx_cap(vcpu);
164 	kvm_apic_set_version(vcpu);
165 	kvm_x86_ops->cpuid_update(vcpu);
166 	r = kvm_update_cpuid(vcpu);
167 
168 out_free:
169 	vfree(cpuid_entries);
170 out:
171 	return r;
172 }
173 
174 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
175 			      struct kvm_cpuid2 *cpuid,
176 			      struct kvm_cpuid_entry2 __user *entries)
177 {
178 	int r;
179 
180 	r = -E2BIG;
181 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
182 		goto out;
183 	r = -EFAULT;
184 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
185 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
186 		goto out;
187 	vcpu->arch.cpuid_nent = cpuid->nent;
188 	kvm_apic_set_version(vcpu);
189 	kvm_x86_ops->cpuid_update(vcpu);
190 	r = kvm_update_cpuid(vcpu);
191 out:
192 	return r;
193 }
194 
195 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
196 			      struct kvm_cpuid2 *cpuid,
197 			      struct kvm_cpuid_entry2 __user *entries)
198 {
199 	int r;
200 
201 	r = -E2BIG;
202 	if (cpuid->nent < vcpu->arch.cpuid_nent)
203 		goto out;
204 	r = -EFAULT;
205 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
206 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
207 		goto out;
208 	return 0;
209 
210 out:
211 	cpuid->nent = vcpu->arch.cpuid_nent;
212 	return r;
213 }
214 
215 static void cpuid_mask(u32 *word, int wordnum)
216 {
217 	*word &= boot_cpu_data.x86_capability[wordnum];
218 }
219 
220 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
221 			   u32 index)
222 {
223 	entry->function = function;
224 	entry->index = index;
225 	cpuid_count(entry->function, entry->index,
226 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
227 	entry->flags = 0;
228 }
229 
230 #define F(x) bit(X86_FEATURE_##x)
231 
232 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
233 				   u32 func, u32 index, int *nent, int maxnent)
234 {
235 	switch (func) {
236 	case 0:
237 		entry->eax = 1;		/* only one leaf currently */
238 		++*nent;
239 		break;
240 	case 1:
241 		entry->ecx = F(MOVBE);
242 		++*nent;
243 		break;
244 	default:
245 		break;
246 	}
247 
248 	entry->function = func;
249 	entry->index = index;
250 
251 	return 0;
252 }
253 
254 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
255 				 u32 index, int *nent, int maxnent)
256 {
257 	int r;
258 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
259 #ifdef CONFIG_X86_64
260 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
261 				? F(GBPAGES) : 0;
262 	unsigned f_lm = F(LM);
263 #else
264 	unsigned f_gbpages = 0;
265 	unsigned f_lm = 0;
266 #endif
267 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
268 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
269 	unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
270 
271 	/* cpuid 1.edx */
272 	const u32 kvm_supported_word0_x86_features =
273 		F(FPU) | F(VME) | F(DE) | F(PSE) |
274 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
275 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
276 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
277 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
278 		0 /* Reserved, DS, ACPI */ | F(MMX) |
279 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
280 		0 /* HTT, TM, Reserved, PBE */;
281 	/* cpuid 0x80000001.edx */
282 	const u32 kvm_supported_word1_x86_features =
283 		F(FPU) | F(VME) | F(DE) | F(PSE) |
284 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
285 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
286 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
287 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
288 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
289 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
290 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
291 	/* cpuid 1.ecx */
292 	const u32 kvm_supported_word4_x86_features =
293 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
294 		 * but *not* advertised to guests via CPUID ! */
295 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
296 		0 /* DS-CPL, VMX, SMX, EST */ |
297 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
298 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
299 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
300 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
301 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
302 		F(F16C) | F(RDRAND);
303 	/* cpuid 0x80000001.ecx */
304 	const u32 kvm_supported_word6_x86_features =
305 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
306 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
307 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
308 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
309 
310 	/* cpuid 0xC0000001.edx */
311 	const u32 kvm_supported_word5_x86_features =
312 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
313 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
314 		F(PMM) | F(PMM_EN);
315 
316 	/* cpuid 7.0.ebx */
317 	const u32 kvm_supported_word9_x86_features =
318 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
319 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
320 		F(ADX) | F(SMAP);
321 
322 	/* all calls to cpuid_count() should be made on the same cpu */
323 	get_cpu();
324 
325 	r = -E2BIG;
326 
327 	if (*nent >= maxnent)
328 		goto out;
329 
330 	do_cpuid_1_ent(entry, function, index);
331 	++*nent;
332 
333 	switch (function) {
334 	case 0:
335 		entry->eax = min(entry->eax, (u32)0xd);
336 		break;
337 	case 1:
338 		entry->edx &= kvm_supported_word0_x86_features;
339 		cpuid_mask(&entry->edx, 0);
340 		entry->ecx &= kvm_supported_word4_x86_features;
341 		cpuid_mask(&entry->ecx, 4);
342 		/* we support x2apic emulation even if host does not support
343 		 * it since we emulate x2apic in software */
344 		entry->ecx |= F(X2APIC);
345 		break;
346 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
347 	 * may return different values. This forces us to get_cpu() before
348 	 * issuing the first command, and also to emulate this annoying behavior
349 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
350 	case 2: {
351 		int t, times = entry->eax & 0xff;
352 
353 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
354 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
355 		for (t = 1; t < times; ++t) {
356 			if (*nent >= maxnent)
357 				goto out;
358 
359 			do_cpuid_1_ent(&entry[t], function, 0);
360 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
361 			++*nent;
362 		}
363 		break;
364 	}
365 	/* function 4 has additional index. */
366 	case 4: {
367 		int i, cache_type;
368 
369 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
370 		/* read more entries until cache_type is zero */
371 		for (i = 1; ; ++i) {
372 			if (*nent >= maxnent)
373 				goto out;
374 
375 			cache_type = entry[i - 1].eax & 0x1f;
376 			if (!cache_type)
377 				break;
378 			do_cpuid_1_ent(&entry[i], function, i);
379 			entry[i].flags |=
380 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
381 			++*nent;
382 		}
383 		break;
384 	}
385 	case 7: {
386 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
387 		/* Mask ebx against host capability word 9 */
388 		if (index == 0) {
389 			entry->ebx &= kvm_supported_word9_x86_features;
390 			cpuid_mask(&entry->ebx, 9);
391 			// TSC_ADJUST is emulated
392 			entry->ebx |= F(TSC_ADJUST);
393 		} else
394 			entry->ebx = 0;
395 		entry->eax = 0;
396 		entry->ecx = 0;
397 		entry->edx = 0;
398 		break;
399 	}
400 	case 9:
401 		break;
402 	case 0xa: { /* Architectural Performance Monitoring */
403 		struct x86_pmu_capability cap;
404 		union cpuid10_eax eax;
405 		union cpuid10_edx edx;
406 
407 		perf_get_x86_pmu_capability(&cap);
408 
409 		/*
410 		 * Only support guest architectural pmu on a host
411 		 * with architectural pmu.
412 		 */
413 		if (!cap.version)
414 			memset(&cap, 0, sizeof(cap));
415 
416 		eax.split.version_id = min(cap.version, 2);
417 		eax.split.num_counters = cap.num_counters_gp;
418 		eax.split.bit_width = cap.bit_width_gp;
419 		eax.split.mask_length = cap.events_mask_len;
420 
421 		edx.split.num_counters_fixed = cap.num_counters_fixed;
422 		edx.split.bit_width_fixed = cap.bit_width_fixed;
423 		edx.split.reserved = 0;
424 
425 		entry->eax = eax.full;
426 		entry->ebx = cap.events_mask;
427 		entry->ecx = 0;
428 		entry->edx = edx.full;
429 		break;
430 	}
431 	/* function 0xb has additional index. */
432 	case 0xb: {
433 		int i, level_type;
434 
435 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
436 		/* read more entries until level_type is zero */
437 		for (i = 1; ; ++i) {
438 			if (*nent >= maxnent)
439 				goto out;
440 
441 			level_type = entry[i - 1].ecx & 0xff00;
442 			if (!level_type)
443 				break;
444 			do_cpuid_1_ent(&entry[i], function, i);
445 			entry[i].flags |=
446 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
447 			++*nent;
448 		}
449 		break;
450 	}
451 	case 0xd: {
452 		int idx, i;
453 		u64 supported = kvm_supported_xcr0();
454 
455 		entry->eax &= supported;
456 		entry->edx &= supported >> 32;
457 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
458 		for (idx = 1, i = 1; idx < 64; ++idx) {
459 			u64 mask = ((u64)1 << idx);
460 			if (*nent >= maxnent)
461 				goto out;
462 
463 			do_cpuid_1_ent(&entry[i], function, idx);
464 			if (entry[i].eax == 0 || !(supported & mask))
465 				continue;
466 			entry[i].flags |=
467 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
468 			++*nent;
469 			++i;
470 		}
471 		break;
472 	}
473 	case KVM_CPUID_SIGNATURE: {
474 		static const char signature[12] = "KVMKVMKVM\0\0";
475 		const u32 *sigptr = (const u32 *)signature;
476 		entry->eax = KVM_CPUID_FEATURES;
477 		entry->ebx = sigptr[0];
478 		entry->ecx = sigptr[1];
479 		entry->edx = sigptr[2];
480 		break;
481 	}
482 	case KVM_CPUID_FEATURES:
483 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
484 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
485 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
486 			     (1 << KVM_FEATURE_ASYNC_PF) |
487 			     (1 << KVM_FEATURE_PV_EOI) |
488 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
489 			     (1 << KVM_FEATURE_PV_UNHALT);
490 
491 		if (sched_info_on())
492 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
493 
494 		entry->ebx = 0;
495 		entry->ecx = 0;
496 		entry->edx = 0;
497 		break;
498 	case 0x80000000:
499 		entry->eax = min(entry->eax, 0x8000001a);
500 		break;
501 	case 0x80000001:
502 		entry->edx &= kvm_supported_word1_x86_features;
503 		cpuid_mask(&entry->edx, 1);
504 		entry->ecx &= kvm_supported_word6_x86_features;
505 		cpuid_mask(&entry->ecx, 6);
506 		break;
507 	case 0x80000007: /* Advanced power management */
508 		/* invariant TSC is CPUID.80000007H:EDX[8] */
509 		entry->edx &= (1 << 8);
510 		/* mask against host */
511 		entry->edx &= boot_cpu_data.x86_power;
512 		entry->eax = entry->ebx = entry->ecx = 0;
513 		break;
514 	case 0x80000008: {
515 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
516 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
517 		unsigned phys_as = entry->eax & 0xff;
518 
519 		if (!g_phys_as)
520 			g_phys_as = phys_as;
521 		entry->eax = g_phys_as | (virt_as << 8);
522 		entry->ebx = entry->edx = 0;
523 		break;
524 	}
525 	case 0x80000019:
526 		entry->ecx = entry->edx = 0;
527 		break;
528 	case 0x8000001a:
529 		break;
530 	case 0x8000001d:
531 		break;
532 	/*Add support for Centaur's CPUID instruction*/
533 	case 0xC0000000:
534 		/*Just support up to 0xC0000004 now*/
535 		entry->eax = min(entry->eax, 0xC0000004);
536 		break;
537 	case 0xC0000001:
538 		entry->edx &= kvm_supported_word5_x86_features;
539 		cpuid_mask(&entry->edx, 5);
540 		break;
541 	case 3: /* Processor serial number */
542 	case 5: /* MONITOR/MWAIT */
543 	case 6: /* Thermal management */
544 	case 0xC0000002:
545 	case 0xC0000003:
546 	case 0xC0000004:
547 	default:
548 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
549 		break;
550 	}
551 
552 	kvm_x86_ops->set_supported_cpuid(function, entry);
553 
554 	r = 0;
555 
556 out:
557 	put_cpu();
558 
559 	return r;
560 }
561 
562 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
563 			u32 idx, int *nent, int maxnent, unsigned int type)
564 {
565 	if (type == KVM_GET_EMULATED_CPUID)
566 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
567 
568 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
569 }
570 
571 #undef F
572 
573 struct kvm_cpuid_param {
574 	u32 func;
575 	u32 idx;
576 	bool has_leaf_count;
577 	bool (*qualifier)(const struct kvm_cpuid_param *param);
578 };
579 
580 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
581 {
582 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
583 }
584 
585 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
586 				 __u32 num_entries, unsigned int ioctl_type)
587 {
588 	int i;
589 	__u32 pad[3];
590 
591 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
592 		return false;
593 
594 	/*
595 	 * We want to make sure that ->padding is being passed clean from
596 	 * userspace in case we want to use it for something in the future.
597 	 *
598 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
599 	 * have to give ourselves satisfied only with the emulated side. /me
600 	 * sheds a tear.
601 	 */
602 	for (i = 0; i < num_entries; i++) {
603 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
604 			return true;
605 
606 		if (pad[0] || pad[1] || pad[2])
607 			return true;
608 	}
609 	return false;
610 }
611 
612 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
613 			    struct kvm_cpuid_entry2 __user *entries,
614 			    unsigned int type)
615 {
616 	struct kvm_cpuid_entry2 *cpuid_entries;
617 	int limit, nent = 0, r = -E2BIG, i;
618 	u32 func;
619 	static const struct kvm_cpuid_param param[] = {
620 		{ .func = 0, .has_leaf_count = true },
621 		{ .func = 0x80000000, .has_leaf_count = true },
622 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
623 		{ .func = KVM_CPUID_SIGNATURE },
624 		{ .func = KVM_CPUID_FEATURES },
625 	};
626 
627 	if (cpuid->nent < 1)
628 		goto out;
629 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
630 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
631 
632 	if (sanity_check_entries(entries, cpuid->nent, type))
633 		return -EINVAL;
634 
635 	r = -ENOMEM;
636 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
637 	if (!cpuid_entries)
638 		goto out;
639 
640 	r = 0;
641 	for (i = 0; i < ARRAY_SIZE(param); i++) {
642 		const struct kvm_cpuid_param *ent = &param[i];
643 
644 		if (ent->qualifier && !ent->qualifier(ent))
645 			continue;
646 
647 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
648 				&nent, cpuid->nent, type);
649 
650 		if (r)
651 			goto out_free;
652 
653 		if (!ent->has_leaf_count)
654 			continue;
655 
656 		limit = cpuid_entries[nent - 1].eax;
657 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
658 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
659 				     &nent, cpuid->nent, type);
660 
661 		if (r)
662 			goto out_free;
663 	}
664 
665 	r = -EFAULT;
666 	if (copy_to_user(entries, cpuid_entries,
667 			 nent * sizeof(struct kvm_cpuid_entry2)))
668 		goto out_free;
669 	cpuid->nent = nent;
670 	r = 0;
671 
672 out_free:
673 	vfree(cpuid_entries);
674 out:
675 	return r;
676 }
677 
678 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
679 {
680 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
681 	int j, nent = vcpu->arch.cpuid_nent;
682 
683 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
684 	/* when no next entry is found, the current entry[i] is reselected */
685 	for (j = i + 1; ; j = (j + 1) % nent) {
686 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
687 		if (ej->function == e->function) {
688 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
689 			return j;
690 		}
691 	}
692 	return 0; /* silence gcc, even though control never reaches here */
693 }
694 
695 /* find an entry with matching function, matching index (if needed), and that
696  * should be read next (if it's stateful) */
697 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
698 	u32 function, u32 index)
699 {
700 	if (e->function != function)
701 		return 0;
702 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
703 		return 0;
704 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
705 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
706 		return 0;
707 	return 1;
708 }
709 
710 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
711 					      u32 function, u32 index)
712 {
713 	int i;
714 	struct kvm_cpuid_entry2 *best = NULL;
715 
716 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
717 		struct kvm_cpuid_entry2 *e;
718 
719 		e = &vcpu->arch.cpuid_entries[i];
720 		if (is_matching_cpuid_entry(e, function, index)) {
721 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
722 				move_to_next_stateful_cpuid_entry(vcpu, i);
723 			best = e;
724 			break;
725 		}
726 	}
727 	return best;
728 }
729 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
730 
731 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
732 {
733 	struct kvm_cpuid_entry2 *best;
734 
735 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
736 	if (!best || best->eax < 0x80000008)
737 		goto not_found;
738 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
739 	if (best)
740 		return best->eax & 0xff;
741 not_found:
742 	return 36;
743 }
744 EXPORT_SYMBOL_GPL(cpuid_maxphyaddr);
745 
746 /*
747  * If no match is found, check whether we exceed the vCPU's limit
748  * and return the content of the highest valid _standard_ leaf instead.
749  * This is to satisfy the CPUID specification.
750  */
751 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
752                                                   u32 function, u32 index)
753 {
754 	struct kvm_cpuid_entry2 *maxlevel;
755 
756 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
757 	if (!maxlevel || maxlevel->eax >= function)
758 		return NULL;
759 	if (function & 0x80000000) {
760 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
761 		if (!maxlevel)
762 			return NULL;
763 	}
764 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
765 }
766 
767 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
768 {
769 	u32 function = *eax, index = *ecx;
770 	struct kvm_cpuid_entry2 *best;
771 
772 	best = kvm_find_cpuid_entry(vcpu, function, index);
773 
774 	if (!best)
775 		best = check_cpuid_limit(vcpu, function, index);
776 
777 	/*
778 	 * Perfmon not yet supported for L2 guest.
779 	 */
780 	if (is_guest_mode(vcpu) && function == 0xa)
781 		best = NULL;
782 
783 	if (best) {
784 		*eax = best->eax;
785 		*ebx = best->ebx;
786 		*ecx = best->ecx;
787 		*edx = best->edx;
788 	} else
789 		*eax = *ebx = *ecx = *edx = 0;
790 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
791 }
792 EXPORT_SYMBOL_GPL(kvm_cpuid);
793 
794 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
795 {
796 	u32 function, eax, ebx, ecx, edx;
797 
798 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
799 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
800 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
801 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
802 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
803 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
804 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
805 	kvm_x86_ops->skip_emulated_instruction(vcpu);
806 }
807 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
808