xref: /linux/arch/x86/kvm/cpuid.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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, bool compacted)
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, offset;
35 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
36 			offset = compacted ? ret : ebx;
37 			ret = max(ret, offset + eax);
38 		}
39 
40 		xstate_bv >>= 1;
41 		feature_bit++;
42 	}
43 
44 	return ret;
45 }
46 
47 u64 kvm_supported_xcr0(void)
48 {
49 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
50 
51 	if (!kvm_x86_ops->mpx_supported())
52 		xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR);
53 
54 	return xcr0;
55 }
56 
57 #define F(x) bit(X86_FEATURE_##x)
58 
59 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
60 {
61 	struct kvm_cpuid_entry2 *best;
62 	struct kvm_lapic *apic = vcpu->arch.apic;
63 
64 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
65 	if (!best)
66 		return 0;
67 
68 	/* Update OSXSAVE bit */
69 	if (cpu_has_xsave && best->function == 0x1) {
70 		best->ecx &= ~F(OSXSAVE);
71 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
72 			best->ecx |= F(OSXSAVE);
73 	}
74 
75 	if (apic) {
76 		if (best->ecx & F(TSC_DEADLINE_TIMER))
77 			apic->lapic_timer.timer_mode_mask = 3 << 17;
78 		else
79 			apic->lapic_timer.timer_mode_mask = 1 << 17;
80 	}
81 
82 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
83 	if (!best) {
84 		vcpu->arch.guest_supported_xcr0 = 0;
85 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
86 	} else {
87 		vcpu->arch.guest_supported_xcr0 =
88 			(best->eax | ((u64)best->edx << 32)) &
89 			kvm_supported_xcr0();
90 		vcpu->arch.guest_xstate_size = best->ebx =
91 			xstate_required_size(vcpu->arch.xcr0, false);
92 	}
93 
94 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
95 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
96 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
97 
98 	/*
99 	 * The existing code assumes virtual address is 48-bit in the canonical
100 	 * address checks; exit if it is ever changed.
101 	 */
102 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
103 	if (best && ((best->eax & 0xff00) >> 8) != 48 &&
104 		((best->eax & 0xff00) >> 8) != 0)
105 		return -EINVAL;
106 
107 	kvm_pmu_cpuid_update(vcpu);
108 	return 0;
109 }
110 
111 static int is_efer_nx(void)
112 {
113 	unsigned long long efer = 0;
114 
115 	rdmsrl_safe(MSR_EFER, &efer);
116 	return efer & EFER_NX;
117 }
118 
119 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
120 {
121 	int i;
122 	struct kvm_cpuid_entry2 *e, *entry;
123 
124 	entry = NULL;
125 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
126 		e = &vcpu->arch.cpuid_entries[i];
127 		if (e->function == 0x80000001) {
128 			entry = e;
129 			break;
130 		}
131 	}
132 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
133 		entry->edx &= ~F(NX);
134 		printk(KERN_INFO "kvm: guest NX capability removed\n");
135 	}
136 }
137 
138 /* when an old userspace process fills a new kernel module */
139 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
140 			     struct kvm_cpuid *cpuid,
141 			     struct kvm_cpuid_entry __user *entries)
142 {
143 	int r, i;
144 	struct kvm_cpuid_entry *cpuid_entries;
145 
146 	r = -E2BIG;
147 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
148 		goto out;
149 	r = -ENOMEM;
150 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
151 	if (!cpuid_entries)
152 		goto out;
153 	r = -EFAULT;
154 	if (copy_from_user(cpuid_entries, entries,
155 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
156 		goto out_free;
157 	for (i = 0; i < cpuid->nent; i++) {
158 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
159 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
160 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
161 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
162 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
163 		vcpu->arch.cpuid_entries[i].index = 0;
164 		vcpu->arch.cpuid_entries[i].flags = 0;
165 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
166 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
167 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
168 	}
169 	vcpu->arch.cpuid_nent = cpuid->nent;
170 	cpuid_fix_nx_cap(vcpu);
171 	kvm_apic_set_version(vcpu);
172 	kvm_x86_ops->cpuid_update(vcpu);
173 	r = kvm_update_cpuid(vcpu);
174 
175 out_free:
176 	vfree(cpuid_entries);
177 out:
178 	return r;
179 }
180 
181 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
182 			      struct kvm_cpuid2 *cpuid,
183 			      struct kvm_cpuid_entry2 __user *entries)
184 {
185 	int r;
186 
187 	r = -E2BIG;
188 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
189 		goto out;
190 	r = -EFAULT;
191 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
192 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
193 		goto out;
194 	vcpu->arch.cpuid_nent = cpuid->nent;
195 	kvm_apic_set_version(vcpu);
196 	kvm_x86_ops->cpuid_update(vcpu);
197 	r = kvm_update_cpuid(vcpu);
198 out:
199 	return r;
200 }
201 
202 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
203 			      struct kvm_cpuid2 *cpuid,
204 			      struct kvm_cpuid_entry2 __user *entries)
205 {
206 	int r;
207 
208 	r = -E2BIG;
209 	if (cpuid->nent < vcpu->arch.cpuid_nent)
210 		goto out;
211 	r = -EFAULT;
212 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
213 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
214 		goto out;
215 	return 0;
216 
217 out:
218 	cpuid->nent = vcpu->arch.cpuid_nent;
219 	return r;
220 }
221 
222 static void cpuid_mask(u32 *word, int wordnum)
223 {
224 	*word &= boot_cpu_data.x86_capability[wordnum];
225 }
226 
227 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
228 			   u32 index)
229 {
230 	entry->function = function;
231 	entry->index = index;
232 	cpuid_count(entry->function, entry->index,
233 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
234 	entry->flags = 0;
235 }
236 
237 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
238 				   u32 func, u32 index, int *nent, int maxnent)
239 {
240 	switch (func) {
241 	case 0:
242 		entry->eax = 1;		/* only one leaf currently */
243 		++*nent;
244 		break;
245 	case 1:
246 		entry->ecx = F(MOVBE);
247 		++*nent;
248 		break;
249 	default:
250 		break;
251 	}
252 
253 	entry->function = func;
254 	entry->index = index;
255 
256 	return 0;
257 }
258 
259 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
260 				 u32 index, int *nent, int maxnent)
261 {
262 	int r;
263 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
264 #ifdef CONFIG_X86_64
265 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
266 				? F(GBPAGES) : 0;
267 	unsigned f_lm = F(LM);
268 #else
269 	unsigned f_gbpages = 0;
270 	unsigned f_lm = 0;
271 #endif
272 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
273 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
274 	unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
275 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
276 
277 	/* cpuid 1.edx */
278 	const u32 kvm_supported_word0_x86_features =
279 		F(FPU) | F(VME) | F(DE) | F(PSE) |
280 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
281 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
282 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
283 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
284 		0 /* Reserved, DS, ACPI */ | F(MMX) |
285 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
286 		0 /* HTT, TM, Reserved, PBE */;
287 	/* cpuid 0x80000001.edx */
288 	const u32 kvm_supported_word1_x86_features =
289 		F(FPU) | F(VME) | F(DE) | F(PSE) |
290 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
291 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
292 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
293 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
294 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
295 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
296 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
297 	/* cpuid 1.ecx */
298 	const u32 kvm_supported_word4_x86_features =
299 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
300 		 * but *not* advertised to guests via CPUID ! */
301 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
302 		0 /* DS-CPL, VMX, SMX, EST */ |
303 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
304 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
305 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
306 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
307 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
308 		F(F16C) | F(RDRAND);
309 	/* cpuid 0x80000001.ecx */
310 	const u32 kvm_supported_word6_x86_features =
311 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
312 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
313 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
314 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
315 
316 	/* cpuid 0xC0000001.edx */
317 	const u32 kvm_supported_word5_x86_features =
318 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
319 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
320 		F(PMM) | F(PMM_EN);
321 
322 	/* cpuid 7.0.ebx */
323 	const u32 kvm_supported_word9_x86_features =
324 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
325 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
326 		F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
327 		F(AVX512CD);
328 
329 	/* cpuid 0xD.1.eax */
330 	const u32 kvm_supported_word10_x86_features =
331 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
332 
333 	/* all calls to cpuid_count() should be made on the same cpu */
334 	get_cpu();
335 
336 	r = -E2BIG;
337 
338 	if (*nent >= maxnent)
339 		goto out;
340 
341 	do_cpuid_1_ent(entry, function, index);
342 	++*nent;
343 
344 	switch (function) {
345 	case 0:
346 		entry->eax = min(entry->eax, (u32)0xd);
347 		break;
348 	case 1:
349 		entry->edx &= kvm_supported_word0_x86_features;
350 		cpuid_mask(&entry->edx, 0);
351 		entry->ecx &= kvm_supported_word4_x86_features;
352 		cpuid_mask(&entry->ecx, 4);
353 		/* we support x2apic emulation even if host does not support
354 		 * it since we emulate x2apic in software */
355 		entry->ecx |= F(X2APIC);
356 		break;
357 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
358 	 * may return different values. This forces us to get_cpu() before
359 	 * issuing the first command, and also to emulate this annoying behavior
360 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
361 	case 2: {
362 		int t, times = entry->eax & 0xff;
363 
364 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
365 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
366 		for (t = 1; t < times; ++t) {
367 			if (*nent >= maxnent)
368 				goto out;
369 
370 			do_cpuid_1_ent(&entry[t], function, 0);
371 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
372 			++*nent;
373 		}
374 		break;
375 	}
376 	/* function 4 has additional index. */
377 	case 4: {
378 		int i, cache_type;
379 
380 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
381 		/* read more entries until cache_type is zero */
382 		for (i = 1; ; ++i) {
383 			if (*nent >= maxnent)
384 				goto out;
385 
386 			cache_type = entry[i - 1].eax & 0x1f;
387 			if (!cache_type)
388 				break;
389 			do_cpuid_1_ent(&entry[i], function, i);
390 			entry[i].flags |=
391 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
392 			++*nent;
393 		}
394 		break;
395 	}
396 	case 7: {
397 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
398 		/* Mask ebx against host capability word 9 */
399 		if (index == 0) {
400 			entry->ebx &= kvm_supported_word9_x86_features;
401 			cpuid_mask(&entry->ebx, 9);
402 			// TSC_ADJUST is emulated
403 			entry->ebx |= F(TSC_ADJUST);
404 		} else
405 			entry->ebx = 0;
406 		entry->eax = 0;
407 		entry->ecx = 0;
408 		entry->edx = 0;
409 		break;
410 	}
411 	case 9:
412 		break;
413 	case 0xa: { /* Architectural Performance Monitoring */
414 		struct x86_pmu_capability cap;
415 		union cpuid10_eax eax;
416 		union cpuid10_edx edx;
417 
418 		perf_get_x86_pmu_capability(&cap);
419 
420 		/*
421 		 * Only support guest architectural pmu on a host
422 		 * with architectural pmu.
423 		 */
424 		if (!cap.version)
425 			memset(&cap, 0, sizeof(cap));
426 
427 		eax.split.version_id = min(cap.version, 2);
428 		eax.split.num_counters = cap.num_counters_gp;
429 		eax.split.bit_width = cap.bit_width_gp;
430 		eax.split.mask_length = cap.events_mask_len;
431 
432 		edx.split.num_counters_fixed = cap.num_counters_fixed;
433 		edx.split.bit_width_fixed = cap.bit_width_fixed;
434 		edx.split.reserved = 0;
435 
436 		entry->eax = eax.full;
437 		entry->ebx = cap.events_mask;
438 		entry->ecx = 0;
439 		entry->edx = edx.full;
440 		break;
441 	}
442 	/* function 0xb has additional index. */
443 	case 0xb: {
444 		int i, level_type;
445 
446 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
447 		/* read more entries until level_type is zero */
448 		for (i = 1; ; ++i) {
449 			if (*nent >= maxnent)
450 				goto out;
451 
452 			level_type = entry[i - 1].ecx & 0xff00;
453 			if (!level_type)
454 				break;
455 			do_cpuid_1_ent(&entry[i], function, i);
456 			entry[i].flags |=
457 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
458 			++*nent;
459 		}
460 		break;
461 	}
462 	case 0xd: {
463 		int idx, i;
464 		u64 supported = kvm_supported_xcr0();
465 
466 		entry->eax &= supported;
467 		entry->ebx = xstate_required_size(supported, false);
468 		entry->ecx = entry->ebx;
469 		entry->edx &= supported >> 32;
470 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
471 		if (!supported)
472 			break;
473 
474 		for (idx = 1, i = 1; idx < 64; ++idx) {
475 			u64 mask = ((u64)1 << idx);
476 			if (*nent >= maxnent)
477 				goto out;
478 
479 			do_cpuid_1_ent(&entry[i], function, idx);
480 			if (idx == 1) {
481 				entry[i].eax &= kvm_supported_word10_x86_features;
482 				entry[i].ebx = 0;
483 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
484 					entry[i].ebx =
485 						xstate_required_size(supported,
486 								     true);
487 			} else {
488 				if (entry[i].eax == 0 || !(supported & mask))
489 					continue;
490 				if (WARN_ON_ONCE(entry[i].ecx & 1))
491 					continue;
492 			}
493 			entry[i].ecx = 0;
494 			entry[i].edx = 0;
495 			entry[i].flags |=
496 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
497 			++*nent;
498 			++i;
499 		}
500 		break;
501 	}
502 	case KVM_CPUID_SIGNATURE: {
503 		static const char signature[12] = "KVMKVMKVM\0\0";
504 		const u32 *sigptr = (const u32 *)signature;
505 		entry->eax = KVM_CPUID_FEATURES;
506 		entry->ebx = sigptr[0];
507 		entry->ecx = sigptr[1];
508 		entry->edx = sigptr[2];
509 		break;
510 	}
511 	case KVM_CPUID_FEATURES:
512 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
513 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
514 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
515 			     (1 << KVM_FEATURE_ASYNC_PF) |
516 			     (1 << KVM_FEATURE_PV_EOI) |
517 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
518 			     (1 << KVM_FEATURE_PV_UNHALT);
519 
520 		if (sched_info_on())
521 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
522 
523 		entry->ebx = 0;
524 		entry->ecx = 0;
525 		entry->edx = 0;
526 		break;
527 	case 0x80000000:
528 		entry->eax = min(entry->eax, 0x8000001a);
529 		break;
530 	case 0x80000001:
531 		entry->edx &= kvm_supported_word1_x86_features;
532 		cpuid_mask(&entry->edx, 1);
533 		entry->ecx &= kvm_supported_word6_x86_features;
534 		cpuid_mask(&entry->ecx, 6);
535 		break;
536 	case 0x80000007: /* Advanced power management */
537 		/* invariant TSC is CPUID.80000007H:EDX[8] */
538 		entry->edx &= (1 << 8);
539 		/* mask against host */
540 		entry->edx &= boot_cpu_data.x86_power;
541 		entry->eax = entry->ebx = entry->ecx = 0;
542 		break;
543 	case 0x80000008: {
544 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
545 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
546 		unsigned phys_as = entry->eax & 0xff;
547 
548 		if (!g_phys_as)
549 			g_phys_as = phys_as;
550 		entry->eax = g_phys_as | (virt_as << 8);
551 		entry->ebx = entry->edx = 0;
552 		break;
553 	}
554 	case 0x80000019:
555 		entry->ecx = entry->edx = 0;
556 		break;
557 	case 0x8000001a:
558 		break;
559 	case 0x8000001d:
560 		break;
561 	/*Add support for Centaur's CPUID instruction*/
562 	case 0xC0000000:
563 		/*Just support up to 0xC0000004 now*/
564 		entry->eax = min(entry->eax, 0xC0000004);
565 		break;
566 	case 0xC0000001:
567 		entry->edx &= kvm_supported_word5_x86_features;
568 		cpuid_mask(&entry->edx, 5);
569 		break;
570 	case 3: /* Processor serial number */
571 	case 5: /* MONITOR/MWAIT */
572 	case 6: /* Thermal management */
573 	case 0xC0000002:
574 	case 0xC0000003:
575 	case 0xC0000004:
576 	default:
577 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
578 		break;
579 	}
580 
581 	kvm_x86_ops->set_supported_cpuid(function, entry);
582 
583 	r = 0;
584 
585 out:
586 	put_cpu();
587 
588 	return r;
589 }
590 
591 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
592 			u32 idx, int *nent, int maxnent, unsigned int type)
593 {
594 	if (type == KVM_GET_EMULATED_CPUID)
595 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
596 
597 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
598 }
599 
600 #undef F
601 
602 struct kvm_cpuid_param {
603 	u32 func;
604 	u32 idx;
605 	bool has_leaf_count;
606 	bool (*qualifier)(const struct kvm_cpuid_param *param);
607 };
608 
609 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
610 {
611 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
612 }
613 
614 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
615 				 __u32 num_entries, unsigned int ioctl_type)
616 {
617 	int i;
618 	__u32 pad[3];
619 
620 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
621 		return false;
622 
623 	/*
624 	 * We want to make sure that ->padding is being passed clean from
625 	 * userspace in case we want to use it for something in the future.
626 	 *
627 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
628 	 * have to give ourselves satisfied only with the emulated side. /me
629 	 * sheds a tear.
630 	 */
631 	for (i = 0; i < num_entries; i++) {
632 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
633 			return true;
634 
635 		if (pad[0] || pad[1] || pad[2])
636 			return true;
637 	}
638 	return false;
639 }
640 
641 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
642 			    struct kvm_cpuid_entry2 __user *entries,
643 			    unsigned int type)
644 {
645 	struct kvm_cpuid_entry2 *cpuid_entries;
646 	int limit, nent = 0, r = -E2BIG, i;
647 	u32 func;
648 	static const struct kvm_cpuid_param param[] = {
649 		{ .func = 0, .has_leaf_count = true },
650 		{ .func = 0x80000000, .has_leaf_count = true },
651 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
652 		{ .func = KVM_CPUID_SIGNATURE },
653 		{ .func = KVM_CPUID_FEATURES },
654 	};
655 
656 	if (cpuid->nent < 1)
657 		goto out;
658 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
659 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
660 
661 	if (sanity_check_entries(entries, cpuid->nent, type))
662 		return -EINVAL;
663 
664 	r = -ENOMEM;
665 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
666 	if (!cpuid_entries)
667 		goto out;
668 
669 	r = 0;
670 	for (i = 0; i < ARRAY_SIZE(param); i++) {
671 		const struct kvm_cpuid_param *ent = &param[i];
672 
673 		if (ent->qualifier && !ent->qualifier(ent))
674 			continue;
675 
676 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
677 				&nent, cpuid->nent, type);
678 
679 		if (r)
680 			goto out_free;
681 
682 		if (!ent->has_leaf_count)
683 			continue;
684 
685 		limit = cpuid_entries[nent - 1].eax;
686 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
687 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
688 				     &nent, cpuid->nent, type);
689 
690 		if (r)
691 			goto out_free;
692 	}
693 
694 	r = -EFAULT;
695 	if (copy_to_user(entries, cpuid_entries,
696 			 nent * sizeof(struct kvm_cpuid_entry2)))
697 		goto out_free;
698 	cpuid->nent = nent;
699 	r = 0;
700 
701 out_free:
702 	vfree(cpuid_entries);
703 out:
704 	return r;
705 }
706 
707 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
708 {
709 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
710 	int j, nent = vcpu->arch.cpuid_nent;
711 
712 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
713 	/* when no next entry is found, the current entry[i] is reselected */
714 	for (j = i + 1; ; j = (j + 1) % nent) {
715 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
716 		if (ej->function == e->function) {
717 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
718 			return j;
719 		}
720 	}
721 	return 0; /* silence gcc, even though control never reaches here */
722 }
723 
724 /* find an entry with matching function, matching index (if needed), and that
725  * should be read next (if it's stateful) */
726 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
727 	u32 function, u32 index)
728 {
729 	if (e->function != function)
730 		return 0;
731 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
732 		return 0;
733 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
734 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
735 		return 0;
736 	return 1;
737 }
738 
739 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
740 					      u32 function, u32 index)
741 {
742 	int i;
743 	struct kvm_cpuid_entry2 *best = NULL;
744 
745 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
746 		struct kvm_cpuid_entry2 *e;
747 
748 		e = &vcpu->arch.cpuid_entries[i];
749 		if (is_matching_cpuid_entry(e, function, index)) {
750 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
751 				move_to_next_stateful_cpuid_entry(vcpu, i);
752 			best = e;
753 			break;
754 		}
755 	}
756 	return best;
757 }
758 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
759 
760 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
761 {
762 	struct kvm_cpuid_entry2 *best;
763 
764 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
765 	if (!best || best->eax < 0x80000008)
766 		goto not_found;
767 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
768 	if (best)
769 		return best->eax & 0xff;
770 not_found:
771 	return 36;
772 }
773 EXPORT_SYMBOL_GPL(cpuid_maxphyaddr);
774 
775 /*
776  * If no match is found, check whether we exceed the vCPU's limit
777  * and return the content of the highest valid _standard_ leaf instead.
778  * This is to satisfy the CPUID specification.
779  */
780 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
781                                                   u32 function, u32 index)
782 {
783 	struct kvm_cpuid_entry2 *maxlevel;
784 
785 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
786 	if (!maxlevel || maxlevel->eax >= function)
787 		return NULL;
788 	if (function & 0x80000000) {
789 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
790 		if (!maxlevel)
791 			return NULL;
792 	}
793 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
794 }
795 
796 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
797 {
798 	u32 function = *eax, index = *ecx;
799 	struct kvm_cpuid_entry2 *best;
800 
801 	best = kvm_find_cpuid_entry(vcpu, function, index);
802 
803 	if (!best)
804 		best = check_cpuid_limit(vcpu, function, index);
805 
806 	/*
807 	 * Perfmon not yet supported for L2 guest.
808 	 */
809 	if (is_guest_mode(vcpu) && function == 0xa)
810 		best = NULL;
811 
812 	if (best) {
813 		*eax = best->eax;
814 		*ebx = best->ebx;
815 		*ecx = best->ecx;
816 		*edx = best->edx;
817 	} else
818 		*eax = *ebx = *ecx = *edx = 0;
819 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
820 }
821 EXPORT_SYMBOL_GPL(kvm_cpuid);
822 
823 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
824 {
825 	u32 function, eax, ebx, ecx, edx;
826 
827 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
828 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
829 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
830 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
831 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
832 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
833 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
834 	kvm_x86_ops->skip_emulated_instruction(vcpu);
835 }
836 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
837