9bffb5228f31670b790051c8aa329c2692c8ad92
[muen/linux.git] / arch / x86 / kvm / cpuid.c
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 <linux/sched/stat.h>
20
21 #include <asm/processor.h>
22 #include <asm/user.h>
23 #include <asm/fpu/xstate.h>
24 #include "cpuid.h"
25 #include "lapic.h"
26 #include "mmu.h"
27 #include "trace.h"
28 #include "pmu.h"
29
30 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
31 {
32         int feature_bit = 0;
33         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
34
35         xstate_bv &= XFEATURE_MASK_EXTEND;
36         while (xstate_bv) {
37                 if (xstate_bv & 0x1) {
38                         u32 eax, ebx, ecx, edx, offset;
39                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
40                         offset = compacted ? ret : ebx;
41                         ret = max(ret, offset + eax);
42                 }
43
44                 xstate_bv >>= 1;
45                 feature_bit++;
46         }
47
48         return ret;
49 }
50
51 bool kvm_mpx_supported(void)
52 {
53         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
54                  && kvm_x86_ops->mpx_supported());
55 }
56 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
57
58 u64 kvm_supported_xcr0(void)
59 {
60         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
61
62         if (!kvm_mpx_supported())
63                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
64
65         return xcr0;
66 }
67
68 #define F(x) bit(X86_FEATURE_##x)
69
70 /* For scattered features from cpufeatures.h; we currently expose none */
71 #define KF(x) bit(KVM_CPUID_BIT_##x)
72
73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
74 {
75         struct kvm_cpuid_entry2 *best;
76         struct kvm_lapic *apic = vcpu->arch.apic;
77
78         best = kvm_find_cpuid_entry(vcpu, 1, 0);
79         if (!best)
80                 return 0;
81
82         /* Update OSXSAVE bit */
83         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
84                 best->ecx &= ~F(OSXSAVE);
85                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
86                         best->ecx |= F(OSXSAVE);
87         }
88
89         best->edx &= ~F(APIC);
90         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
91                 best->edx |= F(APIC);
92
93         if (apic) {
94                 if (best->ecx & F(TSC_DEADLINE_TIMER))
95                         apic->lapic_timer.timer_mode_mask = 3 << 17;
96                 else
97                         apic->lapic_timer.timer_mode_mask = 1 << 17;
98         }
99
100         best = kvm_find_cpuid_entry(vcpu, 7, 0);
101         if (best) {
102                 /* Update OSPKE bit */
103                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
104                         best->ecx &= ~F(OSPKE);
105                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
106                                 best->ecx |= F(OSPKE);
107                 }
108         }
109
110         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
111         if (!best) {
112                 vcpu->arch.guest_supported_xcr0 = 0;
113                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
114         } else {
115                 vcpu->arch.guest_supported_xcr0 =
116                         (best->eax | ((u64)best->edx << 32)) &
117                         kvm_supported_xcr0();
118                 vcpu->arch.guest_xstate_size = best->ebx =
119                         xstate_required_size(vcpu->arch.xcr0, false);
120         }
121
122         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
123         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
124                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
125
126         /*
127          * The existing code assumes virtual address is 48-bit or 57-bit in the
128          * canonical address checks; exit if it is ever changed.
129          */
130         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
131         if (best) {
132                 int vaddr_bits = (best->eax & 0xff00) >> 8;
133
134                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
135                         return -EINVAL;
136         }
137
138         best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
139         if (kvm_hlt_in_guest(vcpu->kvm) && best &&
140                 (best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
141                 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
142
143         /* Update physical-address width */
144         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
145         kvm_mmu_reset_context(vcpu);
146
147         kvm_pmu_refresh(vcpu);
148         return 0;
149 }
150
151 static int is_efer_nx(void)
152 {
153         unsigned long long efer = 0;
154
155         rdmsrl_safe(MSR_EFER, &efer);
156         return efer & EFER_NX;
157 }
158
159 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
160 {
161         int i;
162         struct kvm_cpuid_entry2 *e, *entry;
163
164         entry = NULL;
165         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
166                 e = &vcpu->arch.cpuid_entries[i];
167                 if (e->function == 0x80000001) {
168                         entry = e;
169                         break;
170                 }
171         }
172         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
173                 entry->edx &= ~F(NX);
174                 printk(KERN_INFO "kvm: guest NX capability removed\n");
175         }
176 }
177
178 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
179 {
180         struct kvm_cpuid_entry2 *best;
181
182         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
183         if (!best || best->eax < 0x80000008)
184                 goto not_found;
185         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
186         if (best)
187                 return best->eax & 0xff;
188 not_found:
189         return 36;
190 }
191 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
192
193 /* when an old userspace process fills a new kernel module */
194 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
195                              struct kvm_cpuid *cpuid,
196                              struct kvm_cpuid_entry __user *entries)
197 {
198         int r, i;
199         struct kvm_cpuid_entry *cpuid_entries = NULL;
200
201         r = -E2BIG;
202         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
203                 goto out;
204         r = -ENOMEM;
205         if (cpuid->nent) {
206                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
207                                         cpuid->nent);
208                 if (!cpuid_entries)
209                         goto out;
210                 r = -EFAULT;
211                 if (copy_from_user(cpuid_entries, entries,
212                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
213                         goto out;
214         }
215         for (i = 0; i < cpuid->nent; i++) {
216                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
217                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
218                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
219                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
220                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
221                 vcpu->arch.cpuid_entries[i].index = 0;
222                 vcpu->arch.cpuid_entries[i].flags = 0;
223                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
224                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
225                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
226         }
227         vcpu->arch.cpuid_nent = cpuid->nent;
228         cpuid_fix_nx_cap(vcpu);
229         kvm_apic_set_version(vcpu);
230         kvm_x86_ops->cpuid_update(vcpu);
231         r = kvm_update_cpuid(vcpu);
232
233 out:
234         vfree(cpuid_entries);
235         return r;
236 }
237
238 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
239                               struct kvm_cpuid2 *cpuid,
240                               struct kvm_cpuid_entry2 __user *entries)
241 {
242         int r;
243
244         r = -E2BIG;
245         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
246                 goto out;
247         r = -EFAULT;
248         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
249                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
250                 goto out;
251         vcpu->arch.cpuid_nent = cpuid->nent;
252         kvm_apic_set_version(vcpu);
253         kvm_x86_ops->cpuid_update(vcpu);
254         r = kvm_update_cpuid(vcpu);
255 out:
256         return r;
257 }
258
259 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
260                               struct kvm_cpuid2 *cpuid,
261                               struct kvm_cpuid_entry2 __user *entries)
262 {
263         int r;
264
265         r = -E2BIG;
266         if (cpuid->nent < vcpu->arch.cpuid_nent)
267                 goto out;
268         r = -EFAULT;
269         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
270                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
271                 goto out;
272         return 0;
273
274 out:
275         cpuid->nent = vcpu->arch.cpuid_nent;
276         return r;
277 }
278
279 static void cpuid_mask(u32 *word, int wordnum)
280 {
281         *word &= boot_cpu_data.x86_capability[wordnum];
282 }
283
284 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
285                            u32 index)
286 {
287         entry->function = function;
288         entry->index = index;
289         cpuid_count(entry->function, entry->index,
290                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
291         entry->flags = 0;
292 }
293
294 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
295                                    u32 func, u32 index, int *nent, int maxnent)
296 {
297         switch (func) {
298         case 0:
299                 entry->eax = 7;
300                 ++*nent;
301                 break;
302         case 1:
303                 entry->ecx = F(MOVBE);
304                 ++*nent;
305                 break;
306         case 7:
307                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
308                 if (index == 0)
309                         entry->ecx = F(RDPID);
310                 ++*nent;
311         default:
312                 break;
313         }
314
315         entry->function = func;
316         entry->index = index;
317
318         return 0;
319 }
320
321 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
322                                  u32 index, int *nent, int maxnent)
323 {
324         int r;
325         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
326 #ifdef CONFIG_X86_64
327         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
328                                 ? F(GBPAGES) : 0;
329         unsigned f_lm = F(LM);
330 #else
331         unsigned f_gbpages = 0;
332         unsigned f_lm = 0;
333 #endif
334         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
335         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
336         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
337         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
338         unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
339
340         /* cpuid 1.edx */
341         const u32 kvm_cpuid_1_edx_x86_features =
342                 F(FPU) | F(VME) | F(DE) | F(PSE) |
343                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
344                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
345                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
346                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
347                 0 /* Reserved, DS, ACPI */ | F(MMX) |
348                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
349                 0 /* HTT, TM, Reserved, PBE */;
350         /* cpuid 0x80000001.edx */
351         const u32 kvm_cpuid_8000_0001_edx_x86_features =
352                 F(FPU) | F(VME) | F(DE) | F(PSE) |
353                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
354                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
355                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
356                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
357                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
358                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
359                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
360         /* cpuid 1.ecx */
361         const u32 kvm_cpuid_1_ecx_x86_features =
362                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
363                  * but *not* advertised to guests via CPUID ! */
364                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
365                 0 /* DS-CPL, VMX, SMX, EST */ |
366                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
367                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
368                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
369                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
370                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
371                 F(F16C) | F(RDRAND);
372         /* cpuid 0x80000001.ecx */
373         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
374                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
375                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
376                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
377                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
378                 F(TOPOEXT) | F(PERFCTR_CORE);
379
380         /* cpuid 0x80000008.ebx */
381         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
382                 F(AMD_IBPB) | F(AMD_IBRS) | F(VIRT_SSBD);
383
384         /* cpuid 0xC0000001.edx */
385         const u32 kvm_cpuid_C000_0001_edx_x86_features =
386                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
387                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
388                 F(PMM) | F(PMM_EN);
389
390         /* cpuid 7.0.ebx */
391         const u32 kvm_cpuid_7_0_ebx_x86_features =
392                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
393                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
394                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
395                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
396                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
397
398         /* cpuid 0xD.1.eax */
399         const u32 kvm_cpuid_D_1_eax_x86_features =
400                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
401
402         /* cpuid 7.0.ecx*/
403         const u32 kvm_cpuid_7_0_ecx_x86_features =
404                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
405                 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
406                 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG);
407
408         /* cpuid 7.0.edx*/
409         const u32 kvm_cpuid_7_0_edx_x86_features =
410                 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) | F(SSBD) |
411                 F(ARCH_CAPABILITIES);
412
413         /* all calls to cpuid_count() should be made on the same cpu */
414         get_cpu();
415
416         r = -E2BIG;
417
418         if (*nent >= maxnent)
419                 goto out;
420
421         do_cpuid_1_ent(entry, function, index);
422         ++*nent;
423
424         switch (function) {
425         case 0:
426                 entry->eax = min(entry->eax, (u32)0xd);
427                 break;
428         case 1:
429                 entry->edx &= kvm_cpuid_1_edx_x86_features;
430                 cpuid_mask(&entry->edx, CPUID_1_EDX);
431                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
432                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
433                 /* we support x2apic emulation even if host does not support
434                  * it since we emulate x2apic in software */
435                 entry->ecx |= F(X2APIC);
436                 break;
437         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
438          * may return different values. This forces us to get_cpu() before
439          * issuing the first command, and also to emulate this annoying behavior
440          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
441         case 2: {
442                 int t, times = entry->eax & 0xff;
443
444                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
445                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
446                 for (t = 1; t < times; ++t) {
447                         if (*nent >= maxnent)
448                                 goto out;
449
450                         do_cpuid_1_ent(&entry[t], function, 0);
451                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
452                         ++*nent;
453                 }
454                 break;
455         }
456         /* function 4 has additional index. */
457         case 4: {
458                 int i, cache_type;
459
460                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
461                 /* read more entries until cache_type is zero */
462                 for (i = 1; ; ++i) {
463                         if (*nent >= maxnent)
464                                 goto out;
465
466                         cache_type = entry[i - 1].eax & 0x1f;
467                         if (!cache_type)
468                                 break;
469                         do_cpuid_1_ent(&entry[i], function, i);
470                         entry[i].flags |=
471                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
472                         ++*nent;
473                 }
474                 break;
475         }
476         case 6: /* Thermal management */
477                 entry->eax = 0x4; /* allow ARAT */
478                 entry->ebx = 0;
479                 entry->ecx = 0;
480                 entry->edx = 0;
481                 break;
482         case 7: {
483                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
484                 /* Mask ebx against host capability word 9 */
485                 if (index == 0) {
486                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
487                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
488                         // TSC_ADJUST is emulated
489                         entry->ebx |= F(TSC_ADJUST);
490                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
491                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
492                         entry->ecx |= f_umip;
493                         /* PKU is not yet implemented for shadow paging. */
494                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
495                                 entry->ecx &= ~F(PKU);
496                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
497                         cpuid_mask(&entry->edx, CPUID_7_EDX);
498                         /*
499                          * We emulate ARCH_CAPABILITIES in software even
500                          * if the host doesn't support it.
501                          */
502                         entry->edx |= F(ARCH_CAPABILITIES);
503                 } else {
504                         entry->ebx = 0;
505                         entry->ecx = 0;
506                         entry->edx = 0;
507                 }
508                 entry->eax = 0;
509                 break;
510         }
511         case 9:
512                 break;
513         case 0xa: { /* Architectural Performance Monitoring */
514                 struct x86_pmu_capability cap;
515                 union cpuid10_eax eax;
516                 union cpuid10_edx edx;
517
518                 perf_get_x86_pmu_capability(&cap);
519
520                 /*
521                  * Only support guest architectural pmu on a host
522                  * with architectural pmu.
523                  */
524                 if (!cap.version)
525                         memset(&cap, 0, sizeof(cap));
526
527                 eax.split.version_id = min(cap.version, 2);
528                 eax.split.num_counters = cap.num_counters_gp;
529                 eax.split.bit_width = cap.bit_width_gp;
530                 eax.split.mask_length = cap.events_mask_len;
531
532                 edx.split.num_counters_fixed = cap.num_counters_fixed;
533                 edx.split.bit_width_fixed = cap.bit_width_fixed;
534                 edx.split.reserved = 0;
535
536                 entry->eax = eax.full;
537                 entry->ebx = cap.events_mask;
538                 entry->ecx = 0;
539                 entry->edx = edx.full;
540                 break;
541         }
542         /* function 0xb has additional index. */
543         case 0xb: {
544                 int i, level_type;
545
546                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
547                 /* read more entries until level_type is zero */
548                 for (i = 1; ; ++i) {
549                         if (*nent >= maxnent)
550                                 goto out;
551
552                         level_type = entry[i - 1].ecx & 0xff00;
553                         if (!level_type)
554                                 break;
555                         do_cpuid_1_ent(&entry[i], function, i);
556                         entry[i].flags |=
557                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
558                         ++*nent;
559                 }
560                 break;
561         }
562         case 0xd: {
563                 int idx, i;
564                 u64 supported = kvm_supported_xcr0();
565
566                 entry->eax &= supported;
567                 entry->ebx = xstate_required_size(supported, false);
568                 entry->ecx = entry->ebx;
569                 entry->edx &= supported >> 32;
570                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
571                 if (!supported)
572                         break;
573
574                 for (idx = 1, i = 1; idx < 64; ++idx) {
575                         u64 mask = ((u64)1 << idx);
576                         if (*nent >= maxnent)
577                                 goto out;
578
579                         do_cpuid_1_ent(&entry[i], function, idx);
580                         if (idx == 1) {
581                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
582                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
583                                 entry[i].ebx = 0;
584                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
585                                         entry[i].ebx =
586                                                 xstate_required_size(supported,
587                                                                      true);
588                         } else {
589                                 if (entry[i].eax == 0 || !(supported & mask))
590                                         continue;
591                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
592                                         continue;
593                         }
594                         entry[i].ecx = 0;
595                         entry[i].edx = 0;
596                         entry[i].flags |=
597                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
598                         ++*nent;
599                         ++i;
600                 }
601                 break;
602         }
603         case KVM_CPUID_SIGNATURE: {
604                 static const char signature[12] = "KVMKVMKVM\0\0";
605                 const u32 *sigptr = (const u32 *)signature;
606                 entry->eax = KVM_CPUID_FEATURES;
607                 entry->ebx = sigptr[0];
608                 entry->ecx = sigptr[1];
609                 entry->edx = sigptr[2];
610                 break;
611         }
612         case KVM_CPUID_FEATURES:
613                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
614                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
615                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
616                              (1 << KVM_FEATURE_ASYNC_PF) |
617                              (1 << KVM_FEATURE_PV_EOI) |
618                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
619                              (1 << KVM_FEATURE_PV_UNHALT) |
620                              (1 << KVM_FEATURE_PV_TLB_FLUSH) |
621                              (1 << KVM_FEATURE_ASYNC_PF_VMEXIT);
622
623                 if (sched_info_on())
624                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
625
626                 entry->ebx = 0;
627                 entry->ecx = 0;
628                 entry->edx = 0;
629                 break;
630         case 0x80000000:
631                 entry->eax = min(entry->eax, 0x8000001f);
632                 break;
633         case 0x80000001:
634                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
635                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
636                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
637                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
638                 break;
639         case 0x80000007: /* Advanced power management */
640                 /* invariant TSC is CPUID.80000007H:EDX[8] */
641                 entry->edx &= (1 << 8);
642                 /* mask against host */
643                 entry->edx &= boot_cpu_data.x86_power;
644                 entry->eax = entry->ebx = entry->ecx = 0;
645                 break;
646         case 0x80000008: {
647                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
648                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
649                 unsigned phys_as = entry->eax & 0xff;
650
651                 if (!g_phys_as)
652                         g_phys_as = phys_as;
653                 entry->eax = g_phys_as | (virt_as << 8);
654                 entry->edx = 0;
655                 /*
656                  * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
657                  * hardware cpuid
658                  */
659                 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
660                         entry->ebx |= F(AMD_IBPB);
661                 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
662                         entry->ebx |= F(AMD_IBRS);
663                 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
664                         entry->ebx |= F(VIRT_SSBD);
665                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
666                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
667                 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
668                         entry->ebx |= F(VIRT_SSBD);
669                 break;
670         }
671         case 0x80000019:
672                 entry->ecx = entry->edx = 0;
673                 break;
674         case 0x8000001a:
675                 break;
676         case 0x8000001d:
677                 break;
678         /*Add support for Centaur's CPUID instruction*/
679         case 0xC0000000:
680                 /*Just support up to 0xC0000004 now*/
681                 entry->eax = min(entry->eax, 0xC0000004);
682                 break;
683         case 0xC0000001:
684                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
685                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
686                 break;
687         case 3: /* Processor serial number */
688         case 5: /* MONITOR/MWAIT */
689         case 0xC0000002:
690         case 0xC0000003:
691         case 0xC0000004:
692         default:
693                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
694                 break;
695         }
696
697         kvm_x86_ops->set_supported_cpuid(function, entry);
698
699         r = 0;
700
701 out:
702         put_cpu();
703
704         return r;
705 }
706
707 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
708                         u32 idx, int *nent, int maxnent, unsigned int type)
709 {
710         if (type == KVM_GET_EMULATED_CPUID)
711                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
712
713         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
714 }
715
716 #undef F
717
718 struct kvm_cpuid_param {
719         u32 func;
720         u32 idx;
721         bool has_leaf_count;
722         bool (*qualifier)(const struct kvm_cpuid_param *param);
723 };
724
725 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
726 {
727         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
728 }
729
730 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
731                                  __u32 num_entries, unsigned int ioctl_type)
732 {
733         int i;
734         __u32 pad[3];
735
736         if (ioctl_type != KVM_GET_EMULATED_CPUID)
737                 return false;
738
739         /*
740          * We want to make sure that ->padding is being passed clean from
741          * userspace in case we want to use it for something in the future.
742          *
743          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
744          * have to give ourselves satisfied only with the emulated side. /me
745          * sheds a tear.
746          */
747         for (i = 0; i < num_entries; i++) {
748                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
749                         return true;
750
751                 if (pad[0] || pad[1] || pad[2])
752                         return true;
753         }
754         return false;
755 }
756
757 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
758                             struct kvm_cpuid_entry2 __user *entries,
759                             unsigned int type)
760 {
761         struct kvm_cpuid_entry2 *cpuid_entries;
762         int limit, nent = 0, r = -E2BIG, i;
763         u32 func;
764         static const struct kvm_cpuid_param param[] = {
765                 { .func = 0, .has_leaf_count = true },
766                 { .func = 0x80000000, .has_leaf_count = true },
767                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
768                 { .func = KVM_CPUID_SIGNATURE },
769                 { .func = KVM_CPUID_FEATURES },
770         };
771
772         if (cpuid->nent < 1)
773                 goto out;
774         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
775                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
776
777         if (sanity_check_entries(entries, cpuid->nent, type))
778                 return -EINVAL;
779
780         r = -ENOMEM;
781         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
782         if (!cpuid_entries)
783                 goto out;
784
785         r = 0;
786         for (i = 0; i < ARRAY_SIZE(param); i++) {
787                 const struct kvm_cpuid_param *ent = &param[i];
788
789                 if (ent->qualifier && !ent->qualifier(ent))
790                         continue;
791
792                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
793                                 &nent, cpuid->nent, type);
794
795                 if (r)
796                         goto out_free;
797
798                 if (!ent->has_leaf_count)
799                         continue;
800
801                 limit = cpuid_entries[nent - 1].eax;
802                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
803                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
804                                      &nent, cpuid->nent, type);
805
806                 if (r)
807                         goto out_free;
808         }
809
810         r = -EFAULT;
811         if (copy_to_user(entries, cpuid_entries,
812                          nent * sizeof(struct kvm_cpuid_entry2)))
813                 goto out_free;
814         cpuid->nent = nent;
815         r = 0;
816
817 out_free:
818         vfree(cpuid_entries);
819 out:
820         return r;
821 }
822
823 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
824 {
825         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
826         struct kvm_cpuid_entry2 *ej;
827         int j = i;
828         int nent = vcpu->arch.cpuid_nent;
829
830         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
831         /* when no next entry is found, the current entry[i] is reselected */
832         do {
833                 j = (j + 1) % nent;
834                 ej = &vcpu->arch.cpuid_entries[j];
835         } while (ej->function != e->function);
836
837         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
838
839         return j;
840 }
841
842 /* find an entry with matching function, matching index (if needed), and that
843  * should be read next (if it's stateful) */
844 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
845         u32 function, u32 index)
846 {
847         if (e->function != function)
848                 return 0;
849         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
850                 return 0;
851         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
852             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
853                 return 0;
854         return 1;
855 }
856
857 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
858                                               u32 function, u32 index)
859 {
860         int i;
861         struct kvm_cpuid_entry2 *best = NULL;
862
863         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
864                 struct kvm_cpuid_entry2 *e;
865
866                 e = &vcpu->arch.cpuid_entries[i];
867                 if (is_matching_cpuid_entry(e, function, index)) {
868                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
869                                 move_to_next_stateful_cpuid_entry(vcpu, i);
870                         best = e;
871                         break;
872                 }
873         }
874         return best;
875 }
876 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
877
878 /*
879  * If no match is found, check whether we exceed the vCPU's limit
880  * and return the content of the highest valid _standard_ leaf instead.
881  * This is to satisfy the CPUID specification.
882  */
883 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
884                                                   u32 function, u32 index)
885 {
886         struct kvm_cpuid_entry2 *maxlevel;
887
888         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
889         if (!maxlevel || maxlevel->eax >= function)
890                 return NULL;
891         if (function & 0x80000000) {
892                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
893                 if (!maxlevel)
894                         return NULL;
895         }
896         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
897 }
898
899 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
900                u32 *ecx, u32 *edx, bool check_limit)
901 {
902         u32 function = *eax, index = *ecx;
903         struct kvm_cpuid_entry2 *best;
904         bool entry_found = true;
905
906         best = kvm_find_cpuid_entry(vcpu, function, index);
907
908         if (!best) {
909                 entry_found = false;
910                 if (!check_limit)
911                         goto out;
912
913                 best = check_cpuid_limit(vcpu, function, index);
914         }
915
916 out:
917         if (best) {
918                 *eax = best->eax;
919                 *ebx = best->ebx;
920                 *ecx = best->ecx;
921                 *edx = best->edx;
922         } else
923                 *eax = *ebx = *ecx = *edx = 0;
924         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
925         return entry_found;
926 }
927 EXPORT_SYMBOL_GPL(kvm_cpuid);
928
929 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
930 {
931         u32 eax, ebx, ecx, edx;
932
933         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
934                 return 1;
935
936         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
937         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
938         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
939         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
940         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
941         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
942         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
943         return kvm_skip_emulated_instruction(vcpu);
944 }
945 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);