xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 1d1997db)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  * cpuid support routines
5  *
6  * derived from arch/x86/kvm/x86.c
7  *
8  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
9  * Copyright IBM Corporation, 2008
10  */
11 
12 #include <linux/kvm_host.h>
13 #include <linux/export.h>
14 #include <linux/vmalloc.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/stat.h>
17 
18 #include <asm/processor.h>
19 #include <asm/user.h>
20 #include <asm/fpu/xstate.h>
21 #include "cpuid.h"
22 #include "lapic.h"
23 #include "mmu.h"
24 #include "trace.h"
25 #include "pmu.h"
26 
27 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
28 {
29 	int feature_bit = 0;
30 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
31 
32 	xstate_bv &= XFEATURE_MASK_EXTEND;
33 	while (xstate_bv) {
34 		if (xstate_bv & 0x1) {
35 		        u32 eax, ebx, ecx, edx, offset;
36 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
37 			offset = compacted ? ret : ebx;
38 			ret = max(ret, offset + eax);
39 		}
40 
41 		xstate_bv >>= 1;
42 		feature_bit++;
43 	}
44 
45 	return ret;
46 }
47 
48 bool kvm_mpx_supported(void)
49 {
50 	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
51 		 && kvm_x86_ops->mpx_supported());
52 }
53 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
54 
55 u64 kvm_supported_xcr0(void)
56 {
57 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
58 
59 	if (!kvm_mpx_supported())
60 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
61 
62 	return xcr0;
63 }
64 
65 #define F(x) bit(X86_FEATURE_##x)
66 
67 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
68 {
69 	struct kvm_cpuid_entry2 *best;
70 	struct kvm_lapic *apic = vcpu->arch.apic;
71 
72 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
73 	if (!best)
74 		return 0;
75 
76 	/* Update OSXSAVE bit */
77 	if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
78 		best->ecx &= ~F(OSXSAVE);
79 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
80 			best->ecx |= F(OSXSAVE);
81 	}
82 
83 	best->edx &= ~F(APIC);
84 	if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
85 		best->edx |= F(APIC);
86 
87 	if (apic) {
88 		if (best->ecx & F(TSC_DEADLINE_TIMER))
89 			apic->lapic_timer.timer_mode_mask = 3 << 17;
90 		else
91 			apic->lapic_timer.timer_mode_mask = 1 << 17;
92 	}
93 
94 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
95 	if (best) {
96 		/* Update OSPKE bit */
97 		if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
98 			best->ecx &= ~F(OSPKE);
99 			if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
100 				best->ecx |= F(OSPKE);
101 		}
102 	}
103 
104 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
105 	if (!best) {
106 		vcpu->arch.guest_supported_xcr0 = 0;
107 		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
108 	} else {
109 		vcpu->arch.guest_supported_xcr0 =
110 			(best->eax | ((u64)best->edx << 32)) &
111 			kvm_supported_xcr0();
112 		vcpu->arch.guest_xstate_size = best->ebx =
113 			xstate_required_size(vcpu->arch.xcr0, false);
114 	}
115 
116 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
117 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
118 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
119 
120 	/*
121 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
122 	 * canonical address checks; exit if it is ever changed.
123 	 */
124 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
125 	if (best) {
126 		int vaddr_bits = (best->eax & 0xff00) >> 8;
127 
128 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
129 			return -EINVAL;
130 	}
131 
132 	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
133 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
134 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
135 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
136 
137 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
138 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
139 		if (best) {
140 			if (vcpu->arch.ia32_misc_enable_msr & MSR_IA32_MISC_ENABLE_MWAIT)
141 				best->ecx |= F(MWAIT);
142 			else
143 				best->ecx &= ~F(MWAIT);
144 		}
145 	}
146 
147 	/* Update physical-address width */
148 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
149 	kvm_mmu_reset_context(vcpu);
150 
151 	kvm_pmu_refresh(vcpu);
152 	return 0;
153 }
154 
155 static int is_efer_nx(void)
156 {
157 	unsigned long long efer = 0;
158 
159 	rdmsrl_safe(MSR_EFER, &efer);
160 	return efer & EFER_NX;
161 }
162 
163 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
164 {
165 	int i;
166 	struct kvm_cpuid_entry2 *e, *entry;
167 
168 	entry = NULL;
169 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
170 		e = &vcpu->arch.cpuid_entries[i];
171 		if (e->function == 0x80000001) {
172 			entry = e;
173 			break;
174 		}
175 	}
176 	if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
177 		entry->edx &= ~F(NX);
178 		printk(KERN_INFO "kvm: guest NX capability removed\n");
179 	}
180 }
181 
182 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
183 {
184 	struct kvm_cpuid_entry2 *best;
185 
186 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
187 	if (!best || best->eax < 0x80000008)
188 		goto not_found;
189 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
190 	if (best)
191 		return best->eax & 0xff;
192 not_found:
193 	return 36;
194 }
195 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
196 
197 /* when an old userspace process fills a new kernel module */
198 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
199 			     struct kvm_cpuid *cpuid,
200 			     struct kvm_cpuid_entry __user *entries)
201 {
202 	int r, i;
203 	struct kvm_cpuid_entry *cpuid_entries = NULL;
204 
205 	r = -E2BIG;
206 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
207 		goto out;
208 	r = -ENOMEM;
209 	if (cpuid->nent) {
210 		cpuid_entries =
211 			vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
212 					   cpuid->nent));
213 		if (!cpuid_entries)
214 			goto out;
215 		r = -EFAULT;
216 		if (copy_from_user(cpuid_entries, entries,
217 				   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
218 			goto out;
219 	}
220 	for (i = 0; i < cpuid->nent; i++) {
221 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
222 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
223 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
224 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
225 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
226 		vcpu->arch.cpuid_entries[i].index = 0;
227 		vcpu->arch.cpuid_entries[i].flags = 0;
228 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
229 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
230 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
231 	}
232 	vcpu->arch.cpuid_nent = cpuid->nent;
233 	cpuid_fix_nx_cap(vcpu);
234 	kvm_apic_set_version(vcpu);
235 	kvm_x86_ops->cpuid_update(vcpu);
236 	r = kvm_update_cpuid(vcpu);
237 
238 out:
239 	vfree(cpuid_entries);
240 	return r;
241 }
242 
243 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
244 			      struct kvm_cpuid2 *cpuid,
245 			      struct kvm_cpuid_entry2 __user *entries)
246 {
247 	int r;
248 
249 	r = -E2BIG;
250 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
251 		goto out;
252 	r = -EFAULT;
253 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
254 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
255 		goto out;
256 	vcpu->arch.cpuid_nent = cpuid->nent;
257 	kvm_apic_set_version(vcpu);
258 	kvm_x86_ops->cpuid_update(vcpu);
259 	r = kvm_update_cpuid(vcpu);
260 out:
261 	return r;
262 }
263 
264 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
265 			      struct kvm_cpuid2 *cpuid,
266 			      struct kvm_cpuid_entry2 __user *entries)
267 {
268 	int r;
269 
270 	r = -E2BIG;
271 	if (cpuid->nent < vcpu->arch.cpuid_nent)
272 		goto out;
273 	r = -EFAULT;
274 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
275 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
276 		goto out;
277 	return 0;
278 
279 out:
280 	cpuid->nent = vcpu->arch.cpuid_nent;
281 	return r;
282 }
283 
284 static void cpuid_mask(u32 *word, int wordnum)
285 {
286 	*word &= boot_cpu_data.x86_capability[wordnum];
287 }
288 
289 static void do_host_cpuid(struct kvm_cpuid_entry2 *entry, u32 function,
290 			   u32 index)
291 {
292 	entry->function = function;
293 	entry->index = index;
294 	entry->flags = 0;
295 
296 	cpuid_count(entry->function, entry->index,
297 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
298 
299 	switch (function) {
300 	case 2:
301 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
302 		break;
303 	case 4:
304 	case 7:
305 	case 0xb:
306 	case 0xd:
307 	case 0xf:
308 	case 0x10:
309 	case 0x12:
310 	case 0x14:
311 	case 0x17:
312 	case 0x18:
313 	case 0x1f:
314 	case 0x8000001d:
315 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
316 		break;
317 	}
318 }
319 
320 static int __do_cpuid_func_emulated(struct kvm_cpuid_entry2 *entry,
321 				    u32 func, int *nent, int maxnent)
322 {
323 	entry->function = func;
324 	entry->index = 0;
325 	entry->flags = 0;
326 
327 	switch (func) {
328 	case 0:
329 		entry->eax = 7;
330 		++*nent;
331 		break;
332 	case 1:
333 		entry->ecx = F(MOVBE);
334 		++*nent;
335 		break;
336 	case 7:
337 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
338 		entry->eax = 0;
339 		entry->ecx = F(RDPID);
340 		++*nent;
341 	default:
342 		break;
343 	}
344 
345 	return 0;
346 }
347 
348 static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index)
349 {
350 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
351 	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
352 	unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
353 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
354 	unsigned f_la57;
355 
356 	/* cpuid 7.0.ebx */
357 	const u32 kvm_cpuid_7_0_ebx_x86_features =
358 		F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
359 		F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
360 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
361 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
362 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
363 
364 	/* cpuid 7.0.ecx*/
365 	const u32 kvm_cpuid_7_0_ecx_x86_features =
366 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
367 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
368 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
369 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/;
370 
371 	/* cpuid 7.0.edx*/
372 	const u32 kvm_cpuid_7_0_edx_x86_features =
373 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
374 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
375 		F(MD_CLEAR);
376 
377 	/* cpuid 7.1.eax */
378 	const u32 kvm_cpuid_7_1_eax_x86_features =
379 		F(AVX512_BF16);
380 
381 	switch (index) {
382 	case 0:
383 		entry->eax = min(entry->eax, 1u);
384 		entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
385 		cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
386 		/* TSC_ADJUST is emulated */
387 		entry->ebx |= F(TSC_ADJUST);
388 
389 		entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
390 		f_la57 = entry->ecx & F(LA57);
391 		cpuid_mask(&entry->ecx, CPUID_7_ECX);
392 		/* Set LA57 based on hardware capability. */
393 		entry->ecx |= f_la57;
394 		entry->ecx |= f_umip;
395 		/* PKU is not yet implemented for shadow paging. */
396 		if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
397 			entry->ecx &= ~F(PKU);
398 
399 		entry->edx &= kvm_cpuid_7_0_edx_x86_features;
400 		cpuid_mask(&entry->edx, CPUID_7_EDX);
401 		if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
402 			entry->edx |= F(SPEC_CTRL);
403 		if (boot_cpu_has(X86_FEATURE_STIBP))
404 			entry->edx |= F(INTEL_STIBP);
405 		if (boot_cpu_has(X86_FEATURE_SSBD))
406 			entry->edx |= F(SPEC_CTRL_SSBD);
407 		/*
408 		 * We emulate ARCH_CAPABILITIES in software even
409 		 * if the host doesn't support it.
410 		 */
411 		entry->edx |= F(ARCH_CAPABILITIES);
412 		break;
413 	case 1:
414 		entry->eax &= kvm_cpuid_7_1_eax_x86_features;
415 		entry->ebx = 0;
416 		entry->ecx = 0;
417 		entry->edx = 0;
418 		break;
419 	default:
420 		WARN_ON_ONCE(1);
421 		entry->eax = 0;
422 		entry->ebx = 0;
423 		entry->ecx = 0;
424 		entry->edx = 0;
425 		break;
426 	}
427 }
428 
429 static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,
430 				  int *nent, int maxnent)
431 {
432 	int r;
433 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
434 #ifdef CONFIG_X86_64
435 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
436 				? F(GBPAGES) : 0;
437 	unsigned f_lm = F(LM);
438 #else
439 	unsigned f_gbpages = 0;
440 	unsigned f_lm = 0;
441 #endif
442 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
443 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
444 	unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
445 
446 	/* cpuid 1.edx */
447 	const u32 kvm_cpuid_1_edx_x86_features =
448 		F(FPU) | F(VME) | F(DE) | F(PSE) |
449 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
450 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
451 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
452 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
453 		0 /* Reserved, DS, ACPI */ | F(MMX) |
454 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
455 		0 /* HTT, TM, Reserved, PBE */;
456 	/* cpuid 0x80000001.edx */
457 	const u32 kvm_cpuid_8000_0001_edx_x86_features =
458 		F(FPU) | F(VME) | F(DE) | F(PSE) |
459 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
460 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
461 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
462 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
463 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
464 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
465 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
466 	/* cpuid 1.ecx */
467 	const u32 kvm_cpuid_1_ecx_x86_features =
468 		/* NOTE: MONITOR (and MWAIT) are emulated as NOP,
469 		 * but *not* advertised to guests via CPUID ! */
470 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
471 		0 /* DS-CPL, VMX, SMX, EST */ |
472 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
473 		F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
474 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
475 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
476 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
477 		F(F16C) | F(RDRAND);
478 	/* cpuid 0x80000001.ecx */
479 	const u32 kvm_cpuid_8000_0001_ecx_x86_features =
480 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
481 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
482 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
483 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
484 		F(TOPOEXT) | F(PERFCTR_CORE);
485 
486 	/* cpuid 0x80000008.ebx */
487 	const u32 kvm_cpuid_8000_0008_ebx_x86_features =
488 		F(CLZERO) | F(XSAVEERPTR) |
489 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
490 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON);
491 
492 	/* cpuid 0xC0000001.edx */
493 	const u32 kvm_cpuid_C000_0001_edx_x86_features =
494 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
495 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
496 		F(PMM) | F(PMM_EN);
497 
498 	/* cpuid 0xD.1.eax */
499 	const u32 kvm_cpuid_D_1_eax_x86_features =
500 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
501 
502 	/* all calls to cpuid_count() should be made on the same cpu */
503 	get_cpu();
504 
505 	r = -E2BIG;
506 
507 	if (WARN_ON(*nent >= maxnent))
508 		goto out;
509 
510 	do_host_cpuid(entry, function, 0);
511 	++*nent;
512 
513 	switch (function) {
514 	case 0:
515 		/* Limited to the highest leaf implemented in KVM. */
516 		entry->eax = min(entry->eax, 0x1fU);
517 		break;
518 	case 1:
519 		entry->edx &= kvm_cpuid_1_edx_x86_features;
520 		cpuid_mask(&entry->edx, CPUID_1_EDX);
521 		entry->ecx &= kvm_cpuid_1_ecx_x86_features;
522 		cpuid_mask(&entry->ecx, CPUID_1_ECX);
523 		/* we support x2apic emulation even if host does not support
524 		 * it since we emulate x2apic in software */
525 		entry->ecx |= F(X2APIC);
526 		break;
527 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
528 	 * may return different values. This forces us to get_cpu() before
529 	 * issuing the first command, and also to emulate this annoying behavior
530 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
531 	case 2: {
532 		int t, times = entry->eax & 0xff;
533 
534 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
535 		for (t = 1; t < times; ++t) {
536 			if (*nent >= maxnent)
537 				goto out;
538 
539 			do_host_cpuid(&entry[t], function, 0);
540 			++*nent;
541 		}
542 		break;
543 	}
544 	/* functions 4 and 0x8000001d have additional index. */
545 	case 4:
546 	case 0x8000001d: {
547 		int i, cache_type;
548 
549 		/* read more entries until cache_type is zero */
550 		for (i = 1; ; ++i) {
551 			if (*nent >= maxnent)
552 				goto out;
553 
554 			cache_type = entry[i - 1].eax & 0x1f;
555 			if (!cache_type)
556 				break;
557 			do_host_cpuid(&entry[i], function, i);
558 			++*nent;
559 		}
560 		break;
561 	}
562 	case 6: /* Thermal management */
563 		entry->eax = 0x4; /* allow ARAT */
564 		entry->ebx = 0;
565 		entry->ecx = 0;
566 		entry->edx = 0;
567 		break;
568 	/* function 7 has additional index. */
569 	case 7: {
570 		int i;
571 
572 		for (i = 0; ; ) {
573 			do_cpuid_7_mask(&entry[i], i);
574 			if (i == entry->eax)
575 				break;
576 			if (*nent >= maxnent)
577 				goto out;
578 
579 			++i;
580 			do_host_cpuid(&entry[i], function, i);
581 			++*nent;
582 		}
583 		break;
584 	}
585 	case 9:
586 		break;
587 	case 0xa: { /* Architectural Performance Monitoring */
588 		struct x86_pmu_capability cap;
589 		union cpuid10_eax eax;
590 		union cpuid10_edx edx;
591 
592 		perf_get_x86_pmu_capability(&cap);
593 
594 		/*
595 		 * Only support guest architectural pmu on a host
596 		 * with architectural pmu.
597 		 */
598 		if (!cap.version)
599 			memset(&cap, 0, sizeof(cap));
600 
601 		eax.split.version_id = min(cap.version, 2);
602 		eax.split.num_counters = cap.num_counters_gp;
603 		eax.split.bit_width = cap.bit_width_gp;
604 		eax.split.mask_length = cap.events_mask_len;
605 
606 		edx.split.num_counters_fixed = cap.num_counters_fixed;
607 		edx.split.bit_width_fixed = cap.bit_width_fixed;
608 		edx.split.reserved = 0;
609 
610 		entry->eax = eax.full;
611 		entry->ebx = cap.events_mask;
612 		entry->ecx = 0;
613 		entry->edx = edx.full;
614 		break;
615 	}
616 	/*
617 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
618 	 * thus they can be handled by common code.
619 	 */
620 	case 0x1f:
621 	case 0xb: {
622 		int i;
623 
624 		/*
625 		 * We filled in entry[0] for CPUID(EAX=<function>,
626 		 * ECX=00H) above.  If its level type (ECX[15:8]) is
627 		 * zero, then the leaf is unimplemented, and we're
628 		 * done.  Otherwise, continue to populate entries
629 		 * until the level type (ECX[15:8]) of the previously
630 		 * added entry is zero.
631 		 */
632 		for (i = 1; entry[i - 1].ecx & 0xff00; ++i) {
633 			if (*nent >= maxnent)
634 				goto out;
635 
636 			do_host_cpuid(&entry[i], function, i);
637 			++*nent;
638 		}
639 		break;
640 	}
641 	case 0xd: {
642 		int idx, i;
643 		u64 supported = kvm_supported_xcr0();
644 
645 		entry->eax &= supported;
646 		entry->ebx = xstate_required_size(supported, false);
647 		entry->ecx = entry->ebx;
648 		entry->edx &= supported >> 32;
649 		if (!supported)
650 			break;
651 
652 		for (idx = 1, i = 1; idx < 64; ++idx) {
653 			u64 mask = ((u64)1 << idx);
654 			if (*nent >= maxnent)
655 				goto out;
656 
657 			do_host_cpuid(&entry[i], function, idx);
658 			if (idx == 1) {
659 				entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
660 				cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
661 				entry[i].ebx = 0;
662 				if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
663 					entry[i].ebx =
664 						xstate_required_size(supported,
665 								     true);
666 			} else {
667 				if (entry[i].eax == 0 || !(supported & mask))
668 					continue;
669 				if (WARN_ON_ONCE(entry[i].ecx & 1))
670 					continue;
671 			}
672 			entry[i].ecx = 0;
673 			entry[i].edx = 0;
674 			++*nent;
675 			++i;
676 		}
677 		break;
678 	}
679 	/* Intel PT */
680 	case 0x14: {
681 		int t, times = entry->eax;
682 
683 		if (!f_intel_pt)
684 			break;
685 
686 		for (t = 1; t <= times; ++t) {
687 			if (*nent >= maxnent)
688 				goto out;
689 			do_host_cpuid(&entry[t], function, t);
690 			++*nent;
691 		}
692 		break;
693 	}
694 	case KVM_CPUID_SIGNATURE: {
695 		static const char signature[12] = "KVMKVMKVM\0\0";
696 		const u32 *sigptr = (const u32 *)signature;
697 		entry->eax = KVM_CPUID_FEATURES;
698 		entry->ebx = sigptr[0];
699 		entry->ecx = sigptr[1];
700 		entry->edx = sigptr[2];
701 		break;
702 	}
703 	case KVM_CPUID_FEATURES:
704 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
705 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
706 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
707 			     (1 << KVM_FEATURE_ASYNC_PF) |
708 			     (1 << KVM_FEATURE_PV_EOI) |
709 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
710 			     (1 << KVM_FEATURE_PV_UNHALT) |
711 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
712 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
713 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
714 			     (1 << KVM_FEATURE_POLL_CONTROL) |
715 			     (1 << KVM_FEATURE_PV_SCHED_YIELD);
716 
717 		if (sched_info_on())
718 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
719 
720 		entry->ebx = 0;
721 		entry->ecx = 0;
722 		entry->edx = 0;
723 		break;
724 	case 0x80000000:
725 		entry->eax = min(entry->eax, 0x8000001f);
726 		break;
727 	case 0x80000001:
728 		entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
729 		cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
730 		entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
731 		cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
732 		break;
733 	case 0x80000007: /* Advanced power management */
734 		/* invariant TSC is CPUID.80000007H:EDX[8] */
735 		entry->edx &= (1 << 8);
736 		/* mask against host */
737 		entry->edx &= boot_cpu_data.x86_power;
738 		entry->eax = entry->ebx = entry->ecx = 0;
739 		break;
740 	case 0x80000008: {
741 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
742 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
743 		unsigned phys_as = entry->eax & 0xff;
744 
745 		if (!g_phys_as)
746 			g_phys_as = phys_as;
747 		entry->eax = g_phys_as | (virt_as << 8);
748 		entry->edx = 0;
749 		entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
750 		cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
751 		/*
752 		 * AMD has separate bits for each SPEC_CTRL bit.
753 		 * arch/x86/kernel/cpu/bugs.c is kind enough to
754 		 * record that in cpufeatures so use them.
755 		 */
756 		if (boot_cpu_has(X86_FEATURE_IBPB))
757 			entry->ebx |= F(AMD_IBPB);
758 		if (boot_cpu_has(X86_FEATURE_IBRS))
759 			entry->ebx |= F(AMD_IBRS);
760 		if (boot_cpu_has(X86_FEATURE_STIBP))
761 			entry->ebx |= F(AMD_STIBP);
762 		if (boot_cpu_has(X86_FEATURE_SSBD))
763 			entry->ebx |= F(AMD_SSBD);
764 		if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
765 			entry->ebx |= F(AMD_SSB_NO);
766 		/*
767 		 * The preference is to use SPEC CTRL MSR instead of the
768 		 * VIRT_SPEC MSR.
769 		 */
770 		if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
771 		    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
772 			entry->ebx |= F(VIRT_SSBD);
773 		break;
774 	}
775 	case 0x80000019:
776 		entry->ecx = entry->edx = 0;
777 		break;
778 	case 0x8000001a:
779 	case 0x8000001e:
780 		break;
781 	/* Support memory encryption cpuid if host supports it */
782 	case 0x8000001F:
783 		if (!boot_cpu_has(X86_FEATURE_SEV))
784 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
785 		break;
786 	/*Add support for Centaur's CPUID instruction*/
787 	case 0xC0000000:
788 		/*Just support up to 0xC0000004 now*/
789 		entry->eax = min(entry->eax, 0xC0000004);
790 		break;
791 	case 0xC0000001:
792 		entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
793 		cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
794 		break;
795 	case 3: /* Processor serial number */
796 	case 5: /* MONITOR/MWAIT */
797 	case 0xC0000002:
798 	case 0xC0000003:
799 	case 0xC0000004:
800 	default:
801 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
802 		break;
803 	}
804 
805 	kvm_x86_ops->set_supported_cpuid(function, entry);
806 
807 	r = 0;
808 
809 out:
810 	put_cpu();
811 
812 	return r;
813 }
814 
815 static int do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 func,
816 			 int *nent, int maxnent, unsigned int type)
817 {
818 	if (*nent >= maxnent)
819 		return -E2BIG;
820 
821 	if (type == KVM_GET_EMULATED_CPUID)
822 		return __do_cpuid_func_emulated(entry, func, nent, maxnent);
823 
824 	return __do_cpuid_func(entry, func, nent, maxnent);
825 }
826 
827 struct kvm_cpuid_param {
828 	u32 func;
829 	bool (*qualifier)(const struct kvm_cpuid_param *param);
830 };
831 
832 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
833 {
834 	return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
835 }
836 
837 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
838 				 __u32 num_entries, unsigned int ioctl_type)
839 {
840 	int i;
841 	__u32 pad[3];
842 
843 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
844 		return false;
845 
846 	/*
847 	 * We want to make sure that ->padding is being passed clean from
848 	 * userspace in case we want to use it for something in the future.
849 	 *
850 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
851 	 * have to give ourselves satisfied only with the emulated side. /me
852 	 * sheds a tear.
853 	 */
854 	for (i = 0; i < num_entries; i++) {
855 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
856 			return true;
857 
858 		if (pad[0] || pad[1] || pad[2])
859 			return true;
860 	}
861 	return false;
862 }
863 
864 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
865 			    struct kvm_cpuid_entry2 __user *entries,
866 			    unsigned int type)
867 {
868 	struct kvm_cpuid_entry2 *cpuid_entries;
869 	int limit, nent = 0, r = -E2BIG, i;
870 	u32 func;
871 	static const struct kvm_cpuid_param param[] = {
872 		{ .func = 0 },
873 		{ .func = 0x80000000 },
874 		{ .func = 0xC0000000, .qualifier = is_centaur_cpu },
875 		{ .func = KVM_CPUID_SIGNATURE },
876 	};
877 
878 	if (cpuid->nent < 1)
879 		goto out;
880 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
881 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
882 
883 	if (sanity_check_entries(entries, cpuid->nent, type))
884 		return -EINVAL;
885 
886 	r = -ENOMEM;
887 	cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
888 					   cpuid->nent));
889 	if (!cpuid_entries)
890 		goto out;
891 
892 	r = 0;
893 	for (i = 0; i < ARRAY_SIZE(param); i++) {
894 		const struct kvm_cpuid_param *ent = &param[i];
895 
896 		if (ent->qualifier && !ent->qualifier(ent))
897 			continue;
898 
899 		r = do_cpuid_func(&cpuid_entries[nent], ent->func,
900 				  &nent, cpuid->nent, type);
901 
902 		if (r)
903 			goto out_free;
904 
905 		limit = cpuid_entries[nent - 1].eax;
906 		for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
907 			r = do_cpuid_func(&cpuid_entries[nent], func,
908 				          &nent, cpuid->nent, type);
909 
910 		if (r)
911 			goto out_free;
912 	}
913 
914 	r = -EFAULT;
915 	if (copy_to_user(entries, cpuid_entries,
916 			 nent * sizeof(struct kvm_cpuid_entry2)))
917 		goto out_free;
918 	cpuid->nent = nent;
919 	r = 0;
920 
921 out_free:
922 	vfree(cpuid_entries);
923 out:
924 	return r;
925 }
926 
927 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
928 {
929 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
930 	struct kvm_cpuid_entry2 *ej;
931 	int j = i;
932 	int nent = vcpu->arch.cpuid_nent;
933 
934 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
935 	/* when no next entry is found, the current entry[i] is reselected */
936 	do {
937 		j = (j + 1) % nent;
938 		ej = &vcpu->arch.cpuid_entries[j];
939 	} while (ej->function != e->function);
940 
941 	ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
942 
943 	return j;
944 }
945 
946 /* find an entry with matching function, matching index (if needed), and that
947  * should be read next (if it's stateful) */
948 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
949 	u32 function, u32 index)
950 {
951 	if (e->function != function)
952 		return 0;
953 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
954 		return 0;
955 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
956 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
957 		return 0;
958 	return 1;
959 }
960 
961 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
962 					      u32 function, u32 index)
963 {
964 	int i;
965 	struct kvm_cpuid_entry2 *best = NULL;
966 
967 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
968 		struct kvm_cpuid_entry2 *e;
969 
970 		e = &vcpu->arch.cpuid_entries[i];
971 		if (is_matching_cpuid_entry(e, function, index)) {
972 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
973 				move_to_next_stateful_cpuid_entry(vcpu, i);
974 			best = e;
975 			break;
976 		}
977 	}
978 	return best;
979 }
980 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
981 
982 /*
983  * If the basic or extended CPUID leaf requested is higher than the
984  * maximum supported basic or extended leaf, respectively, then it is
985  * out of range.
986  */
987 static bool cpuid_function_in_range(struct kvm_vcpu *vcpu, u32 function)
988 {
989 	struct kvm_cpuid_entry2 *max;
990 
991 	max = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
992 	return max && function <= max->eax;
993 }
994 
995 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
996 	       u32 *ecx, u32 *edx, bool check_limit)
997 {
998 	u32 function = *eax, index = *ecx;
999 	struct kvm_cpuid_entry2 *entry;
1000 	struct kvm_cpuid_entry2 *max;
1001 	bool found;
1002 
1003 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1004 	found = entry;
1005 	/*
1006 	 * Intel CPUID semantics treats any query for an out-of-range
1007 	 * leaf as if the highest basic leaf (i.e. CPUID.0H:EAX) were
1008 	 * requested. AMD CPUID semantics returns all zeroes for any
1009 	 * undefined leaf, whether or not the leaf is in range.
1010 	 */
1011 	if (!entry && check_limit && !guest_cpuid_is_amd(vcpu) &&
1012 	    !cpuid_function_in_range(vcpu, function)) {
1013 		max = kvm_find_cpuid_entry(vcpu, 0, 0);
1014 		if (max) {
1015 			function = max->eax;
1016 			entry = kvm_find_cpuid_entry(vcpu, function, index);
1017 		}
1018 	}
1019 	if (entry) {
1020 		*eax = entry->eax;
1021 		*ebx = entry->ebx;
1022 		*ecx = entry->ecx;
1023 		*edx = entry->edx;
1024 		if (function == 7 && index == 0) {
1025 			u64 data;
1026 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1027 			    (data & TSX_CTRL_CPUID_CLEAR))
1028 				*ebx &= ~(F(RTM) | F(HLE));
1029 		}
1030 	} else {
1031 		*eax = *ebx = *ecx = *edx = 0;
1032 		/*
1033 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1034 		 * and EDX is always the x2APIC ID, even for undefined
1035 		 * subleaves. Index 1 will exist iff the leaf is
1036 		 * implemented, so we pass through CL iff leaf 1
1037 		 * exists. EDX can be copied from any existing index.
1038 		 */
1039 		if (function == 0xb || function == 0x1f) {
1040 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1041 			if (entry) {
1042 				*ecx = index & 0xff;
1043 				*edx = entry->edx;
1044 			}
1045 		}
1046 	}
1047 	trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, found);
1048 	return found;
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_cpuid);
1051 
1052 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1053 {
1054 	u32 eax, ebx, ecx, edx;
1055 
1056 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1057 		return 1;
1058 
1059 	eax = kvm_rax_read(vcpu);
1060 	ecx = kvm_rcx_read(vcpu);
1061 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
1062 	kvm_rax_write(vcpu, eax);
1063 	kvm_rbx_write(vcpu, ebx);
1064 	kvm_rcx_write(vcpu, ecx);
1065 	kvm_rdx_write(vcpu, edx);
1066 	return kvm_skip_emulated_instruction(vcpu);
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1069