]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kvm/cpuid.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:


14f5cf5c7bf4ea8f70c52773cad70cfe1586a29c
[mirror_ubuntu-artful-kernel.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 /* These are scattered features in cpufeatures.h. */
71 #define KVM_CPUID_BIT_AVX512_4VNNIW 2
72 #define KVM_CPUID_BIT_AVX512_4FMAPS 3
73 #define KVM_CPUID_BIT_SPEC_CTRL 26
74 #define KVM_CPUID_BIT_SSBD 31
75 #define KF(x) bit(KVM_CPUID_BIT_##x)
76
77 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
78 {
79 struct kvm_cpuid_entry2 *best;
80 struct kvm_lapic *apic = vcpu->arch.apic;
81
82 best = kvm_find_cpuid_entry(vcpu, 1, 0);
83 if (!best)
84 return 0;
85
86 /* Update OSXSAVE bit */
87 if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
88 best->ecx &= ~F(OSXSAVE);
89 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
90 best->ecx |= F(OSXSAVE);
91 }
92
93 best->edx &= ~F(APIC);
94 if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
95 best->edx |= F(APIC);
96
97 if (apic) {
98 if (best->ecx & F(TSC_DEADLINE_TIMER))
99 apic->lapic_timer.timer_mode_mask = 3 << 17;
100 else
101 apic->lapic_timer.timer_mode_mask = 1 << 17;
102 }
103
104 best = kvm_find_cpuid_entry(vcpu, 7, 0);
105 if (best) {
106 /* Update OSPKE bit */
107 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
108 best->ecx &= ~F(OSPKE);
109 if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
110 best->ecx |= F(OSPKE);
111 }
112 }
113
114 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
115 if (!best) {
116 vcpu->arch.guest_supported_xcr0 = 0;
117 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
118 } else {
119 vcpu->arch.guest_supported_xcr0 =
120 (best->eax | ((u64)best->edx << 32)) &
121 kvm_supported_xcr0();
122 vcpu->arch.guest_xstate_size = best->ebx =
123 xstate_required_size(vcpu->arch.xcr0, false);
124 }
125
126 best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
127 if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
128 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
129
130 /*
131 * The existing code assumes virtual address is 48-bit in the canonical
132 * address checks; exit if it is ever changed.
133 */
134 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
135 if (best && ((best->eax & 0xff00) >> 8) != 48 &&
136 ((best->eax & 0xff00) >> 8) != 0)
137 return -EINVAL;
138
139 /* Update physical-address width */
140 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
141
142 kvm_pmu_refresh(vcpu);
143 return 0;
144 }
145
146 static int is_efer_nx(void)
147 {
148 unsigned long long efer = 0;
149
150 rdmsrl_safe(MSR_EFER, &efer);
151 return efer & EFER_NX;
152 }
153
154 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
155 {
156 int i;
157 struct kvm_cpuid_entry2 *e, *entry;
158
159 entry = NULL;
160 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
161 e = &vcpu->arch.cpuid_entries[i];
162 if (e->function == 0x80000001) {
163 entry = e;
164 break;
165 }
166 }
167 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
168 entry->edx &= ~F(NX);
169 printk(KERN_INFO "kvm: guest NX capability removed\n");
170 }
171 }
172
173 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
174 {
175 struct kvm_cpuid_entry2 *best;
176
177 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
178 if (!best || best->eax < 0x80000008)
179 goto not_found;
180 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
181 if (best)
182 return best->eax & 0xff;
183 not_found:
184 return 36;
185 }
186 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
187
188 /* when an old userspace process fills a new kernel module */
189 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
190 struct kvm_cpuid *cpuid,
191 struct kvm_cpuid_entry __user *entries)
192 {
193 int r, i;
194 struct kvm_cpuid_entry *cpuid_entries = NULL;
195
196 r = -E2BIG;
197 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
198 goto out;
199 r = -ENOMEM;
200 if (cpuid->nent) {
201 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
202 cpuid->nent);
203 if (!cpuid_entries)
204 goto out;
205 r = -EFAULT;
206 if (copy_from_user(cpuid_entries, entries,
207 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
208 goto out;
209 }
210 for (i = 0; i < cpuid->nent; i++) {
211 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
212 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
213 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
214 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
215 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
216 vcpu->arch.cpuid_entries[i].index = 0;
217 vcpu->arch.cpuid_entries[i].flags = 0;
218 vcpu->arch.cpuid_entries[i].padding[0] = 0;
219 vcpu->arch.cpuid_entries[i].padding[1] = 0;
220 vcpu->arch.cpuid_entries[i].padding[2] = 0;
221 }
222 vcpu->arch.cpuid_nent = cpuid->nent;
223 cpuid_fix_nx_cap(vcpu);
224 kvm_apic_set_version(vcpu);
225 kvm_x86_ops->cpuid_update(vcpu);
226 r = kvm_update_cpuid(vcpu);
227
228 out:
229 vfree(cpuid_entries);
230 return r;
231 }
232
233 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
234 struct kvm_cpuid2 *cpuid,
235 struct kvm_cpuid_entry2 __user *entries)
236 {
237 int r;
238
239 r = -E2BIG;
240 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
241 goto out;
242 r = -EFAULT;
243 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
244 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
245 goto out;
246 vcpu->arch.cpuid_nent = cpuid->nent;
247 kvm_apic_set_version(vcpu);
248 kvm_x86_ops->cpuid_update(vcpu);
249 r = kvm_update_cpuid(vcpu);
250 out:
251 return r;
252 }
253
254 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
255 struct kvm_cpuid2 *cpuid,
256 struct kvm_cpuid_entry2 __user *entries)
257 {
258 int r;
259
260 r = -E2BIG;
261 if (cpuid->nent < vcpu->arch.cpuid_nent)
262 goto out;
263 r = -EFAULT;
264 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
265 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
266 goto out;
267 return 0;
268
269 out:
270 cpuid->nent = vcpu->arch.cpuid_nent;
271 return r;
272 }
273
274 static void cpuid_mask(u32 *word, int wordnum)
275 {
276 *word &= boot_cpu_data.x86_capability[wordnum];
277 }
278
279 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
280 u32 index)
281 {
282 entry->function = function;
283 entry->index = index;
284 cpuid_count(entry->function, entry->index,
285 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
286 entry->flags = 0;
287 }
288
289 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
290 u32 func, u32 index, int *nent, int maxnent)
291 {
292 switch (func) {
293 case 0:
294 entry->eax = 1; /* only one leaf currently */
295 ++*nent;
296 break;
297 case 1:
298 entry->ecx = F(MOVBE);
299 ++*nent;
300 break;
301 default:
302 break;
303 }
304
305 entry->function = func;
306 entry->index = index;
307
308 return 0;
309 }
310
311 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
312 u32 index, int *nent, int maxnent)
313 {
314 int r;
315 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
316 #ifdef CONFIG_X86_64
317 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
318 ? F(GBPAGES) : 0;
319 unsigned f_lm = F(LM);
320 #else
321 unsigned f_gbpages = 0;
322 unsigned f_lm = 0;
323 #endif
324 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
325 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
326 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
327 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
328
329 /* cpuid 1.edx */
330 const u32 kvm_cpuid_1_edx_x86_features =
331 F(FPU) | F(VME) | F(DE) | F(PSE) |
332 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
333 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
334 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
335 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
336 0 /* Reserved, DS, ACPI */ | F(MMX) |
337 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
338 0 /* HTT, TM, Reserved, PBE */;
339 /* cpuid 0x80000001.edx */
340 const u32 kvm_cpuid_8000_0001_edx_x86_features =
341 F(FPU) | F(VME) | F(DE) | F(PSE) |
342 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
343 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
344 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
345 F(PAT) | F(PSE36) | 0 /* Reserved */ |
346 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
347 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
348 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
349 /* cpuid 1.ecx */
350 const u32 kvm_cpuid_1_ecx_x86_features =
351 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
352 * but *not* advertised to guests via CPUID ! */
353 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
354 0 /* DS-CPL, VMX, SMX, EST */ |
355 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
356 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
357 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
358 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
359 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
360 F(F16C) | F(RDRAND);
361 /* cpuid 0x80000001.ecx */
362 const u32 kvm_cpuid_8000_0001_ecx_x86_features =
363 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
364 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
365 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
366 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
367
368 /* cpuid 0xC0000001.edx */
369 const u32 kvm_cpuid_C000_0001_edx_x86_features =
370 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
371 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
372 F(PMM) | F(PMM_EN);
373
374 /* cpuid 7.0.ebx */
375 const u32 kvm_cpuid_7_0_ebx_x86_features =
376 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
377 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
378 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
379 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
380 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
381
382 /* cpuid 0xD.1.eax */
383 const u32 kvm_cpuid_D_1_eax_x86_features =
384 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
385
386 /* cpuid 7.0.ecx*/
387 const u32 kvm_cpuid_7_0_ecx_x86_features =
388 F(AVX512VBMI) | F(PKU) | 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ);
389
390 /* cpuid 7.0.edx*/
391 const u32 kvm_cpuid_7_0_edx_x86_features =
392 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS) |
393 KF(SPEC_CTRL) | KF(SSBD);
394
395 /* cpuid 0x80000008.0.ebx */
396 const u32 kvm_cpuid_80000008_0_ebx_x86_features =
397 F(AMD_IBPB);
398
399 /* all calls to cpuid_count() should be made on the same cpu */
400 get_cpu();
401
402 r = -E2BIG;
403
404 if (*nent >= maxnent)
405 goto out;
406
407 do_cpuid_1_ent(entry, function, index);
408 ++*nent;
409
410 switch (function) {
411 case 0:
412 entry->eax = min(entry->eax, (u32)0xd);
413 break;
414 case 1:
415 entry->edx &= kvm_cpuid_1_edx_x86_features;
416 cpuid_mask(&entry->edx, CPUID_1_EDX);
417 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
418 cpuid_mask(&entry->ecx, CPUID_1_ECX);
419 /* we support x2apic emulation even if host does not support
420 * it since we emulate x2apic in software */
421 entry->ecx |= F(X2APIC);
422 break;
423 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
424 * may return different values. This forces us to get_cpu() before
425 * issuing the first command, and also to emulate this annoying behavior
426 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
427 case 2: {
428 int t, times = entry->eax & 0xff;
429
430 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
431 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
432 for (t = 1; t < times; ++t) {
433 if (*nent >= maxnent)
434 goto out;
435
436 do_cpuid_1_ent(&entry[t], function, 0);
437 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
438 ++*nent;
439 }
440 break;
441 }
442 /* function 4 has additional index. */
443 case 4: {
444 int i, cache_type;
445
446 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
447 /* read more entries until cache_type is zero */
448 for (i = 1; ; ++i) {
449 if (*nent >= maxnent)
450 goto out;
451
452 cache_type = entry[i - 1].eax & 0x1f;
453 if (!cache_type)
454 break;
455 do_cpuid_1_ent(&entry[i], function, i);
456 entry[i].flags |=
457 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
458 ++*nent;
459 }
460 break;
461 }
462 case 6: /* Thermal management */
463 entry->eax = 0x4; /* allow ARAT */
464 entry->ebx = 0;
465 entry->ecx = 0;
466 entry->edx = 0;
467 break;
468 case 7: {
469 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
470 /* Mask ebx against host capability word 9 */
471 if (index == 0) {
472 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
473 cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
474 // TSC_ADJUST is emulated
475 entry->ebx |= F(TSC_ADJUST);
476 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
477 cpuid_mask(&entry->ecx, CPUID_7_ECX);
478 /* PKU is not yet implemented for shadow paging. */
479 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
480 entry->ecx &= ~F(PKU);
481 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
482 entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
483 } else {
484 entry->ebx = 0;
485 entry->ecx = 0;
486 entry->edx = 0;
487 }
488 entry->eax = 0;
489 break;
490 }
491 case 9:
492 break;
493 case 0xa: { /* Architectural Performance Monitoring */
494 struct x86_pmu_capability cap;
495 union cpuid10_eax eax;
496 union cpuid10_edx edx;
497
498 perf_get_x86_pmu_capability(&cap);
499
500 /*
501 * Only support guest architectural pmu on a host
502 * with architectural pmu.
503 */
504 if (!cap.version)
505 memset(&cap, 0, sizeof(cap));
506
507 eax.split.version_id = min(cap.version, 2);
508 eax.split.num_counters = cap.num_counters_gp;
509 eax.split.bit_width = cap.bit_width_gp;
510 eax.split.mask_length = cap.events_mask_len;
511
512 edx.split.num_counters_fixed = cap.num_counters_fixed;
513 edx.split.bit_width_fixed = cap.bit_width_fixed;
514 edx.split.reserved = 0;
515
516 entry->eax = eax.full;
517 entry->ebx = cap.events_mask;
518 entry->ecx = 0;
519 entry->edx = edx.full;
520 break;
521 }
522 /* function 0xb has additional index. */
523 case 0xb: {
524 int i, level_type;
525
526 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
527 /* read more entries until level_type is zero */
528 for (i = 1; ; ++i) {
529 if (*nent >= maxnent)
530 goto out;
531
532 level_type = entry[i - 1].ecx & 0xff00;
533 if (!level_type)
534 break;
535 do_cpuid_1_ent(&entry[i], function, i);
536 entry[i].flags |=
537 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
538 ++*nent;
539 }
540 break;
541 }
542 case 0xd: {
543 int idx, i;
544 u64 supported = kvm_supported_xcr0();
545
546 entry->eax &= supported;
547 entry->ebx = xstate_required_size(supported, false);
548 entry->ecx = entry->ebx;
549 entry->edx &= supported >> 32;
550 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
551 if (!supported)
552 break;
553
554 for (idx = 1, i = 1; idx < 64; ++idx) {
555 u64 mask = ((u64)1 << idx);
556 if (*nent >= maxnent)
557 goto out;
558
559 do_cpuid_1_ent(&entry[i], function, idx);
560 if (idx == 1) {
561 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
562 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
563 entry[i].ebx = 0;
564 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
565 entry[i].ebx =
566 xstate_required_size(supported,
567 true);
568 } else {
569 if (entry[i].eax == 0 || !(supported & mask))
570 continue;
571 if (WARN_ON_ONCE(entry[i].ecx & 1))
572 continue;
573 }
574 entry[i].ecx = 0;
575 entry[i].edx = 0;
576 entry[i].flags |=
577 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
578 ++*nent;
579 ++i;
580 }
581 break;
582 }
583 case KVM_CPUID_SIGNATURE: {
584 static const char signature[12] = "KVMKVMKVM\0\0";
585 const u32 *sigptr = (const u32 *)signature;
586 entry->eax = KVM_CPUID_FEATURES;
587 entry->ebx = sigptr[0];
588 entry->ecx = sigptr[1];
589 entry->edx = sigptr[2];
590 break;
591 }
592 case KVM_CPUID_FEATURES:
593 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
594 (1 << KVM_FEATURE_NOP_IO_DELAY) |
595 (1 << KVM_FEATURE_CLOCKSOURCE2) |
596 (1 << KVM_FEATURE_ASYNC_PF) |
597 (1 << KVM_FEATURE_PV_EOI) |
598 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
599 (1 << KVM_FEATURE_PV_UNHALT);
600
601 if (sched_info_on())
602 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
603
604 entry->ebx = 0;
605 entry->ecx = 0;
606 entry->edx = 0;
607 break;
608 case 0x80000000:
609 entry->eax = min(entry->eax, 0x8000001a);
610 break;
611 case 0x80000001:
612 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
613 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
614 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
615 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
616 break;
617 case 0x80000007: /* Advanced power management */
618 /* invariant TSC is CPUID.80000007H:EDX[8] */
619 entry->edx &= (1 << 8);
620 /* mask against host */
621 entry->edx &= boot_cpu_data.x86_power;
622 entry->eax = entry->ebx = entry->ecx = 0;
623 break;
624 case 0x80000008: {
625 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
626 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
627 unsigned phys_as = entry->eax & 0xff;
628
629 if (!g_phys_as)
630 g_phys_as = phys_as;
631 entry->eax = g_phys_as | (virt_as << 8);
632 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
633 entry->ebx |= F(AMD_IBPB);
634 entry->ebx &= kvm_cpuid_80000008_0_ebx_x86_features;
635 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
636 entry->edx = 0;
637 break;
638 }
639 case 0x80000019:
640 entry->ecx = entry->edx = 0;
641 break;
642 case 0x8000001a:
643 break;
644 case 0x8000001d:
645 break;
646 /*Add support for Centaur's CPUID instruction*/
647 case 0xC0000000:
648 /*Just support up to 0xC0000004 now*/
649 entry->eax = min(entry->eax, 0xC0000004);
650 break;
651 case 0xC0000001:
652 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
653 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
654 break;
655 case 3: /* Processor serial number */
656 case 5: /* MONITOR/MWAIT */
657 case 0xC0000002:
658 case 0xC0000003:
659 case 0xC0000004:
660 default:
661 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
662 break;
663 }
664
665 kvm_x86_ops->set_supported_cpuid(function, entry);
666
667 r = 0;
668
669 out:
670 put_cpu();
671
672 return r;
673 }
674
675 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
676 u32 idx, int *nent, int maxnent, unsigned int type)
677 {
678 if (type == KVM_GET_EMULATED_CPUID)
679 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
680
681 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
682 }
683
684 #undef F
685
686 struct kvm_cpuid_param {
687 u32 func;
688 u32 idx;
689 bool has_leaf_count;
690 bool (*qualifier)(const struct kvm_cpuid_param *param);
691 };
692
693 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
694 {
695 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
696 }
697
698 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
699 __u32 num_entries, unsigned int ioctl_type)
700 {
701 int i;
702 __u32 pad[3];
703
704 if (ioctl_type != KVM_GET_EMULATED_CPUID)
705 return false;
706
707 /*
708 * We want to make sure that ->padding is being passed clean from
709 * userspace in case we want to use it for something in the future.
710 *
711 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
712 * have to give ourselves satisfied only with the emulated side. /me
713 * sheds a tear.
714 */
715 for (i = 0; i < num_entries; i++) {
716 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
717 return true;
718
719 if (pad[0] || pad[1] || pad[2])
720 return true;
721 }
722 return false;
723 }
724
725 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
726 struct kvm_cpuid_entry2 __user *entries,
727 unsigned int type)
728 {
729 struct kvm_cpuid_entry2 *cpuid_entries;
730 int limit, nent = 0, r = -E2BIG, i;
731 u32 func;
732 static const struct kvm_cpuid_param param[] = {
733 { .func = 0, .has_leaf_count = true },
734 { .func = 0x80000000, .has_leaf_count = true },
735 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
736 { .func = KVM_CPUID_SIGNATURE },
737 { .func = KVM_CPUID_FEATURES },
738 };
739
740 if (cpuid->nent < 1)
741 goto out;
742 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
743 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
744
745 if (sanity_check_entries(entries, cpuid->nent, type))
746 return -EINVAL;
747
748 r = -ENOMEM;
749 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
750 if (!cpuid_entries)
751 goto out;
752
753 r = 0;
754 for (i = 0; i < ARRAY_SIZE(param); i++) {
755 const struct kvm_cpuid_param *ent = &param[i];
756
757 if (ent->qualifier && !ent->qualifier(ent))
758 continue;
759
760 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
761 &nent, cpuid->nent, type);
762
763 if (r)
764 goto out_free;
765
766 if (!ent->has_leaf_count)
767 continue;
768
769 limit = cpuid_entries[nent - 1].eax;
770 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
771 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
772 &nent, cpuid->nent, type);
773
774 if (r)
775 goto out_free;
776 }
777
778 r = -EFAULT;
779 if (copy_to_user(entries, cpuid_entries,
780 nent * sizeof(struct kvm_cpuid_entry2)))
781 goto out_free;
782 cpuid->nent = nent;
783 r = 0;
784
785 out_free:
786 vfree(cpuid_entries);
787 out:
788 return r;
789 }
790
791 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
792 {
793 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
794 struct kvm_cpuid_entry2 *ej;
795 int j = i;
796 int nent = vcpu->arch.cpuid_nent;
797
798 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
799 /* when no next entry is found, the current entry[i] is reselected */
800 do {
801 j = (j + 1) % nent;
802 ej = &vcpu->arch.cpuid_entries[j];
803 } while (ej->function != e->function);
804
805 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
806
807 return j;
808 }
809
810 /* find an entry with matching function, matching index (if needed), and that
811 * should be read next (if it's stateful) */
812 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
813 u32 function, u32 index)
814 {
815 if (e->function != function)
816 return 0;
817 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
818 return 0;
819 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
820 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
821 return 0;
822 return 1;
823 }
824
825 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
826 u32 function, u32 index)
827 {
828 int i;
829 struct kvm_cpuid_entry2 *best = NULL;
830
831 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
832 struct kvm_cpuid_entry2 *e;
833
834 e = &vcpu->arch.cpuid_entries[i];
835 if (is_matching_cpuid_entry(e, function, index)) {
836 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
837 move_to_next_stateful_cpuid_entry(vcpu, i);
838 best = e;
839 break;
840 }
841 }
842 return best;
843 }
844 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
845
846 /*
847 * If no match is found, check whether we exceed the vCPU's limit
848 * and return the content of the highest valid _standard_ leaf instead.
849 * This is to satisfy the CPUID specification.
850 */
851 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
852 u32 function, u32 index)
853 {
854 struct kvm_cpuid_entry2 *maxlevel;
855
856 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
857 if (!maxlevel || maxlevel->eax >= function)
858 return NULL;
859 if (function & 0x80000000) {
860 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
861 if (!maxlevel)
862 return NULL;
863 }
864 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
865 }
866
867 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
868 {
869 u32 function = *eax, index = *ecx;
870 struct kvm_cpuid_entry2 *best;
871
872 best = kvm_find_cpuid_entry(vcpu, function, index);
873
874 if (!best)
875 best = check_cpuid_limit(vcpu, function, index);
876
877 if (best) {
878 *eax = best->eax;
879 *ebx = best->ebx;
880 *ecx = best->ecx;
881 *edx = best->edx;
882 } else
883 *eax = *ebx = *ecx = *edx = 0;
884 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
885 }
886 EXPORT_SYMBOL_GPL(kvm_cpuid);
887
888 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
889 {
890 u32 eax, ebx, ecx, edx;
891
892 if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
893 return 1;
894
895 eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
896 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
897 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
898 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
899 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
900 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
901 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
902 return kvm_skip_emulated_instruction(vcpu);
903 }
904 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
Ubuntu Artful kernel mirror