xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 483eb062)
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_FPSSE;
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 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
47 {
48 	struct kvm_cpuid_entry2 *best;
49 	struct kvm_lapic *apic = vcpu->arch.apic;
50 
51 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
52 	if (!best)
53 		return;
54 
55 	/* Update OSXSAVE bit */
56 	if (cpu_has_xsave && best->function == 0x1) {
57 		best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
58 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
59 			best->ecx |= bit(X86_FEATURE_OSXSAVE);
60 	}
61 
62 	if (apic) {
63 		if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
64 			apic->lapic_timer.timer_mode_mask = 3 << 17;
65 		else
66 			apic->lapic_timer.timer_mode_mask = 1 << 17;
67 	}
68 
69 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
70 	if (!best) {
71 		vcpu->arch.guest_supported_xcr0 = 0;
72 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
73 	} else {
74 		vcpu->arch.guest_supported_xcr0 =
75 			(best->eax | ((u64)best->edx << 32)) &
76 			host_xcr0 & KVM_SUPPORTED_XCR0;
77 		vcpu->arch.guest_xstate_size =
78 			xstate_required_size(vcpu->arch.guest_supported_xcr0);
79 	}
80 
81 	kvm_pmu_cpuid_update(vcpu);
82 }
83 
84 static int is_efer_nx(void)
85 {
86 	unsigned long long efer = 0;
87 
88 	rdmsrl_safe(MSR_EFER, &efer);
89 	return efer & EFER_NX;
90 }
91 
92 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
93 {
94 	int i;
95 	struct kvm_cpuid_entry2 *e, *entry;
96 
97 	entry = NULL;
98 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
99 		e = &vcpu->arch.cpuid_entries[i];
100 		if (e->function == 0x80000001) {
101 			entry = e;
102 			break;
103 		}
104 	}
105 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
106 		entry->edx &= ~(1 << 20);
107 		printk(KERN_INFO "kvm: guest NX capability removed\n");
108 	}
109 }
110 
111 /* when an old userspace process fills a new kernel module */
112 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
113 			     struct kvm_cpuid *cpuid,
114 			     struct kvm_cpuid_entry __user *entries)
115 {
116 	int r, i;
117 	struct kvm_cpuid_entry *cpuid_entries;
118 
119 	r = -E2BIG;
120 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
121 		goto out;
122 	r = -ENOMEM;
123 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
124 	if (!cpuid_entries)
125 		goto out;
126 	r = -EFAULT;
127 	if (copy_from_user(cpuid_entries, entries,
128 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
129 		goto out_free;
130 	for (i = 0; i < cpuid->nent; i++) {
131 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
132 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
133 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
134 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
135 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
136 		vcpu->arch.cpuid_entries[i].index = 0;
137 		vcpu->arch.cpuid_entries[i].flags = 0;
138 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
139 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
140 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
141 	}
142 	vcpu->arch.cpuid_nent = cpuid->nent;
143 	cpuid_fix_nx_cap(vcpu);
144 	r = 0;
145 	kvm_apic_set_version(vcpu);
146 	kvm_x86_ops->cpuid_update(vcpu);
147 	kvm_update_cpuid(vcpu);
148 
149 out_free:
150 	vfree(cpuid_entries);
151 out:
152 	return r;
153 }
154 
155 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
156 			      struct kvm_cpuid2 *cpuid,
157 			      struct kvm_cpuid_entry2 __user *entries)
158 {
159 	int r;
160 
161 	r = -E2BIG;
162 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
163 		goto out;
164 	r = -EFAULT;
165 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
166 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
167 		goto out;
168 	vcpu->arch.cpuid_nent = cpuid->nent;
169 	kvm_apic_set_version(vcpu);
170 	kvm_x86_ops->cpuid_update(vcpu);
171 	kvm_update_cpuid(vcpu);
172 	return 0;
173 
174 out:
175 	return r;
176 }
177 
178 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
179 			      struct kvm_cpuid2 *cpuid,
180 			      struct kvm_cpuid_entry2 __user *entries)
181 {
182 	int r;
183 
184 	r = -E2BIG;
185 	if (cpuid->nent < vcpu->arch.cpuid_nent)
186 		goto out;
187 	r = -EFAULT;
188 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
189 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
190 		goto out;
191 	return 0;
192 
193 out:
194 	cpuid->nent = vcpu->arch.cpuid_nent;
195 	return r;
196 }
197 
198 static void cpuid_mask(u32 *word, int wordnum)
199 {
200 	*word &= boot_cpu_data.x86_capability[wordnum];
201 }
202 
203 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
204 			   u32 index)
205 {
206 	entry->function = function;
207 	entry->index = index;
208 	cpuid_count(entry->function, entry->index,
209 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
210 	entry->flags = 0;
211 }
212 
213 static bool supported_xcr0_bit(unsigned bit)
214 {
215 	u64 mask = ((u64)1 << bit);
216 
217 	return mask & KVM_SUPPORTED_XCR0 & host_xcr0;
218 }
219 
220 #define F(x) bit(X86_FEATURE_##x)
221 
222 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
223 				   u32 func, u32 index, int *nent, int maxnent)
224 {
225 	switch (func) {
226 	case 0:
227 		entry->eax = 1;		/* only one leaf currently */
228 		++*nent;
229 		break;
230 	case 1:
231 		entry->ecx = F(MOVBE);
232 		++*nent;
233 		break;
234 	default:
235 		break;
236 	}
237 
238 	entry->function = func;
239 	entry->index = index;
240 
241 	return 0;
242 }
243 
244 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
245 				 u32 index, int *nent, int maxnent)
246 {
247 	int r;
248 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
249 #ifdef CONFIG_X86_64
250 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
251 				? F(GBPAGES) : 0;
252 	unsigned f_lm = F(LM);
253 #else
254 	unsigned f_gbpages = 0;
255 	unsigned f_lm = 0;
256 #endif
257 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
258 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
259 
260 	/* cpuid 1.edx */
261 	const u32 kvm_supported_word0_x86_features =
262 		F(FPU) | F(VME) | F(DE) | F(PSE) |
263 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
264 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
265 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
266 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
267 		0 /* Reserved, DS, ACPI */ | F(MMX) |
268 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
269 		0 /* HTT, TM, Reserved, PBE */;
270 	/* cpuid 0x80000001.edx */
271 	const u32 kvm_supported_word1_x86_features =
272 		F(FPU) | F(VME) | F(DE) | F(PSE) |
273 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
274 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
275 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
276 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
277 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
278 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
279 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
280 	/* cpuid 1.ecx */
281 	const u32 kvm_supported_word4_x86_features =
282 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
283 		0 /* DS-CPL, VMX, SMX, EST */ |
284 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
285 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
286 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
287 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
288 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
289 		F(F16C) | F(RDRAND);
290 	/* cpuid 0x80000001.ecx */
291 	const u32 kvm_supported_word6_x86_features =
292 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
293 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
294 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
295 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
296 
297 	/* cpuid 0xC0000001.edx */
298 	const u32 kvm_supported_word5_x86_features =
299 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
300 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
301 		F(PMM) | F(PMM_EN);
302 
303 	/* cpuid 7.0.ebx */
304 	const u32 kvm_supported_word9_x86_features =
305 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
306 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM);
307 
308 	/* all calls to cpuid_count() should be made on the same cpu */
309 	get_cpu();
310 
311 	r = -E2BIG;
312 
313 	if (*nent >= maxnent)
314 		goto out;
315 
316 	do_cpuid_1_ent(entry, function, index);
317 	++*nent;
318 
319 	switch (function) {
320 	case 0:
321 		entry->eax = min(entry->eax, (u32)0xd);
322 		break;
323 	case 1:
324 		entry->edx &= kvm_supported_word0_x86_features;
325 		cpuid_mask(&entry->edx, 0);
326 		entry->ecx &= kvm_supported_word4_x86_features;
327 		cpuid_mask(&entry->ecx, 4);
328 		/* we support x2apic emulation even if host does not support
329 		 * it since we emulate x2apic in software */
330 		entry->ecx |= F(X2APIC);
331 		break;
332 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
333 	 * may return different values. This forces us to get_cpu() before
334 	 * issuing the first command, and also to emulate this annoying behavior
335 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
336 	case 2: {
337 		int t, times = entry->eax & 0xff;
338 
339 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
340 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
341 		for (t = 1; t < times; ++t) {
342 			if (*nent >= maxnent)
343 				goto out;
344 
345 			do_cpuid_1_ent(&entry[t], function, 0);
346 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
347 			++*nent;
348 		}
349 		break;
350 	}
351 	/* function 4 has additional index. */
352 	case 4: {
353 		int i, cache_type;
354 
355 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
356 		/* read more entries until cache_type is zero */
357 		for (i = 1; ; ++i) {
358 			if (*nent >= maxnent)
359 				goto out;
360 
361 			cache_type = entry[i - 1].eax & 0x1f;
362 			if (!cache_type)
363 				break;
364 			do_cpuid_1_ent(&entry[i], function, i);
365 			entry[i].flags |=
366 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
367 			++*nent;
368 		}
369 		break;
370 	}
371 	case 7: {
372 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
373 		/* Mask ebx against host capability word 9 */
374 		if (index == 0) {
375 			entry->ebx &= kvm_supported_word9_x86_features;
376 			cpuid_mask(&entry->ebx, 9);
377 			// TSC_ADJUST is emulated
378 			entry->ebx |= F(TSC_ADJUST);
379 		} else
380 			entry->ebx = 0;
381 		entry->eax = 0;
382 		entry->ecx = 0;
383 		entry->edx = 0;
384 		break;
385 	}
386 	case 9:
387 		break;
388 	case 0xa: { /* Architectural Performance Monitoring */
389 		struct x86_pmu_capability cap;
390 		union cpuid10_eax eax;
391 		union cpuid10_edx edx;
392 
393 		perf_get_x86_pmu_capability(&cap);
394 
395 		/*
396 		 * Only support guest architectural pmu on a host
397 		 * with architectural pmu.
398 		 */
399 		if (!cap.version)
400 			memset(&cap, 0, sizeof(cap));
401 
402 		eax.split.version_id = min(cap.version, 2);
403 		eax.split.num_counters = cap.num_counters_gp;
404 		eax.split.bit_width = cap.bit_width_gp;
405 		eax.split.mask_length = cap.events_mask_len;
406 
407 		edx.split.num_counters_fixed = cap.num_counters_fixed;
408 		edx.split.bit_width_fixed = cap.bit_width_fixed;
409 		edx.split.reserved = 0;
410 
411 		entry->eax = eax.full;
412 		entry->ebx = cap.events_mask;
413 		entry->ecx = 0;
414 		entry->edx = edx.full;
415 		break;
416 	}
417 	/* function 0xb has additional index. */
418 	case 0xb: {
419 		int i, level_type;
420 
421 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
422 		/* read more entries until level_type is zero */
423 		for (i = 1; ; ++i) {
424 			if (*nent >= maxnent)
425 				goto out;
426 
427 			level_type = entry[i - 1].ecx & 0xff00;
428 			if (!level_type)
429 				break;
430 			do_cpuid_1_ent(&entry[i], function, i);
431 			entry[i].flags |=
432 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
433 			++*nent;
434 		}
435 		break;
436 	}
437 	case 0xd: {
438 		int idx, i;
439 
440 		entry->eax &= host_xcr0 & KVM_SUPPORTED_XCR0;
441 		entry->edx &= (host_xcr0 & KVM_SUPPORTED_XCR0) >> 32;
442 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
443 		for (idx = 1, i = 1; idx < 64; ++idx) {
444 			if (*nent >= maxnent)
445 				goto out;
446 
447 			do_cpuid_1_ent(&entry[i], function, idx);
448 			if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
449 				continue;
450 			entry[i].flags |=
451 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
452 			++*nent;
453 			++i;
454 		}
455 		break;
456 	}
457 	case KVM_CPUID_SIGNATURE: {
458 		static const char signature[12] = "KVMKVMKVM\0\0";
459 		const u32 *sigptr = (const u32 *)signature;
460 		entry->eax = KVM_CPUID_FEATURES;
461 		entry->ebx = sigptr[0];
462 		entry->ecx = sigptr[1];
463 		entry->edx = sigptr[2];
464 		break;
465 	}
466 	case KVM_CPUID_FEATURES:
467 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
468 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
469 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
470 			     (1 << KVM_FEATURE_ASYNC_PF) |
471 			     (1 << KVM_FEATURE_PV_EOI) |
472 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
473 			     (1 << KVM_FEATURE_PV_UNHALT);
474 
475 		if (sched_info_on())
476 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
477 
478 		entry->ebx = 0;
479 		entry->ecx = 0;
480 		entry->edx = 0;
481 		break;
482 	case 0x80000000:
483 		entry->eax = min(entry->eax, 0x8000001a);
484 		break;
485 	case 0x80000001:
486 		entry->edx &= kvm_supported_word1_x86_features;
487 		cpuid_mask(&entry->edx, 1);
488 		entry->ecx &= kvm_supported_word6_x86_features;
489 		cpuid_mask(&entry->ecx, 6);
490 		break;
491 	case 0x80000008: {
492 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
493 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
494 		unsigned phys_as = entry->eax & 0xff;
495 
496 		if (!g_phys_as)
497 			g_phys_as = phys_as;
498 		entry->eax = g_phys_as | (virt_as << 8);
499 		entry->ebx = entry->edx = 0;
500 		break;
501 	}
502 	case 0x80000019:
503 		entry->ecx = entry->edx = 0;
504 		break;
505 	case 0x8000001a:
506 		break;
507 	case 0x8000001d:
508 		break;
509 	/*Add support for Centaur's CPUID instruction*/
510 	case 0xC0000000:
511 		/*Just support up to 0xC0000004 now*/
512 		entry->eax = min(entry->eax, 0xC0000004);
513 		break;
514 	case 0xC0000001:
515 		entry->edx &= kvm_supported_word5_x86_features;
516 		cpuid_mask(&entry->edx, 5);
517 		break;
518 	case 3: /* Processor serial number */
519 	case 5: /* MONITOR/MWAIT */
520 	case 6: /* Thermal management */
521 	case 0x80000007: /* Advanced power management */
522 	case 0xC0000002:
523 	case 0xC0000003:
524 	case 0xC0000004:
525 	default:
526 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
527 		break;
528 	}
529 
530 	kvm_x86_ops->set_supported_cpuid(function, entry);
531 
532 	r = 0;
533 
534 out:
535 	put_cpu();
536 
537 	return r;
538 }
539 
540 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
541 			u32 idx, int *nent, int maxnent, unsigned int type)
542 {
543 	if (type == KVM_GET_EMULATED_CPUID)
544 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
545 
546 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
547 }
548 
549 #undef F
550 
551 struct kvm_cpuid_param {
552 	u32 func;
553 	u32 idx;
554 	bool has_leaf_count;
555 	bool (*qualifier)(const struct kvm_cpuid_param *param);
556 };
557 
558 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
559 {
560 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
561 }
562 
563 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
564 				 __u32 num_entries, unsigned int ioctl_type)
565 {
566 	int i;
567 	__u32 pad[3];
568 
569 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
570 		return false;
571 
572 	/*
573 	 * We want to make sure that ->padding is being passed clean from
574 	 * userspace in case we want to use it for something in the future.
575 	 *
576 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
577 	 * have to give ourselves satisfied only with the emulated side. /me
578 	 * sheds a tear.
579 	 */
580 	for (i = 0; i < num_entries; i++) {
581 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
582 			return true;
583 
584 		if (pad[0] || pad[1] || pad[2])
585 			return true;
586 	}
587 	return false;
588 }
589 
590 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
591 			    struct kvm_cpuid_entry2 __user *entries,
592 			    unsigned int type)
593 {
594 	struct kvm_cpuid_entry2 *cpuid_entries;
595 	int limit, nent = 0, r = -E2BIG, i;
596 	u32 func;
597 	static const struct kvm_cpuid_param param[] = {
598 		{ .func = 0, .has_leaf_count = true },
599 		{ .func = 0x80000000, .has_leaf_count = true },
600 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
601 		{ .func = KVM_CPUID_SIGNATURE },
602 		{ .func = KVM_CPUID_FEATURES },
603 	};
604 
605 	if (cpuid->nent < 1)
606 		goto out;
607 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
608 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
609 
610 	if (sanity_check_entries(entries, cpuid->nent, type))
611 		return -EINVAL;
612 
613 	r = -ENOMEM;
614 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
615 	if (!cpuid_entries)
616 		goto out;
617 
618 	r = 0;
619 	for (i = 0; i < ARRAY_SIZE(param); i++) {
620 		const struct kvm_cpuid_param *ent = &param[i];
621 
622 		if (ent->qualifier && !ent->qualifier(ent))
623 			continue;
624 
625 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
626 				&nent, cpuid->nent, type);
627 
628 		if (r)
629 			goto out_free;
630 
631 		if (!ent->has_leaf_count)
632 			continue;
633 
634 		limit = cpuid_entries[nent - 1].eax;
635 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
636 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
637 				     &nent, cpuid->nent, type);
638 
639 		if (r)
640 			goto out_free;
641 	}
642 
643 	r = -EFAULT;
644 	if (copy_to_user(entries, cpuid_entries,
645 			 nent * sizeof(struct kvm_cpuid_entry2)))
646 		goto out_free;
647 	cpuid->nent = nent;
648 	r = 0;
649 
650 out_free:
651 	vfree(cpuid_entries);
652 out:
653 	return r;
654 }
655 
656 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
657 {
658 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
659 	int j, nent = vcpu->arch.cpuid_nent;
660 
661 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
662 	/* when no next entry is found, the current entry[i] is reselected */
663 	for (j = i + 1; ; j = (j + 1) % nent) {
664 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
665 		if (ej->function == e->function) {
666 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
667 			return j;
668 		}
669 	}
670 	return 0; /* silence gcc, even though control never reaches here */
671 }
672 
673 /* find an entry with matching function, matching index (if needed), and that
674  * should be read next (if it's stateful) */
675 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
676 	u32 function, u32 index)
677 {
678 	if (e->function != function)
679 		return 0;
680 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
681 		return 0;
682 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
683 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
684 		return 0;
685 	return 1;
686 }
687 
688 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
689 					      u32 function, u32 index)
690 {
691 	int i;
692 	struct kvm_cpuid_entry2 *best = NULL;
693 
694 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
695 		struct kvm_cpuid_entry2 *e;
696 
697 		e = &vcpu->arch.cpuid_entries[i];
698 		if (is_matching_cpuid_entry(e, function, index)) {
699 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
700 				move_to_next_stateful_cpuid_entry(vcpu, i);
701 			best = e;
702 			break;
703 		}
704 	}
705 	return best;
706 }
707 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
708 
709 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
710 {
711 	struct kvm_cpuid_entry2 *best;
712 
713 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
714 	if (!best || best->eax < 0x80000008)
715 		goto not_found;
716 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
717 	if (best)
718 		return best->eax & 0xff;
719 not_found:
720 	return 36;
721 }
722 
723 /*
724  * If no match is found, check whether we exceed the vCPU's limit
725  * and return the content of the highest valid _standard_ leaf instead.
726  * This is to satisfy the CPUID specification.
727  */
728 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
729                                                   u32 function, u32 index)
730 {
731 	struct kvm_cpuid_entry2 *maxlevel;
732 
733 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
734 	if (!maxlevel || maxlevel->eax >= function)
735 		return NULL;
736 	if (function & 0x80000000) {
737 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
738 		if (!maxlevel)
739 			return NULL;
740 	}
741 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
742 }
743 
744 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
745 {
746 	u32 function = *eax, index = *ecx;
747 	struct kvm_cpuid_entry2 *best;
748 
749 	best = kvm_find_cpuid_entry(vcpu, function, index);
750 
751 	if (!best)
752 		best = check_cpuid_limit(vcpu, function, index);
753 
754 	if (best) {
755 		*eax = best->eax;
756 		*ebx = best->ebx;
757 		*ecx = best->ecx;
758 		*edx = best->edx;
759 	} else
760 		*eax = *ebx = *ecx = *edx = 0;
761 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
762 }
763 EXPORT_SYMBOL_GPL(kvm_cpuid);
764 
765 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
766 {
767 	u32 function, eax, ebx, ecx, edx;
768 
769 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
770 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
771 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
772 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
773 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
774 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
775 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
776 	kvm_x86_ops->skip_emulated_instruction(vcpu);
777 }
778 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
779