?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.9.199
[linux/fpc-iii.git] / arch / x86 / kvm / cpuid.c
blobfc8236fd249500ec2aaa75827d8af83935a880cb
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/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 if (apic) {
84 if (best->ecx & F(TSC_DEADLINE_TIMER))
85 apic->lapic_timer.timer_mode_mask = 3 << 17;
86 else
87 apic->lapic_timer.timer_mode_mask = 1 << 17;
88 }
89
90 best = kvm_find_cpuid_entry(vcpu, 7, 0);
91 if (best) {
92 /* Update OSPKE bit */
93 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
94 best->ecx &= ~F(OSPKE);
95 if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
96 best->ecx |= F(OSPKE);
97 }
98 }
99
100 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
101 if (!best) {
102 vcpu->arch.guest_supported_xcr0 = 0;
103 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
104 } else {
105 vcpu->arch.guest_supported_xcr0 =
106 (best->eax | ((u64)best->edx << 32)) &
107 kvm_supported_xcr0();
108 vcpu->arch.guest_xstate_size = best->ebx =
109 xstate_required_size(vcpu->arch.xcr0, false);
110 }
111
112 best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
113 if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
114 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
115
116 kvm_x86_ops->fpu_activate(vcpu);
117
118 /*
119 * The existing code assumes virtual address is 48-bit in the canonical
120 * address checks; exit if it is ever changed.
121 */
122 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
123 if (best && ((best->eax & 0xff00) >> 8) != 48 &&
124 ((best->eax & 0xff00) >> 8) != 0)
125 return -EINVAL;
126
127 /* Update physical-address width */
128 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
129
130 kvm_pmu_refresh(vcpu);
131 return 0;
132 }
133
134 static int is_efer_nx(void)
135 {
136 unsigned long long efer = 0;
137
138 rdmsrl_safe(MSR_EFER, &efer);
139 return efer & EFER_NX;
140 }
141
142 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
143 {
144 int i;
145 struct kvm_cpuid_entry2 *e, *entry;
146
147 entry = NULL;
148 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
149 e = &vcpu->arch.cpuid_entries[i];
150 if (e->function == 0x80000001) {
151 entry = e;
152 break;
153 }
154 }
155 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
156 entry->edx &= ~F(NX);
157 printk(KERN_INFO "kvm: guest NX capability removed\n");
158 }
159 }
160
161 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
162 {
163 struct kvm_cpuid_entry2 *best;
164
165 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
166 if (!best || best->eax < 0x80000008)
167 goto not_found;
168 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
169 if (best)
170 return best->eax & 0xff;
171 not_found:
172 return 36;
173 }
174 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
175
176 /* when an old userspace process fills a new kernel module */
177 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
178 struct kvm_cpuid *cpuid,
179 struct kvm_cpuid_entry __user *entries)
180 {
181 int r, i;
182 struct kvm_cpuid_entry *cpuid_entries = NULL;
183
184 r = -E2BIG;
185 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
186 goto out;
187 r = -ENOMEM;
188 if (cpuid->nent) {
189 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
190 cpuid->nent);
191 if (!cpuid_entries)
192 goto out;
193 r = -EFAULT;
194 if (copy_from_user(cpuid_entries, entries,
195 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
196 goto out;
197 }
198 for (i = 0; i < cpuid->nent; i++) {
199 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
200 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
201 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
202 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
203 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
204 vcpu->arch.cpuid_entries[i].index = 0;
205 vcpu->arch.cpuid_entries[i].flags = 0;
206 vcpu->arch.cpuid_entries[i].padding[0] = 0;
207 vcpu->arch.cpuid_entries[i].padding[1] = 0;
208 vcpu->arch.cpuid_entries[i].padding[2] = 0;
209 }
210 vcpu->arch.cpuid_nent = cpuid->nent;
211 cpuid_fix_nx_cap(vcpu);
212 kvm_apic_set_version(vcpu);
213 kvm_x86_ops->cpuid_update(vcpu);
214 r = kvm_update_cpuid(vcpu);
215
216 out:
217 vfree(cpuid_entries);
218 return r;
219 }
220
221 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
222 struct kvm_cpuid2 *cpuid,
223 struct kvm_cpuid_entry2 __user *entries)
224 {
225 int r;
226
227 r = -E2BIG;
228 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
229 goto out;
230 r = -EFAULT;
231 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
232 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
233 goto out;
234 vcpu->arch.cpuid_nent = cpuid->nent;
235 kvm_apic_set_version(vcpu);
236 kvm_x86_ops->cpuid_update(vcpu);
237 r = kvm_update_cpuid(vcpu);
238 out:
239 return r;
240 }
241
242 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
243 struct kvm_cpuid2 *cpuid,
244 struct kvm_cpuid_entry2 __user *entries)
245 {
246 int r;
247
248 r = -E2BIG;
249 if (cpuid->nent < vcpu->arch.cpuid_nent)
250 goto out;
251 r = -EFAULT;
252 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
253 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
254 goto out;
255 return 0;
256
257 out:
258 cpuid->nent = vcpu->arch.cpuid_nent;
259 return r;
260 }
261
262 static void cpuid_mask(u32 *word, int wordnum)
263 {
264 *word &= boot_cpu_data.x86_capability[wordnum];
265 }
266
267 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
268 u32 index)
269 {
270 entry->function = function;
271 entry->index = index;
272 cpuid_count(entry->function, entry->index,
273 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
274 entry->flags = 0;
275 }
276
277 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
278 u32 func, u32 index, int *nent, int maxnent)
279 {
280 switch (func) {
281 case 0:
282 entry->eax = 1; /* only one leaf currently */
283 ++*nent;
284 break;
285 case 1:
286 entry->ecx = F(MOVBE);
287 ++*nent;
288 break;
289 default:
290 break;
291 }
292
293 entry->function = func;
294 entry->index = index;
295
296 return 0;
297 }
298
299 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
300 u32 index, int *nent, int maxnent)
301 {
302 int r;
303 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
304 #ifdef CONFIG_X86_64
305 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
306 ? F(GBPAGES) : 0;
307 unsigned f_lm = F(LM);
308 #else
309 unsigned f_gbpages = 0;
310 unsigned f_lm = 0;
311 #endif
312 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
313 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
314 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
315 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
316
317 /* cpuid 1.edx */
318 const u32 kvm_cpuid_1_edx_x86_features =
319 F(FPU) | F(VME) | F(DE) | F(PSE) |
320 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
321 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
322 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
323 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
324 0 /* Reserved, DS, ACPI */ | F(MMX) |
325 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
326 0 /* HTT, TM, Reserved, PBE */;
327 /* cpuid 0x80000001.edx */
328 const u32 kvm_cpuid_8000_0001_edx_x86_features =
329 F(FPU) | F(VME) | F(DE) | F(PSE) |
330 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
331 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
332 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
333 F(PAT) | F(PSE36) | 0 /* Reserved */ |
334 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
335 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
336 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
337 /* cpuid 1.ecx */
338 const u32 kvm_cpuid_1_ecx_x86_features =
339 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
340 * but *not* advertised to guests via CPUID ! */
341 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
342 0 /* DS-CPL, VMX, SMX, EST */ |
343 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
344 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
345 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
346 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
347 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
348 F(F16C) | F(RDRAND);
349 /* cpuid 0x80000001.ecx */
350 const u32 kvm_cpuid_8000_0001_ecx_x86_features =
351 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
352 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
353 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
354 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
355
356 /* cpuid 0x80000008.ebx */
357 const u32 kvm_cpuid_8000_0008_ebx_x86_features =
358 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
359 F(AMD_SSB_NO) | F(AMD_STIBP);
360
361 /* cpuid 0xC0000001.edx */
362 const u32 kvm_cpuid_C000_0001_edx_x86_features =
363 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
364 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
365 F(PMM) | F(PMM_EN);
366
367 /* cpuid 7.0.ebx */
368 const u32 kvm_cpuid_7_0_ebx_x86_features =
369 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
370 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
371 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
372 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
373 F(AVX512BW) | F(AVX512VL);
374
375 /* cpuid 0xD.1.eax */
376 const u32 kvm_cpuid_D_1_eax_x86_features =
377 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
378
379 /* cpuid 7.0.ecx*/
380 const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
381
382 /* cpuid 7.0.edx*/
383 const u32 kvm_cpuid_7_0_edx_x86_features =
384 F(SPEC_CTRL) | F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) |
385 F(INTEL_STIBP) | F(MD_CLEAR);
386
387 /* all calls to cpuid_count() should be made on the same cpu */
388 get_cpu();
389
390 r = -E2BIG;
391
392 if (*nent >= maxnent)
393 goto out;
394
395 do_cpuid_1_ent(entry, function, index);
396 ++*nent;
397
398 switch (function) {
399 case 0:
400 entry->eax = min(entry->eax, (u32)0xd);
401 break;
402 case 1:
403 entry->edx &= kvm_cpuid_1_edx_x86_features;
404 cpuid_mask(&entry->edx, CPUID_1_EDX);
405 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
406 cpuid_mask(&entry->ecx, CPUID_1_ECX);
407 /* we support x2apic emulation even if host does not support
408 * it since we emulate x2apic in software */
409 entry->ecx |= F(X2APIC);
410 break;
411 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
412 * may return different values. This forces us to get_cpu() before
413 * issuing the first command, and also to emulate this annoying behavior
414 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
415 case 2: {
416 int t, times = entry->eax & 0xff;
417
418 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
419 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
420 for (t = 1; t < times; ++t) {
421 if (*nent >= maxnent)
422 goto out;
423
424 do_cpuid_1_ent(&entry[t], function, 0);
425 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
426 ++*nent;
427 }
428 break;
429 }
430 /* function 4 has additional index. */
431 case 4: {
432 int i, cache_type;
433
434 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
435 /* read more entries until cache_type is zero */
436 for (i = 1; ; ++i) {
437 if (*nent >= maxnent)
438 goto out;
439
440 cache_type = entry[i - 1].eax & 0x1f;
441 if (!cache_type)
442 break;
443 do_cpuid_1_ent(&entry[i], function, i);
444 entry[i].flags |=
445 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
446 ++*nent;
447 }
448 break;
449 }
450 case 6: /* Thermal management */
451 entry->eax = 0x4; /* allow ARAT */
452 entry->ebx = 0;
453 entry->ecx = 0;
454 entry->edx = 0;
455 break;
456 case 7: {
457 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
458 /* Mask ebx against host capability word 9 */
459 if (index == 0) {
460 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
461 cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
462 // TSC_ADJUST is emulated
463 entry->ebx |= F(TSC_ADJUST);
464 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
465 cpuid_mask(&entry->ecx, CPUID_7_ECX);
466 /* PKU is not yet implemented for shadow paging. */
467 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
468 entry->ecx &= ~F(PKU);
469 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
470 cpuid_mask(&entry->edx, CPUID_7_EDX);
471 /*
472 * We emulate ARCH_CAPABILITIES in software even
473 * if the host doesn't support it.
474 */
475 entry->edx |= F(ARCH_CAPABILITIES);
476 } else {
477 entry->ebx = 0;
478 entry->ecx = 0;
479 entry->edx = 0;
480 }
481 entry->eax = 0;
482 break;
483 }
484 case 9:
485 break;
486 case 0xa: { /* Architectural Performance Monitoring */
487 struct x86_pmu_capability cap;
488 union cpuid10_eax eax;
489 union cpuid10_edx edx;
490
491 perf_get_x86_pmu_capability(&cap);
492
493 /*
494 * Only support guest architectural pmu on a host
495 * with architectural pmu.
496 */
497 if (!cap.version)
498 memset(&cap, 0, sizeof(cap));
499
500 eax.split.version_id = min(cap.version, 2);
501 eax.split.num_counters = cap.num_counters_gp;
502 eax.split.bit_width = cap.bit_width_gp;
503 eax.split.mask_length = cap.events_mask_len;
504
505 edx.split.num_counters_fixed = cap.num_counters_fixed;
506 edx.split.bit_width_fixed = cap.bit_width_fixed;
507 edx.split.reserved = 0;
508
509 entry->eax = eax.full;
510 entry->ebx = cap.events_mask;
511 entry->ecx = 0;
512 entry->edx = edx.full;
513 break;
514 }
515 /* function 0xb has additional index. */
516 case 0xb: {
517 int i, level_type;
518
519 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
520 /* read more entries until level_type is zero */
521 for (i = 1; ; ++i) {
522 if (*nent >= maxnent)
523 goto out;
524
525 level_type = entry[i - 1].ecx & 0xff00;
526 if (!level_type)
527 break;
528 do_cpuid_1_ent(&entry[i], function, i);
529 entry[i].flags |=
530 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
531 ++*nent;
532 }
533 break;
534 }
535 case 0xd: {
536 int idx, i;
537 u64 supported = kvm_supported_xcr0();
538
539 entry->eax &= supported;
540 entry->ebx = xstate_required_size(supported, false);
541 entry->ecx = entry->ebx;
542 entry->edx &= supported >> 32;
543 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
544 if (!supported)
545 break;
546
547 for (idx = 1, i = 1; idx < 64; ++idx) {
548 u64 mask = ((u64)1 << idx);
549 if (*nent >= maxnent)
550 goto out;
551
552 do_cpuid_1_ent(&entry[i], function, idx);
553 if (idx == 1) {
554 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
555 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
556 entry[i].ebx = 0;
557 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
558 entry[i].ebx =
559 xstate_required_size(supported,
560 true);
561 } else {
562 if (entry[i].eax == 0 || !(supported & mask))
563 continue;
564 if (WARN_ON_ONCE(entry[i].ecx & 1))
565 continue;
566 }
567 entry[i].ecx = 0;
568 entry[i].edx = 0;
569 entry[i].flags |=
570 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
571 ++*nent;
572 ++i;
573 }
574 break;
575 }
576 case KVM_CPUID_SIGNATURE: {
577 static const char signature[12] = "KVMKVMKVM\0\0";
578 const u32 *sigptr = (const u32 *)signature;
579 entry->eax = KVM_CPUID_FEATURES;
580 entry->ebx = sigptr[0];
581 entry->ecx = sigptr[1];
582 entry->edx = sigptr[2];
583 break;
584 }
585 case KVM_CPUID_FEATURES:
586 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
587 (1 << KVM_FEATURE_NOP_IO_DELAY) |
588 (1 << KVM_FEATURE_CLOCKSOURCE2) |
589 (1 << KVM_FEATURE_ASYNC_PF) |
590 (1 << KVM_FEATURE_PV_EOI) |
591 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
592 (1 << KVM_FEATURE_PV_UNHALT);
593
594 if (sched_info_on())
595 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
596
597 entry->ebx = 0;
598 entry->ecx = 0;
599 entry->edx = 0;
600 break;
601 case 0x80000000:
602 entry->eax = min(entry->eax, 0x8000001a);
603 break;
604 case 0x80000001:
605 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
606 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
607 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
608 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
609 break;
610 case 0x80000007: /* Advanced power management */
611 /* invariant TSC is CPUID.80000007H:EDX[8] */
612 entry->edx &= (1 << 8);
613 /* mask against host */
614 entry->edx &= boot_cpu_data.x86_power;
615 entry->eax = entry->ebx = entry->ecx = 0;
616 break;
617 case 0x80000008: {
618 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
619 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
620 unsigned phys_as = entry->eax & 0xff;
621
622 if (!g_phys_as)
623 g_phys_as = phys_as;
624 entry->eax = g_phys_as | (virt_as << 8);
625 entry->edx = 0;
626 /*
627 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
628 * hardware cpuid
629 */
630 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
631 entry->ebx |= F(AMD_IBPB);
632 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
633 entry->ebx |= F(AMD_IBRS);
634 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
635 entry->ebx |= F(VIRT_SSBD);
636 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
637 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
638 /*
639 * The preference is to use SPEC CTRL MSR instead of the
640 * VIRT_SPEC MSR.
641 */
642 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
643 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
644 entry->ebx |= F(VIRT_SSBD);
645 break;
646 }
647 case 0x80000019:
648 entry->ecx = entry->edx = 0;
649 break;
650 case 0x8000001a:
651 break;
652 case 0x8000001d:
653 break;
654 /*Add support for Centaur's CPUID instruction*/
655 case 0xC0000000:
656 /*Just support up to 0xC0000004 now*/
657 entry->eax = min(entry->eax, 0xC0000004);
658 break;
659 case 0xC0000001:
660 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
661 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
662 break;
663 case 3: /* Processor serial number */
664 case 5: /* MONITOR/MWAIT */
665 case 0xC0000002:
666 case 0xC0000003:
667 case 0xC0000004:
668 default:
669 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
670 break;
671 }
672
673 kvm_x86_ops->set_supported_cpuid(function, entry);
674
675 r = 0;
676
677 out:
678 put_cpu();
679
680 return r;
681 }
682
683 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
684 u32 idx, int *nent, int maxnent, unsigned int type)
685 {
686 if (type == KVM_GET_EMULATED_CPUID)
687 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
688
689 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
690 }
691
692 #undef F
693
694 struct kvm_cpuid_param {
695 u32 func;
696 u32 idx;
697 bool has_leaf_count;
698 bool (*qualifier)(const struct kvm_cpuid_param *param);
699 };
700
701 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
702 {
703 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
704 }
705
706 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
707 __u32 num_entries, unsigned int ioctl_type)
708 {
709 int i;
710 __u32 pad[3];
711
712 if (ioctl_type != KVM_GET_EMULATED_CPUID)
713 return false;
714
715 /*
716 * We want to make sure that ->padding is being passed clean from
717 * userspace in case we want to use it for something in the future.
718 *
719 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
720 * have to give ourselves satisfied only with the emulated side. /me
721 * sheds a tear.
722 */
723 for (i = 0; i < num_entries; i++) {
724 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
725 return true;
726
727 if (pad[0] || pad[1] || pad[2])
728 return true;
729 }
730 return false;
731 }
732
733 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
734 struct kvm_cpuid_entry2 __user *entries,
735 unsigned int type)
736 {
737 struct kvm_cpuid_entry2 *cpuid_entries;
738 int limit, nent = 0, r = -E2BIG, i;
739 u32 func;
740 static const struct kvm_cpuid_param param[] = {
741 { .func = 0, .has_leaf_count = true },
742 { .func = 0x80000000, .has_leaf_count = true },
743 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
744 { .func = KVM_CPUID_SIGNATURE },
745 { .func = KVM_CPUID_FEATURES },
746 };
747
748 if (cpuid->nent < 1)
749 goto out;
750 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
751 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
752
753 if (sanity_check_entries(entries, cpuid->nent, type))
754 return -EINVAL;
755
756 r = -ENOMEM;
757 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
758 if (!cpuid_entries)
759 goto out;
760
761 r = 0;
762 for (i = 0; i < ARRAY_SIZE(param); i++) {
763 const struct kvm_cpuid_param *ent = &param[i];
764
765 if (ent->qualifier && !ent->qualifier(ent))
766 continue;
767
768 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
769 &nent, cpuid->nent, type);
770
771 if (r)
772 goto out_free;
773
774 if (!ent->has_leaf_count)
775 continue;
776
777 limit = cpuid_entries[nent - 1].eax;
778 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
779 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
780 &nent, cpuid->nent, type);
781
782 if (r)
783 goto out_free;
784 }
785
786 r = -EFAULT;
787 if (copy_to_user(entries, cpuid_entries,
788 nent * sizeof(struct kvm_cpuid_entry2)))
789 goto out_free;
790 cpuid->nent = nent;
791 r = 0;
792
793 out_free:
794 vfree(cpuid_entries);
795 out:
796 return r;
797 }
798
799 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
800 {
801 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
802 struct kvm_cpuid_entry2 *ej;
803 int j = i;
804 int nent = vcpu->arch.cpuid_nent;
805
806 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
807 /* when no next entry is found, the current entry[i] is reselected */
808 do {
809 j = (j + 1) % nent;
810 ej = &vcpu->arch.cpuid_entries[j];
811 } while (ej->function != e->function);
812
813 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
814
815 return j;
816 }
817
818 /* find an entry with matching function, matching index (if needed), and that
819 * should be read next (if it's stateful) */
820 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
821 u32 function, u32 index)
822 {
823 if (e->function != function)
824 return 0;
825 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
826 return 0;
827 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
828 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
829 return 0;
830 return 1;
831 }
832
833 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
834 u32 function, u32 index)
835 {
836 int i;
837 struct kvm_cpuid_entry2 *best = NULL;
838
839 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
840 struct kvm_cpuid_entry2 *e;
841
842 e = &vcpu->arch.cpuid_entries[i];
843 if (is_matching_cpuid_entry(e, function, index)) {
844 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
845 move_to_next_stateful_cpuid_entry(vcpu, i);
846 best = e;
847 break;
848 }
849 }
850 return best;
851 }
852 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
853
854 /*
855 * If no match is found, check whether we exceed the vCPU's limit
856 * and return the content of the highest valid _standard_ leaf instead.
857 * This is to satisfy the CPUID specification.
858 */
859 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
860 u32 function, u32 index)
861 {
862 struct kvm_cpuid_entry2 *maxlevel;
863
864 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
865 if (!maxlevel || maxlevel->eax >= function)
866 return NULL;
867 if (function & 0x80000000) {
868 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
869 if (!maxlevel)
870 return NULL;
871 }
872 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
873 }
874
875 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
876 {
877 u32 function = *eax, index = *ecx;
878 struct kvm_cpuid_entry2 *best;
879
880 best = kvm_find_cpuid_entry(vcpu, function, index);
881
882 if (!best)
883 best = check_cpuid_limit(vcpu, function, index);
884
885 if (best) {
886 *eax = best->eax;
887 *ebx = best->ebx;
888 *ecx = best->ecx;
889 *edx = best->edx;
890 } else
891 *eax = *ebx = *ecx = *edx = 0;
892 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
893 }
894 EXPORT_SYMBOL_GPL(kvm_cpuid);
895
896 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
897 {
898 u32 function, eax, ebx, ecx, edx;
899
900 function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
901 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
902 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
903 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
904 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
905 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
906 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
907 kvm_x86_ops->skip_emulated_instruction(vcpu);
908 }
909 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
Linux with added FPC-III support