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
28static 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
49u64 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
61int 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
120static 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
128static 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
147int 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;
157not_found:
158	return 36;
159}
160EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
161
162/* when an old userspace process fills a new kernel module */
163int 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
199out_free:
200	vfree(cpuid_entries);
201out:
202	return r;
203}
204
205int 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);
222out:
223	return r;
224}
225
226int 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
241out:
242	cpuid->nent = vcpu->arch.cpuid_nent;
243	return r;
244}
245
246static void cpuid_mask(u32 *word, int wordnum)
247{
248	*word &= boot_cpu_data.x86_capability[wordnum];
249}
250
251static 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
261static 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
283static 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				cpuid_mask(&entry[i].eax, 10);
513				entry[i].ebx = 0;
514				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
515					entry[i].ebx =
516						xstate_required_size(supported,
517								     true);
518			} else {
519				if (entry[i].eax == 0 || !(supported & mask))
520					continue;
521				if (WARN_ON_ONCE(entry[i].ecx & 1))
522					continue;
523			}
524			entry[i].ecx = 0;
525			entry[i].edx = 0;
526			entry[i].flags |=
527			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
528			++*nent;
529			++i;
530		}
531		break;
532	}
533	case KVM_CPUID_SIGNATURE: {
534		static const char signature[12] = "KVMKVMKVM\0\0";
535		const u32 *sigptr = (const u32 *)signature;
536		entry->eax = KVM_CPUID_FEATURES;
537		entry->ebx = sigptr[0];
538		entry->ecx = sigptr[1];
539		entry->edx = sigptr[2];
540		break;
541	}
542	case KVM_CPUID_FEATURES:
543		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
544			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
545			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
546			     (1 << KVM_FEATURE_ASYNC_PF) |
547			     (1 << KVM_FEATURE_PV_EOI) |
548			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
549			     (1 << KVM_FEATURE_PV_UNHALT);
550
551		if (sched_info_on())
552			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
553
554		entry->ebx = 0;
555		entry->ecx = 0;
556		entry->edx = 0;
557		break;
558	case 0x80000000:
559		entry->eax = min(entry->eax, 0x8000001a);
560		break;
561	case 0x80000001:
562		entry->edx &= kvm_supported_word1_x86_features;
563		cpuid_mask(&entry->edx, 1);
564		entry->ecx &= kvm_supported_word6_x86_features;
565		cpuid_mask(&entry->ecx, 6);
566		break;
567	case 0x80000007: /* Advanced power management */
568		/* invariant TSC is CPUID.80000007H:EDX[8] */
569		entry->edx &= (1 << 8);
570		/* mask against host */
571		entry->edx &= boot_cpu_data.x86_power;
572		entry->eax = entry->ebx = entry->ecx = 0;
573		break;
574	case 0x80000008: {
575		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
576		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
577		unsigned phys_as = entry->eax & 0xff;
578
579		if (!g_phys_as)
580			g_phys_as = phys_as;
581		entry->eax = g_phys_as | (virt_as << 8);
582		entry->ebx = entry->edx = 0;
583		break;
584	}
585	case 0x80000019:
586		entry->ecx = entry->edx = 0;
587		break;
588	case 0x8000001a:
589		break;
590	case 0x8000001d:
591		break;
592	/*Add support for Centaur's CPUID instruction*/
593	case 0xC0000000:
594		/*Just support up to 0xC0000004 now*/
595		entry->eax = min(entry->eax, 0xC0000004);
596		break;
597	case 0xC0000001:
598		entry->edx &= kvm_supported_word5_x86_features;
599		cpuid_mask(&entry->edx, 5);
600		break;
601	case 3: /* Processor serial number */
602	case 5: /* MONITOR/MWAIT */
603	case 0xC0000002:
604	case 0xC0000003:
605	case 0xC0000004:
606	default:
607		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
608		break;
609	}
610
611	kvm_x86_ops->set_supported_cpuid(function, entry);
612
613	r = 0;
614
615out:
616	put_cpu();
617
618	return r;
619}
620
621static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
622			u32 idx, int *nent, int maxnent, unsigned int type)
623{
624	if (type == KVM_GET_EMULATED_CPUID)
625		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
626
627	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
628}
629
630#undef F
631
632struct kvm_cpuid_param {
633	u32 func;
634	u32 idx;
635	bool has_leaf_count;
636	bool (*qualifier)(const struct kvm_cpuid_param *param);
637};
638
639static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
640{
641	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
642}
643
644static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
645				 __u32 num_entries, unsigned int ioctl_type)
646{
647	int i;
648	__u32 pad[3];
649
650	if (ioctl_type != KVM_GET_EMULATED_CPUID)
651		return false;
652
653	/*
654	 * We want to make sure that ->padding is being passed clean from
655	 * userspace in case we want to use it for something in the future.
656	 *
657	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
658	 * have to give ourselves satisfied only with the emulated side. /me
659	 * sheds a tear.
660	 */
661	for (i = 0; i < num_entries; i++) {
662		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
663			return true;
664
665		if (pad[0] || pad[1] || pad[2])
666			return true;
667	}
668	return false;
669}
670
671int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
672			    struct kvm_cpuid_entry2 __user *entries,
673			    unsigned int type)
674{
675	struct kvm_cpuid_entry2 *cpuid_entries;
676	int limit, nent = 0, r = -E2BIG, i;
677	u32 func;
678	static const struct kvm_cpuid_param param[] = {
679		{ .func = 0, .has_leaf_count = true },
680		{ .func = 0x80000000, .has_leaf_count = true },
681		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
682		{ .func = KVM_CPUID_SIGNATURE },
683		{ .func = KVM_CPUID_FEATURES },
684	};
685
686	if (cpuid->nent < 1)
687		goto out;
688	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
689		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
690
691	if (sanity_check_entries(entries, cpuid->nent, type))
692		return -EINVAL;
693
694	r = -ENOMEM;
695	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
696	if (!cpuid_entries)
697		goto out;
698
699	r = 0;
700	for (i = 0; i < ARRAY_SIZE(param); i++) {
701		const struct kvm_cpuid_param *ent = &param[i];
702
703		if (ent->qualifier && !ent->qualifier(ent))
704			continue;
705
706		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
707				&nent, cpuid->nent, type);
708
709		if (r)
710			goto out_free;
711
712		if (!ent->has_leaf_count)
713			continue;
714
715		limit = cpuid_entries[nent - 1].eax;
716		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
717			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
718				     &nent, cpuid->nent, type);
719
720		if (r)
721			goto out_free;
722	}
723
724	r = -EFAULT;
725	if (copy_to_user(entries, cpuid_entries,
726			 nent * sizeof(struct kvm_cpuid_entry2)))
727		goto out_free;
728	cpuid->nent = nent;
729	r = 0;
730
731out_free:
732	vfree(cpuid_entries);
733out:
734	return r;
735}
736
737static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
738{
739	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
740	int j, nent = vcpu->arch.cpuid_nent;
741
742	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
743	/* when no next entry is found, the current entry[i] is reselected */
744	for (j = i + 1; ; j = (j + 1) % nent) {
745		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
746		if (ej->function == e->function) {
747			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
748			return j;
749		}
750	}
751	return 0; /* silence gcc, even though control never reaches here */
752}
753
754/* find an entry with matching function, matching index (if needed), and that
755 * should be read next (if it's stateful) */
756static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
757	u32 function, u32 index)
758{
759	if (e->function != function)
760		return 0;
761	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
762		return 0;
763	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
764	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
765		return 0;
766	return 1;
767}
768
769struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
770					      u32 function, u32 index)
771{
772	int i;
773	struct kvm_cpuid_entry2 *best = NULL;
774
775	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
776		struct kvm_cpuid_entry2 *e;
777
778		e = &vcpu->arch.cpuid_entries[i];
779		if (is_matching_cpuid_entry(e, function, index)) {
780			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
781				move_to_next_stateful_cpuid_entry(vcpu, i);
782			best = e;
783			break;
784		}
785	}
786	return best;
787}
788EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
789
790/*
791 * If no match is found, check whether we exceed the vCPU's limit
792 * and return the content of the highest valid _standard_ leaf instead.
793 * This is to satisfy the CPUID specification.
794 */
795static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
796                                                  u32 function, u32 index)
797{
798	struct kvm_cpuid_entry2 *maxlevel;
799
800	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
801	if (!maxlevel || maxlevel->eax >= function)
802		return NULL;
803	if (function & 0x80000000) {
804		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
805		if (!maxlevel)
806			return NULL;
807	}
808	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
809}
810
811void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
812{
813	u32 function = *eax, index = *ecx;
814	struct kvm_cpuid_entry2 *best;
815
816	best = kvm_find_cpuid_entry(vcpu, function, index);
817
818	if (!best)
819		best = check_cpuid_limit(vcpu, function, index);
820
821	/*
822	 * Perfmon not yet supported for L2 guest.
823	 */
824	if (is_guest_mode(vcpu) && function == 0xa)
825		best = NULL;
826
827	if (best) {
828		*eax = best->eax;
829		*ebx = best->ebx;
830		*ecx = best->ecx;
831		*edx = best->edx;
832	} else
833		*eax = *ebx = *ecx = *edx = 0;
834	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
835}
836EXPORT_SYMBOL_GPL(kvm_cpuid);
837
838void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
839{
840	u32 function, eax, ebx, ecx, edx;
841
842	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
843	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
844	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
845	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
846	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
847	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
848	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
849	kvm_x86_ops->skip_emulated_instruction(vcpu);
850}
851EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
852