xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 84d517f3)
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 void 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;
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 	kvm_pmu_cpuid_update(vcpu);
92 }
93 
94 static int is_efer_nx(void)
95 {
96 	unsigned long long efer = 0;
97 
98 	rdmsrl_safe(MSR_EFER, &efer);
99 	return efer & EFER_NX;
100 }
101 
102 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
103 {
104 	int i;
105 	struct kvm_cpuid_entry2 *e, *entry;
106 
107 	entry = NULL;
108 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
109 		e = &vcpu->arch.cpuid_entries[i];
110 		if (e->function == 0x80000001) {
111 			entry = e;
112 			break;
113 		}
114 	}
115 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
116 		entry->edx &= ~(1 << 20);
117 		printk(KERN_INFO "kvm: guest NX capability removed\n");
118 	}
119 }
120 
121 /* when an old userspace process fills a new kernel module */
122 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
123 			     struct kvm_cpuid *cpuid,
124 			     struct kvm_cpuid_entry __user *entries)
125 {
126 	int r, i;
127 	struct kvm_cpuid_entry *cpuid_entries;
128 
129 	r = -E2BIG;
130 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
131 		goto out;
132 	r = -ENOMEM;
133 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
134 	if (!cpuid_entries)
135 		goto out;
136 	r = -EFAULT;
137 	if (copy_from_user(cpuid_entries, entries,
138 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
139 		goto out_free;
140 	for (i = 0; i < cpuid->nent; i++) {
141 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
142 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
143 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
144 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
145 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
146 		vcpu->arch.cpuid_entries[i].index = 0;
147 		vcpu->arch.cpuid_entries[i].flags = 0;
148 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
149 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
150 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
151 	}
152 	vcpu->arch.cpuid_nent = cpuid->nent;
153 	cpuid_fix_nx_cap(vcpu);
154 	r = 0;
155 	kvm_apic_set_version(vcpu);
156 	kvm_x86_ops->cpuid_update(vcpu);
157 	kvm_update_cpuid(vcpu);
158 
159 out_free:
160 	vfree(cpuid_entries);
161 out:
162 	return r;
163 }
164 
165 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
166 			      struct kvm_cpuid2 *cpuid,
167 			      struct kvm_cpuid_entry2 __user *entries)
168 {
169 	int r;
170 
171 	r = -E2BIG;
172 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
173 		goto out;
174 	r = -EFAULT;
175 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
176 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
177 		goto out;
178 	vcpu->arch.cpuid_nent = cpuid->nent;
179 	kvm_apic_set_version(vcpu);
180 	kvm_x86_ops->cpuid_update(vcpu);
181 	kvm_update_cpuid(vcpu);
182 	return 0;
183 
184 out:
185 	return r;
186 }
187 
188 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
189 			      struct kvm_cpuid2 *cpuid,
190 			      struct kvm_cpuid_entry2 __user *entries)
191 {
192 	int r;
193 
194 	r = -E2BIG;
195 	if (cpuid->nent < vcpu->arch.cpuid_nent)
196 		goto out;
197 	r = -EFAULT;
198 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
199 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
200 		goto out;
201 	return 0;
202 
203 out:
204 	cpuid->nent = vcpu->arch.cpuid_nent;
205 	return r;
206 }
207 
208 static void cpuid_mask(u32 *word, int wordnum)
209 {
210 	*word &= boot_cpu_data.x86_capability[wordnum];
211 }
212 
213 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
214 			   u32 index)
215 {
216 	entry->function = function;
217 	entry->index = index;
218 	cpuid_count(entry->function, entry->index,
219 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
220 	entry->flags = 0;
221 }
222 
223 #define F(x) bit(X86_FEATURE_##x)
224 
225 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
226 				   u32 func, u32 index, int *nent, int maxnent)
227 {
228 	switch (func) {
229 	case 0:
230 		entry->eax = 1;		/* only one leaf currently */
231 		++*nent;
232 		break;
233 	case 1:
234 		entry->ecx = F(MOVBE);
235 		++*nent;
236 		break;
237 	default:
238 		break;
239 	}
240 
241 	entry->function = func;
242 	entry->index = index;
243 
244 	return 0;
245 }
246 
247 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
248 				 u32 index, int *nent, int maxnent)
249 {
250 	int r;
251 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
252 #ifdef CONFIG_X86_64
253 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
254 				? F(GBPAGES) : 0;
255 	unsigned f_lm = F(LM);
256 #else
257 	unsigned f_gbpages = 0;
258 	unsigned f_lm = 0;
259 #endif
260 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
261 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
262 	unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
263 
264 	/* cpuid 1.edx */
265 	const u32 kvm_supported_word0_x86_features =
266 		F(FPU) | F(VME) | F(DE) | F(PSE) |
267 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
268 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
269 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
270 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
271 		0 /* Reserved, DS, ACPI */ | F(MMX) |
272 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
273 		0 /* HTT, TM, Reserved, PBE */;
274 	/* cpuid 0x80000001.edx */
275 	const u32 kvm_supported_word1_x86_features =
276 		F(FPU) | F(VME) | F(DE) | F(PSE) |
277 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
278 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
279 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
280 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
281 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
282 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
283 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
284 	/* cpuid 1.ecx */
285 	const u32 kvm_supported_word4_x86_features =
286 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
287 		 * but *not* advertised to guests via CPUID ! */
288 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
289 		0 /* DS-CPL, VMX, SMX, EST */ |
290 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
291 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
292 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
293 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
294 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
295 		F(F16C) | F(RDRAND);
296 	/* cpuid 0x80000001.ecx */
297 	const u32 kvm_supported_word6_x86_features =
298 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
299 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
300 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
301 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
302 
303 	/* cpuid 0xC0000001.edx */
304 	const u32 kvm_supported_word5_x86_features =
305 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
306 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
307 		F(PMM) | F(PMM_EN);
308 
309 	/* cpuid 7.0.ebx */
310 	const u32 kvm_supported_word9_x86_features =
311 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
312 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
313 		F(ADX) | F(SMAP);
314 
315 	/* all calls to cpuid_count() should be made on the same cpu */
316 	get_cpu();
317 
318 	r = -E2BIG;
319 
320 	if (*nent >= maxnent)
321 		goto out;
322 
323 	do_cpuid_1_ent(entry, function, index);
324 	++*nent;
325 
326 	switch (function) {
327 	case 0:
328 		entry->eax = min(entry->eax, (u32)0xd);
329 		break;
330 	case 1:
331 		entry->edx &= kvm_supported_word0_x86_features;
332 		cpuid_mask(&entry->edx, 0);
333 		entry->ecx &= kvm_supported_word4_x86_features;
334 		cpuid_mask(&entry->ecx, 4);
335 		/* we support x2apic emulation even if host does not support
336 		 * it since we emulate x2apic in software */
337 		entry->ecx |= F(X2APIC);
338 		break;
339 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
340 	 * may return different values. This forces us to get_cpu() before
341 	 * issuing the first command, and also to emulate this annoying behavior
342 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
343 	case 2: {
344 		int t, times = entry->eax & 0xff;
345 
346 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
347 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
348 		for (t = 1; t < times; ++t) {
349 			if (*nent >= maxnent)
350 				goto out;
351 
352 			do_cpuid_1_ent(&entry[t], function, 0);
353 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
354 			++*nent;
355 		}
356 		break;
357 	}
358 	/* function 4 has additional index. */
359 	case 4: {
360 		int i, cache_type;
361 
362 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
363 		/* read more entries until cache_type is zero */
364 		for (i = 1; ; ++i) {
365 			if (*nent >= maxnent)
366 				goto out;
367 
368 			cache_type = entry[i - 1].eax & 0x1f;
369 			if (!cache_type)
370 				break;
371 			do_cpuid_1_ent(&entry[i], function, i);
372 			entry[i].flags |=
373 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
374 			++*nent;
375 		}
376 		break;
377 	}
378 	case 7: {
379 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
380 		/* Mask ebx against host capability word 9 */
381 		if (index == 0) {
382 			entry->ebx &= kvm_supported_word9_x86_features;
383 			cpuid_mask(&entry->ebx, 9);
384 			// TSC_ADJUST is emulated
385 			entry->ebx |= F(TSC_ADJUST);
386 		} else
387 			entry->ebx = 0;
388 		entry->eax = 0;
389 		entry->ecx = 0;
390 		entry->edx = 0;
391 		break;
392 	}
393 	case 9:
394 		break;
395 	case 0xa: { /* Architectural Performance Monitoring */
396 		struct x86_pmu_capability cap;
397 		union cpuid10_eax eax;
398 		union cpuid10_edx edx;
399 
400 		perf_get_x86_pmu_capability(&cap);
401 
402 		/*
403 		 * Only support guest architectural pmu on a host
404 		 * with architectural pmu.
405 		 */
406 		if (!cap.version)
407 			memset(&cap, 0, sizeof(cap));
408 
409 		eax.split.version_id = min(cap.version, 2);
410 		eax.split.num_counters = cap.num_counters_gp;
411 		eax.split.bit_width = cap.bit_width_gp;
412 		eax.split.mask_length = cap.events_mask_len;
413 
414 		edx.split.num_counters_fixed = cap.num_counters_fixed;
415 		edx.split.bit_width_fixed = cap.bit_width_fixed;
416 		edx.split.reserved = 0;
417 
418 		entry->eax = eax.full;
419 		entry->ebx = cap.events_mask;
420 		entry->ecx = 0;
421 		entry->edx = edx.full;
422 		break;
423 	}
424 	/* function 0xb has additional index. */
425 	case 0xb: {
426 		int i, level_type;
427 
428 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
429 		/* read more entries until level_type is zero */
430 		for (i = 1; ; ++i) {
431 			if (*nent >= maxnent)
432 				goto out;
433 
434 			level_type = entry[i - 1].ecx & 0xff00;
435 			if (!level_type)
436 				break;
437 			do_cpuid_1_ent(&entry[i], function, i);
438 			entry[i].flags |=
439 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
440 			++*nent;
441 		}
442 		break;
443 	}
444 	case 0xd: {
445 		int idx, i;
446 		u64 supported = kvm_supported_xcr0();
447 
448 		entry->eax &= supported;
449 		entry->edx &= supported >> 32;
450 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
451 		for (idx = 1, i = 1; idx < 64; ++idx) {
452 			u64 mask = ((u64)1 << idx);
453 			if (*nent >= maxnent)
454 				goto out;
455 
456 			do_cpuid_1_ent(&entry[i], function, idx);
457 			if (entry[i].eax == 0 || !(supported & mask))
458 				continue;
459 			entry[i].flags |=
460 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
461 			++*nent;
462 			++i;
463 		}
464 		break;
465 	}
466 	case KVM_CPUID_SIGNATURE: {
467 		static const char signature[12] = "KVMKVMKVM\0\0";
468 		const u32 *sigptr = (const u32 *)signature;
469 		entry->eax = KVM_CPUID_FEATURES;
470 		entry->ebx = sigptr[0];
471 		entry->ecx = sigptr[1];
472 		entry->edx = sigptr[2];
473 		break;
474 	}
475 	case KVM_CPUID_FEATURES:
476 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
477 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
478 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
479 			     (1 << KVM_FEATURE_ASYNC_PF) |
480 			     (1 << KVM_FEATURE_PV_EOI) |
481 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
482 			     (1 << KVM_FEATURE_PV_UNHALT);
483 
484 		if (sched_info_on())
485 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
486 
487 		entry->ebx = 0;
488 		entry->ecx = 0;
489 		entry->edx = 0;
490 		break;
491 	case 0x80000000:
492 		entry->eax = min(entry->eax, 0x8000001a);
493 		break;
494 	case 0x80000001:
495 		entry->edx &= kvm_supported_word1_x86_features;
496 		cpuid_mask(&entry->edx, 1);
497 		entry->ecx &= kvm_supported_word6_x86_features;
498 		cpuid_mask(&entry->ecx, 6);
499 		break;
500 	case 0x80000007: /* Advanced power management */
501 		/* invariant TSC is CPUID.80000007H:EDX[8] */
502 		entry->edx &= (1 << 8);
503 		/* mask against host */
504 		entry->edx &= boot_cpu_data.x86_power;
505 		entry->eax = entry->ebx = entry->ecx = 0;
506 		break;
507 	case 0x80000008: {
508 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
509 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
510 		unsigned phys_as = entry->eax & 0xff;
511 
512 		if (!g_phys_as)
513 			g_phys_as = phys_as;
514 		entry->eax = g_phys_as | (virt_as << 8);
515 		entry->ebx = entry->edx = 0;
516 		break;
517 	}
518 	case 0x80000019:
519 		entry->ecx = entry->edx = 0;
520 		break;
521 	case 0x8000001a:
522 		break;
523 	case 0x8000001d:
524 		break;
525 	/*Add support for Centaur's CPUID instruction*/
526 	case 0xC0000000:
527 		/*Just support up to 0xC0000004 now*/
528 		entry->eax = min(entry->eax, 0xC0000004);
529 		break;
530 	case 0xC0000001:
531 		entry->edx &= kvm_supported_word5_x86_features;
532 		cpuid_mask(&entry->edx, 5);
533 		break;
534 	case 3: /* Processor serial number */
535 	case 5: /* MONITOR/MWAIT */
536 	case 6: /* Thermal management */
537 	case 0xC0000002:
538 	case 0xC0000003:
539 	case 0xC0000004:
540 	default:
541 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
542 		break;
543 	}
544 
545 	kvm_x86_ops->set_supported_cpuid(function, entry);
546 
547 	r = 0;
548 
549 out:
550 	put_cpu();
551 
552 	return r;
553 }
554 
555 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
556 			u32 idx, int *nent, int maxnent, unsigned int type)
557 {
558 	if (type == KVM_GET_EMULATED_CPUID)
559 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
560 
561 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
562 }
563 
564 #undef F
565 
566 struct kvm_cpuid_param {
567 	u32 func;
568 	u32 idx;
569 	bool has_leaf_count;
570 	bool (*qualifier)(const struct kvm_cpuid_param *param);
571 };
572 
573 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
574 {
575 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
576 }
577 
578 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
579 				 __u32 num_entries, unsigned int ioctl_type)
580 {
581 	int i;
582 	__u32 pad[3];
583 
584 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
585 		return false;
586 
587 	/*
588 	 * We want to make sure that ->padding is being passed clean from
589 	 * userspace in case we want to use it for something in the future.
590 	 *
591 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
592 	 * have to give ourselves satisfied only with the emulated side. /me
593 	 * sheds a tear.
594 	 */
595 	for (i = 0; i < num_entries; i++) {
596 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
597 			return true;
598 
599 		if (pad[0] || pad[1] || pad[2])
600 			return true;
601 	}
602 	return false;
603 }
604 
605 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
606 			    struct kvm_cpuid_entry2 __user *entries,
607 			    unsigned int type)
608 {
609 	struct kvm_cpuid_entry2 *cpuid_entries;
610 	int limit, nent = 0, r = -E2BIG, i;
611 	u32 func;
612 	static const struct kvm_cpuid_param param[] = {
613 		{ .func = 0, .has_leaf_count = true },
614 		{ .func = 0x80000000, .has_leaf_count = true },
615 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
616 		{ .func = KVM_CPUID_SIGNATURE },
617 		{ .func = KVM_CPUID_FEATURES },
618 	};
619 
620 	if (cpuid->nent < 1)
621 		goto out;
622 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
623 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
624 
625 	if (sanity_check_entries(entries, cpuid->nent, type))
626 		return -EINVAL;
627 
628 	r = -ENOMEM;
629 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
630 	if (!cpuid_entries)
631 		goto out;
632 
633 	r = 0;
634 	for (i = 0; i < ARRAY_SIZE(param); i++) {
635 		const struct kvm_cpuid_param *ent = &param[i];
636 
637 		if (ent->qualifier && !ent->qualifier(ent))
638 			continue;
639 
640 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
641 				&nent, cpuid->nent, type);
642 
643 		if (r)
644 			goto out_free;
645 
646 		if (!ent->has_leaf_count)
647 			continue;
648 
649 		limit = cpuid_entries[nent - 1].eax;
650 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
651 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
652 				     &nent, cpuid->nent, type);
653 
654 		if (r)
655 			goto out_free;
656 	}
657 
658 	r = -EFAULT;
659 	if (copy_to_user(entries, cpuid_entries,
660 			 nent * sizeof(struct kvm_cpuid_entry2)))
661 		goto out_free;
662 	cpuid->nent = nent;
663 	r = 0;
664 
665 out_free:
666 	vfree(cpuid_entries);
667 out:
668 	return r;
669 }
670 
671 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
672 {
673 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
674 	int j, nent = vcpu->arch.cpuid_nent;
675 
676 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
677 	/* when no next entry is found, the current entry[i] is reselected */
678 	for (j = i + 1; ; j = (j + 1) % nent) {
679 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
680 		if (ej->function == e->function) {
681 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
682 			return j;
683 		}
684 	}
685 	return 0; /* silence gcc, even though control never reaches here */
686 }
687 
688 /* find an entry with matching function, matching index (if needed), and that
689  * should be read next (if it's stateful) */
690 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
691 	u32 function, u32 index)
692 {
693 	if (e->function != function)
694 		return 0;
695 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
696 		return 0;
697 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
698 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
699 		return 0;
700 	return 1;
701 }
702 
703 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
704 					      u32 function, u32 index)
705 {
706 	int i;
707 	struct kvm_cpuid_entry2 *best = NULL;
708 
709 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
710 		struct kvm_cpuid_entry2 *e;
711 
712 		e = &vcpu->arch.cpuid_entries[i];
713 		if (is_matching_cpuid_entry(e, function, index)) {
714 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
715 				move_to_next_stateful_cpuid_entry(vcpu, i);
716 			best = e;
717 			break;
718 		}
719 	}
720 	return best;
721 }
722 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
723 
724 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
725 {
726 	struct kvm_cpuid_entry2 *best;
727 
728 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
729 	if (!best || best->eax < 0x80000008)
730 		goto not_found;
731 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
732 	if (best)
733 		return best->eax & 0xff;
734 not_found:
735 	return 36;
736 }
737 EXPORT_SYMBOL_GPL(cpuid_maxphyaddr);
738 
739 /*
740  * If no match is found, check whether we exceed the vCPU's limit
741  * and return the content of the highest valid _standard_ leaf instead.
742  * This is to satisfy the CPUID specification.
743  */
744 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
745                                                   u32 function, u32 index)
746 {
747 	struct kvm_cpuid_entry2 *maxlevel;
748 
749 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
750 	if (!maxlevel || maxlevel->eax >= function)
751 		return NULL;
752 	if (function & 0x80000000) {
753 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
754 		if (!maxlevel)
755 			return NULL;
756 	}
757 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
758 }
759 
760 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
761 {
762 	u32 function = *eax, index = *ecx;
763 	struct kvm_cpuid_entry2 *best;
764 
765 	best = kvm_find_cpuid_entry(vcpu, function, index);
766 
767 	if (!best)
768 		best = check_cpuid_limit(vcpu, function, index);
769 
770 	if (best) {
771 		*eax = best->eax;
772 		*ebx = best->ebx;
773 		*ecx = best->ecx;
774 		*edx = best->edx;
775 	} else
776 		*eax = *ebx = *ecx = *edx = 0;
777 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
778 }
779 EXPORT_SYMBOL_GPL(kvm_cpuid);
780 
781 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
782 {
783 	u32 function, eax, ebx, ecx, edx;
784 
785 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
786 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
787 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
788 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
789 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
790 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
791 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
792 	kvm_x86_ops->skip_emulated_instruction(vcpu);
793 }
794 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
795