xref: /linux/arch/x86/kvm/cpuid.c (revision 293d5b43948309434568f4dcbb36cce4c3c51bd5)
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/export.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 
28 static 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 
49 bool kvm_mpx_supported(void)
50 {
51 	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
52 		 && kvm_x86_ops->mpx_supported());
53 }
54 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
55 
56 u64 kvm_supported_xcr0(void)
57 {
58 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
59 
60 	if (!kvm_mpx_supported())
61 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
62 
63 	return xcr0;
64 }
65 
66 #define F(x) bit(X86_FEATURE_##x)
67 
68 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
69 {
70 	struct kvm_cpuid_entry2 *best;
71 	struct kvm_lapic *apic = vcpu->arch.apic;
72 
73 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
74 	if (!best)
75 		return 0;
76 
77 	/* Update OSXSAVE bit */
78 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
79 		best->ecx &= ~F(OSXSAVE);
80 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
81 			best->ecx |= F(OSXSAVE);
82 	}
83 
84 	if (apic) {
85 		if (best->ecx & F(TSC_DEADLINE_TIMER))
86 			apic->lapic_timer.timer_mode_mask = 3 << 17;
87 		else
88 			apic->lapic_timer.timer_mode_mask = 1 << 17;
89 	}
90 
91 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
92 	if (best) {
93 		/* Update OSPKE bit */
94 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
95 			best->ecx &= ~F(OSPKE);
96 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
97 				best->ecx |= F(OSPKE);
98 		}
99 	}
100 
101 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
102 	if (!best) {
103 		vcpu->arch.guest_supported_xcr0 = 0;
104 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
105 	} else {
106 		vcpu->arch.guest_supported_xcr0 =
107 			(best->eax | ((u64)best->edx << 32)) &
108 			kvm_supported_xcr0();
109 		vcpu->arch.guest_xstate_size = best->ebx =
110 			xstate_required_size(vcpu->arch.xcr0, false);
111 	}
112 
113 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
114 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
115 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
116 
117 	if (use_eager_fpu())
118 		kvm_x86_ops->fpu_activate(vcpu);
119 
120 	/*
121 	 * The existing code assumes virtual address is 48-bit in the canonical
122 	 * address checks; exit if it is ever changed.
123 	 */
124 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
125 	if (best && ((best->eax & 0xff00) >> 8) != 48 &&
126 		((best->eax & 0xff00) >> 8) != 0)
127 		return -EINVAL;
128 
129 	/* Update physical-address width */
130 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
131 
132 	kvm_pmu_refresh(vcpu);
133 	return 0;
134 }
135 
136 static int is_efer_nx(void)
137 {
138 	unsigned long long efer = 0;
139 
140 	rdmsrl_safe(MSR_EFER, &efer);
141 	return efer & EFER_NX;
142 }
143 
144 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
145 {
146 	int i;
147 	struct kvm_cpuid_entry2 *e, *entry;
148 
149 	entry = NULL;
150 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
151 		e = &vcpu->arch.cpuid_entries[i];
152 		if (e->function == 0x80000001) {
153 			entry = e;
154 			break;
155 		}
156 	}
157 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
158 		entry->edx &= ~F(NX);
159 		printk(KERN_INFO "kvm: guest NX capability removed\n");
160 	}
161 }
162 
163 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
164 {
165 	struct kvm_cpuid_entry2 *best;
166 
167 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
168 	if (!best || best->eax < 0x80000008)
169 		goto not_found;
170 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
171 	if (best)
172 		return best->eax & 0xff;
173 not_found:
174 	return 36;
175 }
176 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
177 
178 /* when an old userspace process fills a new kernel module */
179 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
180 			     struct kvm_cpuid *cpuid,
181 			     struct kvm_cpuid_entry __user *entries)
182 {
183 	int r, i;
184 	struct kvm_cpuid_entry *cpuid_entries = NULL;
185 
186 	r = -E2BIG;
187 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
188 		goto out;
189 	r = -ENOMEM;
190 	if (cpuid->nent) {
191 		cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
192 					cpuid->nent);
193 		if (!cpuid_entries)
194 			goto out;
195 		r = -EFAULT;
196 		if (copy_from_user(cpuid_entries, entries,
197 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
198 			goto out;
199 	}
200 	for (i = 0; i < cpuid->nent; i++) {
201 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
202 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
203 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
204 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
205 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
206 		vcpu->arch.cpuid_entries[i].index = 0;
207 		vcpu->arch.cpuid_entries[i].flags = 0;
208 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
209 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
210 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
211 	}
212 	vcpu->arch.cpuid_nent = cpuid->nent;
213 	cpuid_fix_nx_cap(vcpu);
214 	kvm_apic_set_version(vcpu);
215 	kvm_x86_ops->cpuid_update(vcpu);
216 	r = kvm_update_cpuid(vcpu);
217 
218 out:
219 	vfree(cpuid_entries);
220 	return r;
221 }
222 
223 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
224 			      struct kvm_cpuid2 *cpuid,
225 			      struct kvm_cpuid_entry2 __user *entries)
226 {
227 	int r;
228 
229 	r = -E2BIG;
230 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
231 		goto out;
232 	r = -EFAULT;
233 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
234 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
235 		goto out;
236 	vcpu->arch.cpuid_nent = cpuid->nent;
237 	kvm_apic_set_version(vcpu);
238 	kvm_x86_ops->cpuid_update(vcpu);
239 	r = kvm_update_cpuid(vcpu);
240 out:
241 	return r;
242 }
243 
244 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
245 			      struct kvm_cpuid2 *cpuid,
246 			      struct kvm_cpuid_entry2 __user *entries)
247 {
248 	int r;
249 
250 	r = -E2BIG;
251 	if (cpuid->nent < vcpu->arch.cpuid_nent)
252 		goto out;
253 	r = -EFAULT;
254 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
255 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
256 		goto out;
257 	return 0;
258 
259 out:
260 	cpuid->nent = vcpu->arch.cpuid_nent;
261 	return r;
262 }
263 
264 static void cpuid_mask(u32 *word, int wordnum)
265 {
266 	*word &= boot_cpu_data.x86_capability[wordnum];
267 }
268 
269 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
270 			   u32 index)
271 {
272 	entry->function = function;
273 	entry->index = index;
274 	cpuid_count(entry->function, entry->index,
275 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
276 	entry->flags = 0;
277 }
278 
279 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
280 				   u32 func, u32 index, int *nent, int maxnent)
281 {
282 	switch (func) {
283 	case 0:
284 		entry->eax = 1;		/* only one leaf currently */
285 		++*nent;
286 		break;
287 	case 1:
288 		entry->ecx = F(MOVBE);
289 		++*nent;
290 		break;
291 	default:
292 		break;
293 	}
294 
295 	entry->function = func;
296 	entry->index = index;
297 
298 	return 0;
299 }
300 
301 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
302 				 u32 index, int *nent, int maxnent)
303 {
304 	int r;
305 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
306 #ifdef CONFIG_X86_64
307 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
308 				? F(GBPAGES) : 0;
309 	unsigned f_lm = F(LM);
310 #else
311 	unsigned f_gbpages = 0;
312 	unsigned f_lm = 0;
313 #endif
314 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
315 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
316 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
317 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
318 
319 	/* cpuid 1.edx */
320 	const u32 kvm_cpuid_1_edx_x86_features =
321 		F(FPU) | F(VME) | F(DE) | F(PSE) |
322 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
323 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
324 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
325 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
326 		0 /* Reserved, DS, ACPI */ | F(MMX) |
327 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
328 		0 /* HTT, TM, Reserved, PBE */;
329 	/* cpuid 0x80000001.edx */
330 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
331 		F(FPU) | F(VME) | F(DE) | F(PSE) |
332 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
333 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
334 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
335 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
336 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
337 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
338 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
339 	/* cpuid 1.ecx */
340 	const u32 kvm_cpuid_1_ecx_x86_features =
341 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
342 		 * but *not* advertised to guests via CPUID ! */
343 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
344 		0 /* DS-CPL, VMX, SMX, EST */ |
345 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
346 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
347 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
348 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
349 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
350 		F(F16C) | F(RDRAND);
351 	/* cpuid 0x80000001.ecx */
352 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
353 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
354 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
355 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
356 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
357 
358 	/* cpuid 0xC0000001.edx */
359 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
360 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
361 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
362 		F(PMM) | F(PMM_EN);
363 
364 	/* cpuid 7.0.ebx */
365 	const u32 kvm_cpuid_7_0_ebx_x86_features =
366 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
367 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
368 		F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
369 		F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB);
370 
371 	/* cpuid 0xD.1.eax */
372 	const u32 kvm_cpuid_D_1_eax_x86_features =
373 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
374 
375 	/* cpuid 7.0.ecx*/
376 	const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
377 
378 	/* all calls to cpuid_count() should be made on the same cpu */
379 	get_cpu();
380 
381 	r = -E2BIG;
382 
383 	if (*nent >= maxnent)
384 		goto out;
385 
386 	do_cpuid_1_ent(entry, function, index);
387 	++*nent;
388 
389 	switch (function) {
390 	case 0:
391 		entry->eax = min(entry->eax, (u32)0xd);
392 		break;
393 	case 1:
394 		entry->edx &= kvm_cpuid_1_edx_x86_features;
395 		cpuid_mask(&entry->edx, CPUID_1_EDX);
396 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
397 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
398 		/* we support x2apic emulation even if host does not support
399 		 * it since we emulate x2apic in software */
400 		entry->ecx |= F(X2APIC);
401 		break;
402 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
403 	 * may return different values. This forces us to get_cpu() before
404 	 * issuing the first command, and also to emulate this annoying behavior
405 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
406 	case 2: {
407 		int t, times = entry->eax & 0xff;
408 
409 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
410 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
411 		for (t = 1; t < times; ++t) {
412 			if (*nent >= maxnent)
413 				goto out;
414 
415 			do_cpuid_1_ent(&entry[t], function, 0);
416 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
417 			++*nent;
418 		}
419 		break;
420 	}
421 	/* function 4 has additional index. */
422 	case 4: {
423 		int i, cache_type;
424 
425 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
426 		/* read more entries until cache_type is zero */
427 		for (i = 1; ; ++i) {
428 			if (*nent >= maxnent)
429 				goto out;
430 
431 			cache_type = entry[i - 1].eax & 0x1f;
432 			if (!cache_type)
433 				break;
434 			do_cpuid_1_ent(&entry[i], function, i);
435 			entry[i].flags |=
436 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
437 			++*nent;
438 		}
439 		break;
440 	}
441 	case 6: /* Thermal management */
442 		entry->eax = 0x4; /* allow ARAT */
443 		entry->ebx = 0;
444 		entry->ecx = 0;
445 		entry->edx = 0;
446 		break;
447 	case 7: {
448 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
449 		/* Mask ebx against host capability word 9 */
450 		if (index == 0) {
451 			entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
452 			cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
453 			// TSC_ADJUST is emulated
454 			entry->ebx |= F(TSC_ADJUST);
455 			entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
456 			cpuid_mask(&entry->ecx, CPUID_7_ECX);
457 			/* PKU is not yet implemented for shadow paging. */
458 			if (!tdp_enabled)
459 				entry->ecx &= ~F(PKU);
460 		} else {
461 			entry->ebx = 0;
462 			entry->ecx = 0;
463 		}
464 		entry->eax = 0;
465 		entry->edx = 0;
466 		break;
467 	}
468 	case 9:
469 		break;
470 	case 0xa: { /* Architectural Performance Monitoring */
471 		struct x86_pmu_capability cap;
472 		union cpuid10_eax eax;
473 		union cpuid10_edx edx;
474 
475 		perf_get_x86_pmu_capability(&cap);
476 
477 		/*
478 		 * Only support guest architectural pmu on a host
479 		 * with architectural pmu.
480 		 */
481 		if (!cap.version)
482 			memset(&cap, 0, sizeof(cap));
483 
484 		eax.split.version_id = min(cap.version, 2);
485 		eax.split.num_counters = cap.num_counters_gp;
486 		eax.split.bit_width = cap.bit_width_gp;
487 		eax.split.mask_length = cap.events_mask_len;
488 
489 		edx.split.num_counters_fixed = cap.num_counters_fixed;
490 		edx.split.bit_width_fixed = cap.bit_width_fixed;
491 		edx.split.reserved = 0;
492 
493 		entry->eax = eax.full;
494 		entry->ebx = cap.events_mask;
495 		entry->ecx = 0;
496 		entry->edx = edx.full;
497 		break;
498 	}
499 	/* function 0xb has additional index. */
500 	case 0xb: {
501 		int i, level_type;
502 
503 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
504 		/* read more entries until level_type is zero */
505 		for (i = 1; ; ++i) {
506 			if (*nent >= maxnent)
507 				goto out;
508 
509 			level_type = entry[i - 1].ecx & 0xff00;
510 			if (!level_type)
511 				break;
512 			do_cpuid_1_ent(&entry[i], function, i);
513 			entry[i].flags |=
514 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
515 			++*nent;
516 		}
517 		break;
518 	}
519 	case 0xd: {
520 		int idx, i;
521 		u64 supported = kvm_supported_xcr0();
522 
523 		entry->eax &= supported;
524 		entry->ebx = xstate_required_size(supported, false);
525 		entry->ecx = entry->ebx;
526 		entry->edx &= supported >> 32;
527 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
528 		if (!supported)
529 			break;
530 
531 		for (idx = 1, i = 1; idx < 64; ++idx) {
532 			u64 mask = ((u64)1 << idx);
533 			if (*nent >= maxnent)
534 				goto out;
535 
536 			do_cpuid_1_ent(&entry[i], function, idx);
537 			if (idx == 1) {
538 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
539 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
540 				entry[i].ebx = 0;
541 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
542 					entry[i].ebx =
543 						xstate_required_size(supported,
544 								     true);
545 			} else {
546 				if (entry[i].eax == 0 || !(supported & mask))
547 					continue;
548 				if (WARN_ON_ONCE(entry[i].ecx & 1))
549 					continue;
550 			}
551 			entry[i].ecx = 0;
552 			entry[i].edx = 0;
553 			entry[i].flags |=
554 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
555 			++*nent;
556 			++i;
557 		}
558 		break;
559 	}
560 	case KVM_CPUID_SIGNATURE: {
561 		static const char signature[12] = "KVMKVMKVM\0\0";
562 		const u32 *sigptr = (const u32 *)signature;
563 		entry->eax = KVM_CPUID_FEATURES;
564 		entry->ebx = sigptr[0];
565 		entry->ecx = sigptr[1];
566 		entry->edx = sigptr[2];
567 		break;
568 	}
569 	case KVM_CPUID_FEATURES:
570 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
571 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
572 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
573 			     (1 << KVM_FEATURE_ASYNC_PF) |
574 			     (1 << KVM_FEATURE_PV_EOI) |
575 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
576 			     (1 << KVM_FEATURE_PV_UNHALT);
577 
578 		if (sched_info_on())
579 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
580 
581 		entry->ebx = 0;
582 		entry->ecx = 0;
583 		entry->edx = 0;
584 		break;
585 	case 0x80000000:
586 		entry->eax = min(entry->eax, 0x8000001a);
587 		break;
588 	case 0x80000001:
589 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
590 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
591 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
592 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
593 		break;
594 	case 0x80000007: /* Advanced power management */
595 		/* invariant TSC is CPUID.80000007H:EDX[8] */
596 		entry->edx &= (1 << 8);
597 		/* mask against host */
598 		entry->edx &= boot_cpu_data.x86_power;
599 		entry->eax = entry->ebx = entry->ecx = 0;
600 		break;
601 	case 0x80000008: {
602 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
603 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
604 		unsigned phys_as = entry->eax & 0xff;
605 
606 		if (!g_phys_as)
607 			g_phys_as = phys_as;
608 		entry->eax = g_phys_as | (virt_as << 8);
609 		entry->ebx = entry->edx = 0;
610 		break;
611 	}
612 	case 0x80000019:
613 		entry->ecx = entry->edx = 0;
614 		break;
615 	case 0x8000001a:
616 		break;
617 	case 0x8000001d:
618 		break;
619 	/*Add support for Centaur's CPUID instruction*/
620 	case 0xC0000000:
621 		/*Just support up to 0xC0000004 now*/
622 		entry->eax = min(entry->eax, 0xC0000004);
623 		break;
624 	case 0xC0000001:
625 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
626 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
627 		break;
628 	case 3: /* Processor serial number */
629 	case 5: /* MONITOR/MWAIT */
630 	case 0xC0000002:
631 	case 0xC0000003:
632 	case 0xC0000004:
633 	default:
634 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
635 		break;
636 	}
637 
638 	kvm_x86_ops->set_supported_cpuid(function, entry);
639 
640 	r = 0;
641 
642 out:
643 	put_cpu();
644 
645 	return r;
646 }
647 
648 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
649 			u32 idx, int *nent, int maxnent, unsigned int type)
650 {
651 	if (type == KVM_GET_EMULATED_CPUID)
652 		return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
653 
654 	return __do_cpuid_ent(entry, func, idx, nent, maxnent);
655 }
656 
657 #undef F
658 
659 struct kvm_cpuid_param {
660 	u32 func;
661 	u32 idx;
662 	bool has_leaf_count;
663 	bool (*qualifier)(const struct kvm_cpuid_param *param);
664 };
665 
666 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
667 {
668 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
669 }
670 
671 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
672 				 __u32 num_entries, unsigned int ioctl_type)
673 {
674 	int i;
675 	__u32 pad[3];
676 
677 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
678 		return false;
679 
680 	/*
681 	 * We want to make sure that ->padding is being passed clean from
682 	 * userspace in case we want to use it for something in the future.
683 	 *
684 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
685 	 * have to give ourselves satisfied only with the emulated side. /me
686 	 * sheds a tear.
687 	 */
688 	for (i = 0; i < num_entries; i++) {
689 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
690 			return true;
691 
692 		if (pad[0] || pad[1] || pad[2])
693 			return true;
694 	}
695 	return false;
696 }
697 
698 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
699 			    struct kvm_cpuid_entry2 __user *entries,
700 			    unsigned int type)
701 {
702 	struct kvm_cpuid_entry2 *cpuid_entries;
703 	int limit, nent = 0, r = -E2BIG, i;
704 	u32 func;
705 	static const struct kvm_cpuid_param param[] = {
706 		{ .func = 0, .has_leaf_count = true },
707 		{ .func = 0x80000000, .has_leaf_count = true },
708 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
709 		{ .func = KVM_CPUID_SIGNATURE },
710 		{ .func = KVM_CPUID_FEATURES },
711 	};
712 
713 	if (cpuid->nent < 1)
714 		goto out;
715 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
716 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
717 
718 	if (sanity_check_entries(entries, cpuid->nent, type))
719 		return -EINVAL;
720 
721 	r = -ENOMEM;
722 	cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
723 	if (!cpuid_entries)
724 		goto out;
725 
726 	r = 0;
727 	for (i = 0; i < ARRAY_SIZE(param); i++) {
728 		const struct kvm_cpuid_param *ent = &param[i];
729 
730 		if (ent->qualifier && !ent->qualifier(ent))
731 			continue;
732 
733 		r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
734 				&nent, cpuid->nent, type);
735 
736 		if (r)
737 			goto out_free;
738 
739 		if (!ent->has_leaf_count)
740 			continue;
741 
742 		limit = cpuid_entries[nent - 1].eax;
743 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
744 			r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
745 				     &nent, cpuid->nent, type);
746 
747 		if (r)
748 			goto out_free;
749 	}
750 
751 	r = -EFAULT;
752 	if (copy_to_user(entries, cpuid_entries,
753 			 nent * sizeof(struct kvm_cpuid_entry2)))
754 		goto out_free;
755 	cpuid->nent = nent;
756 	r = 0;
757 
758 out_free:
759 	vfree(cpuid_entries);
760 out:
761 	return r;
762 }
763 
764 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
765 {
766 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
767 	int j, nent = vcpu->arch.cpuid_nent;
768 
769 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
770 	/* when no next entry is found, the current entry[i] is reselected */
771 	for (j = i + 1; ; j = (j + 1) % nent) {
772 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
773 		if (ej->function == e->function) {
774 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
775 			return j;
776 		}
777 	}
778 	return 0; /* silence gcc, even though control never reaches here */
779 }
780 
781 /* find an entry with matching function, matching index (if needed), and that
782  * should be read next (if it's stateful) */
783 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
784 	u32 function, u32 index)
785 {
786 	if (e->function != function)
787 		return 0;
788 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
789 		return 0;
790 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
791 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
792 		return 0;
793 	return 1;
794 }
795 
796 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
797 					      u32 function, u32 index)
798 {
799 	int i;
800 	struct kvm_cpuid_entry2 *best = NULL;
801 
802 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
803 		struct kvm_cpuid_entry2 *e;
804 
805 		e = &vcpu->arch.cpuid_entries[i];
806 		if (is_matching_cpuid_entry(e, function, index)) {
807 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
808 				move_to_next_stateful_cpuid_entry(vcpu, i);
809 			best = e;
810 			break;
811 		}
812 	}
813 	return best;
814 }
815 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
816 
817 /*
818  * If no match is found, check whether we exceed the vCPU's limit
819  * and return the content of the highest valid _standard_ leaf instead.
820  * This is to satisfy the CPUID specification.
821  */
822 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
823                                                   u32 function, u32 index)
824 {
825 	struct kvm_cpuid_entry2 *maxlevel;
826 
827 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
828 	if (!maxlevel || maxlevel->eax >= function)
829 		return NULL;
830 	if (function & 0x80000000) {
831 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
832 		if (!maxlevel)
833 			return NULL;
834 	}
835 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
836 }
837 
838 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
839 {
840 	u32 function = *eax, index = *ecx;
841 	struct kvm_cpuid_entry2 *best;
842 
843 	best = kvm_find_cpuid_entry(vcpu, function, index);
844 
845 	if (!best)
846 		best = check_cpuid_limit(vcpu, function, index);
847 
848 	/*
849 	 * Perfmon not yet supported for L2 guest.
850 	 */
851 	if (is_guest_mode(vcpu) && function == 0xa)
852 		best = NULL;
853 
854 	if (best) {
855 		*eax = best->eax;
856 		*ebx = best->ebx;
857 		*ecx = best->ecx;
858 		*edx = best->edx;
859 	} else
860 		*eax = *ebx = *ecx = *edx = 0;
861 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
862 }
863 EXPORT_SYMBOL_GPL(kvm_cpuid);
864 
865 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
866 {
867 	u32 function, eax, ebx, ecx, edx;
868 
869 	function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
870 	ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
871 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
872 	kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
873 	kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
874 	kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
875 	kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
876 	kvm_x86_ops->skip_emulated_instruction(vcpu);
877 }
878 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
879