Loading...
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/module.h>
17#include <linux/vmalloc.h>
18#include <linux/uaccess.h>
19#include <asm/user.h>
20#include <asm/xsave.h>
21#include "cpuid.h"
22#include "lapic.h"
23#include "mmu.h"
24#include "trace.h"
25
26static u32 xstate_required_size(u64 xstate_bv)
27{
28 int feature_bit = 0;
29 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
30
31 xstate_bv &= XSTATE_EXTEND_MASK;
32 while (xstate_bv) {
33 if (xstate_bv & 0x1) {
34 u32 eax, ebx, ecx, edx;
35 cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
36 ret = max(ret, eax + ebx);
37 }
38
39 xstate_bv >>= 1;
40 feature_bit++;
41 }
42
43 return ret;
44}
45
46u64 kvm_supported_xcr0(void)
47{
48 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
49
50 if (!kvm_x86_ops->mpx_supported())
51 xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR);
52
53 return xcr0;
54}
55
56void kvm_update_cpuid(struct kvm_vcpu *vcpu)
57{
58 struct kvm_cpuid_entry2 *best;
59 struct kvm_lapic *apic = vcpu->arch.apic;
60
61 best = kvm_find_cpuid_entry(vcpu, 1, 0);
62 if (!best)
63 return;
64
65 /* Update OSXSAVE bit */
66 if (cpu_has_xsave && best->function == 0x1) {
67 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
68 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
69 best->ecx |= bit(X86_FEATURE_OSXSAVE);
70 }
71
72 if (apic) {
73 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
74 apic->lapic_timer.timer_mode_mask = 3 << 17;
75 else
76 apic->lapic_timer.timer_mode_mask = 1 << 17;
77 }
78
79 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
80 if (!best) {
81 vcpu->arch.guest_supported_xcr0 = 0;
82 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
83 } else {
84 vcpu->arch.guest_supported_xcr0 =
85 (best->eax | ((u64)best->edx << 32)) &
86 kvm_supported_xcr0();
87 vcpu->arch.guest_xstate_size = best->ebx =
88 xstate_required_size(vcpu->arch.xcr0);
89 }
90
91 kvm_pmu_cpuid_update(vcpu);
92}
93
94static int is_efer_nx(void)
95{
96 unsigned long long efer = 0;
97
98 rdmsrl_safe(MSR_EFER, &efer);
99 return efer & EFER_NX;
100}
101
102static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
103{
104 int i;
105 struct kvm_cpuid_entry2 *e, *entry;
106
107 entry = NULL;
108 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
109 e = &vcpu->arch.cpuid_entries[i];
110 if (e->function == 0x80000001) {
111 entry = e;
112 break;
113 }
114 }
115 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
116 entry->edx &= ~(1 << 20);
117 printk(KERN_INFO "kvm: guest NX capability removed\n");
118 }
119}
120
121/* when an old userspace process fills a new kernel module */
122int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
123 struct kvm_cpuid *cpuid,
124 struct kvm_cpuid_entry __user *entries)
125{
126 int r, i;
127 struct kvm_cpuid_entry *cpuid_entries;
128
129 r = -E2BIG;
130 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
131 goto out;
132 r = -ENOMEM;
133 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
134 if (!cpuid_entries)
135 goto out;
136 r = -EFAULT;
137 if (copy_from_user(cpuid_entries, entries,
138 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
139 goto out_free;
140 for (i = 0; i < cpuid->nent; i++) {
141 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
142 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
143 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
144 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
145 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
146 vcpu->arch.cpuid_entries[i].index = 0;
147 vcpu->arch.cpuid_entries[i].flags = 0;
148 vcpu->arch.cpuid_entries[i].padding[0] = 0;
149 vcpu->arch.cpuid_entries[i].padding[1] = 0;
150 vcpu->arch.cpuid_entries[i].padding[2] = 0;
151 }
152 vcpu->arch.cpuid_nent = cpuid->nent;
153 cpuid_fix_nx_cap(vcpu);
154 r = 0;
155 kvm_apic_set_version(vcpu);
156 kvm_x86_ops->cpuid_update(vcpu);
157 kvm_update_cpuid(vcpu);
158
159out_free:
160 vfree(cpuid_entries);
161out:
162 return r;
163}
164
165int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
166 struct kvm_cpuid2 *cpuid,
167 struct kvm_cpuid_entry2 __user *entries)
168{
169 int r;
170
171 r = -E2BIG;
172 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
173 goto out;
174 r = -EFAULT;
175 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
176 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
177 goto out;
178 vcpu->arch.cpuid_nent = cpuid->nent;
179 kvm_apic_set_version(vcpu);
180 kvm_x86_ops->cpuid_update(vcpu);
181 kvm_update_cpuid(vcpu);
182 return 0;
183
184out:
185 return r;
186}
187
188int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
189 struct kvm_cpuid2 *cpuid,
190 struct kvm_cpuid_entry2 __user *entries)
191{
192 int r;
193
194 r = -E2BIG;
195 if (cpuid->nent < vcpu->arch.cpuid_nent)
196 goto out;
197 r = -EFAULT;
198 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
199 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
200 goto out;
201 return 0;
202
203out:
204 cpuid->nent = vcpu->arch.cpuid_nent;
205 return r;
206}
207
208static void cpuid_mask(u32 *word, int wordnum)
209{
210 *word &= boot_cpu_data.x86_capability[wordnum];
211}
212
213static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
214 u32 index)
215{
216 entry->function = function;
217 entry->index = index;
218 cpuid_count(entry->function, entry->index,
219 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
220 entry->flags = 0;
221}
222
223#define F(x) bit(X86_FEATURE_##x)
224
225static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
226 u32 func, u32 index, int *nent, int maxnent)
227{
228 switch (func) {
229 case 0:
230 entry->eax = 1; /* only one leaf currently */
231 ++*nent;
232 break;
233 case 1:
234 entry->ecx = F(MOVBE);
235 ++*nent;
236 break;
237 default:
238 break;
239 }
240
241 entry->function = func;
242 entry->index = index;
243
244 return 0;
245}
246
247static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
248 u32 index, int *nent, int maxnent)
249{
250 int r;
251 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
252#ifdef CONFIG_X86_64
253 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
254 ? F(GBPAGES) : 0;
255 unsigned f_lm = F(LM);
256#else
257 unsigned f_gbpages = 0;
258 unsigned f_lm = 0;
259#endif
260 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
261 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
262 unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
263
264 /* cpuid 1.edx */
265 const u32 kvm_supported_word0_x86_features =
266 F(FPU) | F(VME) | F(DE) | F(PSE) |
267 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
268 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
269 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
270 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
271 0 /* Reserved, DS, ACPI */ | F(MMX) |
272 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
273 0 /* HTT, TM, Reserved, PBE */;
274 /* cpuid 0x80000001.edx */
275 const u32 kvm_supported_word1_x86_features =
276 F(FPU) | F(VME) | F(DE) | F(PSE) |
277 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
278 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
279 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
280 F(PAT) | F(PSE36) | 0 /* Reserved */ |
281 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
282 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
283 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
284 /* cpuid 1.ecx */
285 const u32 kvm_supported_word4_x86_features =
286 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
287 0 /* DS-CPL, VMX, SMX, EST */ |
288 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
289 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
290 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
291 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
292 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
293 F(F16C) | F(RDRAND);
294 /* cpuid 0x80000001.ecx */
295 const u32 kvm_supported_word6_x86_features =
296 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
297 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
298 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
299 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
300
301 /* cpuid 0xC0000001.edx */
302 const u32 kvm_supported_word5_x86_features =
303 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
304 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
305 F(PMM) | F(PMM_EN);
306
307 /* cpuid 7.0.ebx */
308 const u32 kvm_supported_word9_x86_features =
309 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
310 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
311 F(ADX) | F(SMAP);
312
313 /* all calls to cpuid_count() should be made on the same cpu */
314 get_cpu();
315
316 r = -E2BIG;
317
318 if (*nent >= maxnent)
319 goto out;
320
321 do_cpuid_1_ent(entry, function, index);
322 ++*nent;
323
324 switch (function) {
325 case 0:
326 entry->eax = min(entry->eax, (u32)0xd);
327 break;
328 case 1:
329 entry->edx &= kvm_supported_word0_x86_features;
330 cpuid_mask(&entry->edx, 0);
331 entry->ecx &= kvm_supported_word4_x86_features;
332 cpuid_mask(&entry->ecx, 4);
333 /* we support x2apic emulation even if host does not support
334 * it since we emulate x2apic in software */
335 entry->ecx |= F(X2APIC);
336 break;
337 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
338 * may return different values. This forces us to get_cpu() before
339 * issuing the first command, and also to emulate this annoying behavior
340 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
341 case 2: {
342 int t, times = entry->eax & 0xff;
343
344 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
345 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
346 for (t = 1; t < times; ++t) {
347 if (*nent >= maxnent)
348 goto out;
349
350 do_cpuid_1_ent(&entry[t], function, 0);
351 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
352 ++*nent;
353 }
354 break;
355 }
356 /* function 4 has additional index. */
357 case 4: {
358 int i, cache_type;
359
360 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
361 /* read more entries until cache_type is zero */
362 for (i = 1; ; ++i) {
363 if (*nent >= maxnent)
364 goto out;
365
366 cache_type = entry[i - 1].eax & 0x1f;
367 if (!cache_type)
368 break;
369 do_cpuid_1_ent(&entry[i], function, i);
370 entry[i].flags |=
371 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
372 ++*nent;
373 }
374 break;
375 }
376 case 7: {
377 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
378 /* Mask ebx against host capability word 9 */
379 if (index == 0) {
380 entry->ebx &= kvm_supported_word9_x86_features;
381 cpuid_mask(&entry->ebx, 9);
382 // TSC_ADJUST is emulated
383 entry->ebx |= F(TSC_ADJUST);
384 } else
385 entry->ebx = 0;
386 entry->eax = 0;
387 entry->ecx = 0;
388 entry->edx = 0;
389 break;
390 }
391 case 9:
392 break;
393 case 0xa: { /* Architectural Performance Monitoring */
394 struct x86_pmu_capability cap;
395 union cpuid10_eax eax;
396 union cpuid10_edx edx;
397
398 perf_get_x86_pmu_capability(&cap);
399
400 /*
401 * Only support guest architectural pmu on a host
402 * with architectural pmu.
403 */
404 if (!cap.version)
405 memset(&cap, 0, sizeof(cap));
406
407 eax.split.version_id = min(cap.version, 2);
408 eax.split.num_counters = cap.num_counters_gp;
409 eax.split.bit_width = cap.bit_width_gp;
410 eax.split.mask_length = cap.events_mask_len;
411
412 edx.split.num_counters_fixed = cap.num_counters_fixed;
413 edx.split.bit_width_fixed = cap.bit_width_fixed;
414 edx.split.reserved = 0;
415
416 entry->eax = eax.full;
417 entry->ebx = cap.events_mask;
418 entry->ecx = 0;
419 entry->edx = edx.full;
420 break;
421 }
422 /* function 0xb has additional index. */
423 case 0xb: {
424 int i, level_type;
425
426 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
427 /* read more entries until level_type is zero */
428 for (i = 1; ; ++i) {
429 if (*nent >= maxnent)
430 goto out;
431
432 level_type = entry[i - 1].ecx & 0xff00;
433 if (!level_type)
434 break;
435 do_cpuid_1_ent(&entry[i], function, i);
436 entry[i].flags |=
437 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
438 ++*nent;
439 }
440 break;
441 }
442 case 0xd: {
443 int idx, i;
444 u64 supported = kvm_supported_xcr0();
445
446 entry->eax &= supported;
447 entry->edx &= supported >> 32;
448 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
449 for (idx = 1, i = 1; idx < 64; ++idx) {
450 u64 mask = ((u64)1 << idx);
451 if (*nent >= maxnent)
452 goto out;
453
454 do_cpuid_1_ent(&entry[i], function, idx);
455 if (entry[i].eax == 0 || !(supported & mask))
456 continue;
457 entry[i].flags |=
458 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
459 ++*nent;
460 ++i;
461 }
462 break;
463 }
464 case KVM_CPUID_SIGNATURE: {
465 static const char signature[12] = "KVMKVMKVM\0\0";
466 const u32 *sigptr = (const u32 *)signature;
467 entry->eax = KVM_CPUID_FEATURES;
468 entry->ebx = sigptr[0];
469 entry->ecx = sigptr[1];
470 entry->edx = sigptr[2];
471 break;
472 }
473 case KVM_CPUID_FEATURES:
474 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
475 (1 << KVM_FEATURE_NOP_IO_DELAY) |
476 (1 << KVM_FEATURE_CLOCKSOURCE2) |
477 (1 << KVM_FEATURE_ASYNC_PF) |
478 (1 << KVM_FEATURE_PV_EOI) |
479 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
480 (1 << KVM_FEATURE_PV_UNHALT);
481
482 if (sched_info_on())
483 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
484
485 entry->ebx = 0;
486 entry->ecx = 0;
487 entry->edx = 0;
488 break;
489 case 0x80000000:
490 entry->eax = min(entry->eax, 0x8000001a);
491 break;
492 case 0x80000001:
493 entry->edx &= kvm_supported_word1_x86_features;
494 cpuid_mask(&entry->edx, 1);
495 entry->ecx &= kvm_supported_word6_x86_features;
496 cpuid_mask(&entry->ecx, 6);
497 break;
498 case 0x80000008: {
499 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
500 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
501 unsigned phys_as = entry->eax & 0xff;
502
503 if (!g_phys_as)
504 g_phys_as = phys_as;
505 entry->eax = g_phys_as | (virt_as << 8);
506 entry->ebx = entry->edx = 0;
507 break;
508 }
509 case 0x80000019:
510 entry->ecx = entry->edx = 0;
511 break;
512 case 0x8000001a:
513 break;
514 case 0x8000001d:
515 break;
516 /*Add support for Centaur's CPUID instruction*/
517 case 0xC0000000:
518 /*Just support up to 0xC0000004 now*/
519 entry->eax = min(entry->eax, 0xC0000004);
520 break;
521 case 0xC0000001:
522 entry->edx &= kvm_supported_word5_x86_features;
523 cpuid_mask(&entry->edx, 5);
524 break;
525 case 3: /* Processor serial number */
526 case 5: /* MONITOR/MWAIT */
527 case 6: /* Thermal management */
528 case 0x80000007: /* Advanced power management */
529 case 0xC0000002:
530 case 0xC0000003:
531 case 0xC0000004:
532 default:
533 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
534 break;
535 }
536
537 kvm_x86_ops->set_supported_cpuid(function, entry);
538
539 r = 0;
540
541out:
542 put_cpu();
543
544 return r;
545}
546
547static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
548 u32 idx, int *nent, int maxnent, unsigned int type)
549{
550 if (type == KVM_GET_EMULATED_CPUID)
551 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
552
553 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
554}
555
556#undef F
557
558struct kvm_cpuid_param {
559 u32 func;
560 u32 idx;
561 bool has_leaf_count;
562 bool (*qualifier)(const struct kvm_cpuid_param *param);
563};
564
565static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
566{
567 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
568}
569
570static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
571 __u32 num_entries, unsigned int ioctl_type)
572{
573 int i;
574 __u32 pad[3];
575
576 if (ioctl_type != KVM_GET_EMULATED_CPUID)
577 return false;
578
579 /*
580 * We want to make sure that ->padding is being passed clean from
581 * userspace in case we want to use it for something in the future.
582 *
583 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
584 * have to give ourselves satisfied only with the emulated side. /me
585 * sheds a tear.
586 */
587 for (i = 0; i < num_entries; i++) {
588 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
589 return true;
590
591 if (pad[0] || pad[1] || pad[2])
592 return true;
593 }
594 return false;
595}
596
597int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
598 struct kvm_cpuid_entry2 __user *entries,
599 unsigned int type)
600{
601 struct kvm_cpuid_entry2 *cpuid_entries;
602 int limit, nent = 0, r = -E2BIG, i;
603 u32 func;
604 static const struct kvm_cpuid_param param[] = {
605 { .func = 0, .has_leaf_count = true },
606 { .func = 0x80000000, .has_leaf_count = true },
607 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
608 { .func = KVM_CPUID_SIGNATURE },
609 { .func = KVM_CPUID_FEATURES },
610 };
611
612 if (cpuid->nent < 1)
613 goto out;
614 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
615 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
616
617 if (sanity_check_entries(entries, cpuid->nent, type))
618 return -EINVAL;
619
620 r = -ENOMEM;
621 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
622 if (!cpuid_entries)
623 goto out;
624
625 r = 0;
626 for (i = 0; i < ARRAY_SIZE(param); i++) {
627 const struct kvm_cpuid_param *ent = ¶m[i];
628
629 if (ent->qualifier && !ent->qualifier(ent))
630 continue;
631
632 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
633 &nent, cpuid->nent, type);
634
635 if (r)
636 goto out_free;
637
638 if (!ent->has_leaf_count)
639 continue;
640
641 limit = cpuid_entries[nent - 1].eax;
642 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
643 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
644 &nent, cpuid->nent, type);
645
646 if (r)
647 goto out_free;
648 }
649
650 r = -EFAULT;
651 if (copy_to_user(entries, cpuid_entries,
652 nent * sizeof(struct kvm_cpuid_entry2)))
653 goto out_free;
654 cpuid->nent = nent;
655 r = 0;
656
657out_free:
658 vfree(cpuid_entries);
659out:
660 return r;
661}
662
663static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
664{
665 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
666 int j, nent = vcpu->arch.cpuid_nent;
667
668 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
669 /* when no next entry is found, the current entry[i] is reselected */
670 for (j = i + 1; ; j = (j + 1) % nent) {
671 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
672 if (ej->function == e->function) {
673 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
674 return j;
675 }
676 }
677 return 0; /* silence gcc, even though control never reaches here */
678}
679
680/* find an entry with matching function, matching index (if needed), and that
681 * should be read next (if it's stateful) */
682static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
683 u32 function, u32 index)
684{
685 if (e->function != function)
686 return 0;
687 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
688 return 0;
689 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
690 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
691 return 0;
692 return 1;
693}
694
695struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
696 u32 function, u32 index)
697{
698 int i;
699 struct kvm_cpuid_entry2 *best = NULL;
700
701 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
702 struct kvm_cpuid_entry2 *e;
703
704 e = &vcpu->arch.cpuid_entries[i];
705 if (is_matching_cpuid_entry(e, function, index)) {
706 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
707 move_to_next_stateful_cpuid_entry(vcpu, i);
708 best = e;
709 break;
710 }
711 }
712 return best;
713}
714EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
715
716int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
717{
718 struct kvm_cpuid_entry2 *best;
719
720 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
721 if (!best || best->eax < 0x80000008)
722 goto not_found;
723 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
724 if (best)
725 return best->eax & 0xff;
726not_found:
727 return 36;
728}
729
730/*
731 * If no match is found, check whether we exceed the vCPU's limit
732 * and return the content of the highest valid _standard_ leaf instead.
733 * This is to satisfy the CPUID specification.
734 */
735static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
736 u32 function, u32 index)
737{
738 struct kvm_cpuid_entry2 *maxlevel;
739
740 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
741 if (!maxlevel || maxlevel->eax >= function)
742 return NULL;
743 if (function & 0x80000000) {
744 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
745 if (!maxlevel)
746 return NULL;
747 }
748 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
749}
750
751void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
752{
753 u32 function = *eax, index = *ecx;
754 struct kvm_cpuid_entry2 *best;
755
756 best = kvm_find_cpuid_entry(vcpu, function, index);
757
758 if (!best)
759 best = check_cpuid_limit(vcpu, function, index);
760
761 if (best) {
762 *eax = best->eax;
763 *ebx = best->ebx;
764 *ecx = best->ecx;
765 *edx = best->edx;
766 } else
767 *eax = *ebx = *ecx = *edx = 0;
768 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
769}
770EXPORT_SYMBOL_GPL(kvm_cpuid);
771
772void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
773{
774 u32 function, eax, ebx, ecx, edx;
775
776 function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
777 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
778 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
779 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
780 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
781 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
782 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
783 kvm_x86_ops->skip_emulated_instruction(vcpu);
784}
785EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
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/processor.h>
20#include <asm/user.h>
21#include <asm/fpu/xstate.h>
22#include "cpuid.h"
23#include "lapic.h"
24#include "mmu.h"
25#include "trace.h"
26#include "pmu.h"
27
28static u32 xstate_required_size(u64 xstate_bv, bool compacted)
29{
30 int feature_bit = 0;
31 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
32
33 xstate_bv &= XFEATURE_MASK_EXTEND;
34 while (xstate_bv) {
35 if (xstate_bv & 0x1) {
36 u32 eax, ebx, ecx, edx, offset;
37 cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
38 offset = compacted ? ret : ebx;
39 ret = max(ret, offset + eax);
40 }
41
42 xstate_bv >>= 1;
43 feature_bit++;
44 }
45
46 return ret;
47}
48
49bool kvm_mpx_supported(void)
50{
51 return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
52 && kvm_x86_ops->mpx_supported());
53}
54EXPORT_SYMBOL_GPL(kvm_mpx_supported);
55
56u64 kvm_supported_xcr0(void)
57{
58 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
59
60 if (!kvm_mpx_supported())
61 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
62
63 return xcr0;
64}
65
66#define F(x) bit(X86_FEATURE_##x)
67
68/* These are scattered features in cpufeatures.h. */
69#define KVM_CPUID_BIT_AVX512_4VNNIW 2
70#define KVM_CPUID_BIT_AVX512_4FMAPS 3
71#define KF(x) bit(KVM_CPUID_BIT_##x)
72
73int 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 kvm_x86_ops->fpu_activate(vcpu);
127
128 /*
129 * The existing code assumes virtual address is 48-bit in the canonical
130 * address checks; exit if it is ever changed.
131 */
132 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
133 if (best && ((best->eax & 0xff00) >> 8) != 48 &&
134 ((best->eax & 0xff00) >> 8) != 0)
135 return -EINVAL;
136
137 /* Update physical-address width */
138 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
139
140 kvm_pmu_refresh(vcpu);
141 return 0;
142}
143
144static int is_efer_nx(void)
145{
146 unsigned long long efer = 0;
147
148 rdmsrl_safe(MSR_EFER, &efer);
149 return efer & EFER_NX;
150}
151
152static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
153{
154 int i;
155 struct kvm_cpuid_entry2 *e, *entry;
156
157 entry = NULL;
158 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
159 e = &vcpu->arch.cpuid_entries[i];
160 if (e->function == 0x80000001) {
161 entry = e;
162 break;
163 }
164 }
165 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
166 entry->edx &= ~F(NX);
167 printk(KERN_INFO "kvm: guest NX capability removed\n");
168 }
169}
170
171int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
172{
173 struct kvm_cpuid_entry2 *best;
174
175 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
176 if (!best || best->eax < 0x80000008)
177 goto not_found;
178 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
179 if (best)
180 return best->eax & 0xff;
181not_found:
182 return 36;
183}
184EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
185
186/* when an old userspace process fills a new kernel module */
187int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
188 struct kvm_cpuid *cpuid,
189 struct kvm_cpuid_entry __user *entries)
190{
191 int r, i;
192 struct kvm_cpuid_entry *cpuid_entries = NULL;
193
194 r = -E2BIG;
195 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
196 goto out;
197 r = -ENOMEM;
198 if (cpuid->nent) {
199 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
200 cpuid->nent);
201 if (!cpuid_entries)
202 goto out;
203 r = -EFAULT;
204 if (copy_from_user(cpuid_entries, entries,
205 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
206 goto out;
207 }
208 for (i = 0; i < cpuid->nent; i++) {
209 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
210 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
211 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
212 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
213 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
214 vcpu->arch.cpuid_entries[i].index = 0;
215 vcpu->arch.cpuid_entries[i].flags = 0;
216 vcpu->arch.cpuid_entries[i].padding[0] = 0;
217 vcpu->arch.cpuid_entries[i].padding[1] = 0;
218 vcpu->arch.cpuid_entries[i].padding[2] = 0;
219 }
220 vcpu->arch.cpuid_nent = cpuid->nent;
221 cpuid_fix_nx_cap(vcpu);
222 kvm_apic_set_version(vcpu);
223 kvm_x86_ops->cpuid_update(vcpu);
224 r = kvm_update_cpuid(vcpu);
225
226out:
227 vfree(cpuid_entries);
228 return r;
229}
230
231int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
232 struct kvm_cpuid2 *cpuid,
233 struct kvm_cpuid_entry2 __user *entries)
234{
235 int r;
236
237 r = -E2BIG;
238 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
239 goto out;
240 r = -EFAULT;
241 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
242 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
243 goto out;
244 vcpu->arch.cpuid_nent = cpuid->nent;
245 kvm_apic_set_version(vcpu);
246 kvm_x86_ops->cpuid_update(vcpu);
247 r = kvm_update_cpuid(vcpu);
248out:
249 return r;
250}
251
252int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
253 struct kvm_cpuid2 *cpuid,
254 struct kvm_cpuid_entry2 __user *entries)
255{
256 int r;
257
258 r = -E2BIG;
259 if (cpuid->nent < vcpu->arch.cpuid_nent)
260 goto out;
261 r = -EFAULT;
262 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
263 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
264 goto out;
265 return 0;
266
267out:
268 cpuid->nent = vcpu->arch.cpuid_nent;
269 return r;
270}
271
272static void cpuid_mask(u32 *word, int wordnum)
273{
274 *word &= boot_cpu_data.x86_capability[wordnum];
275}
276
277static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
278 u32 index)
279{
280 entry->function = function;
281 entry->index = index;
282 cpuid_count(entry->function, entry->index,
283 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
284 entry->flags = 0;
285}
286
287static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
288 u32 func, u32 index, int *nent, int maxnent)
289{
290 switch (func) {
291 case 0:
292 entry->eax = 1; /* only one leaf currently */
293 ++*nent;
294 break;
295 case 1:
296 entry->ecx = F(MOVBE);
297 ++*nent;
298 break;
299 default:
300 break;
301 }
302
303 entry->function = func;
304 entry->index = index;
305
306 return 0;
307}
308
309static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
310 u32 index, int *nent, int maxnent)
311{
312 int r;
313 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
314#ifdef CONFIG_X86_64
315 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
316 ? F(GBPAGES) : 0;
317 unsigned f_lm = F(LM);
318#else
319 unsigned f_gbpages = 0;
320 unsigned f_lm = 0;
321#endif
322 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
323 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
324 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
325 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
326
327 /* cpuid 1.edx */
328 const u32 kvm_cpuid_1_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(SEP) |
332 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
333 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
334 0 /* Reserved, DS, ACPI */ | F(MMX) |
335 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
336 0 /* HTT, TM, Reserved, PBE */;
337 /* cpuid 0x80000001.edx */
338 const u32 kvm_cpuid_8000_0001_edx_x86_features =
339 F(FPU) | F(VME) | F(DE) | F(PSE) |
340 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
341 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
342 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
343 F(PAT) | F(PSE36) | 0 /* Reserved */ |
344 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
345 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
346 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
347 /* cpuid 1.ecx */
348 const u32 kvm_cpuid_1_ecx_x86_features =
349 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
350 * but *not* advertised to guests via CPUID ! */
351 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
352 0 /* DS-CPL, VMX, SMX, EST */ |
353 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
354 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
355 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
356 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
357 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
358 F(F16C) | F(RDRAND);
359 /* cpuid 0x80000001.ecx */
360 const u32 kvm_cpuid_8000_0001_ecx_x86_features =
361 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
362 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
363 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
364 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
365
366 /* cpuid 0xC0000001.edx */
367 const u32 kvm_cpuid_C000_0001_edx_x86_features =
368 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
369 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
370 F(PMM) | F(PMM_EN);
371
372 /* cpuid 7.0.ebx */
373 const u32 kvm_cpuid_7_0_ebx_x86_features =
374 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
375 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
376 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
377 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
378 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
379
380 /* cpuid 0xD.1.eax */
381 const u32 kvm_cpuid_D_1_eax_x86_features =
382 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
383
384 /* cpuid 7.0.ecx*/
385 const u32 kvm_cpuid_7_0_ecx_x86_features =
386 F(AVX512VBMI) | F(PKU) | 0 /*OSPKE*/;
387
388 /* cpuid 7.0.edx*/
389 const u32 kvm_cpuid_7_0_edx_x86_features =
390 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
391
392 /* all calls to cpuid_count() should be made on the same cpu */
393 get_cpu();
394
395 r = -E2BIG;
396
397 if (*nent >= maxnent)
398 goto out;
399
400 do_cpuid_1_ent(entry, function, index);
401 ++*nent;
402
403 switch (function) {
404 case 0:
405 entry->eax = min(entry->eax, (u32)0xd);
406 break;
407 case 1:
408 entry->edx &= kvm_cpuid_1_edx_x86_features;
409 cpuid_mask(&entry->edx, CPUID_1_EDX);
410 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
411 cpuid_mask(&entry->ecx, CPUID_1_ECX);
412 /* we support x2apic emulation even if host does not support
413 * it since we emulate x2apic in software */
414 entry->ecx |= F(X2APIC);
415 break;
416 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
417 * may return different values. This forces us to get_cpu() before
418 * issuing the first command, and also to emulate this annoying behavior
419 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
420 case 2: {
421 int t, times = entry->eax & 0xff;
422
423 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
424 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
425 for (t = 1; t < times; ++t) {
426 if (*nent >= maxnent)
427 goto out;
428
429 do_cpuid_1_ent(&entry[t], function, 0);
430 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
431 ++*nent;
432 }
433 break;
434 }
435 /* function 4 has additional index. */
436 case 4: {
437 int i, cache_type;
438
439 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
440 /* read more entries until cache_type is zero */
441 for (i = 1; ; ++i) {
442 if (*nent >= maxnent)
443 goto out;
444
445 cache_type = entry[i - 1].eax & 0x1f;
446 if (!cache_type)
447 break;
448 do_cpuid_1_ent(&entry[i], function, i);
449 entry[i].flags |=
450 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
451 ++*nent;
452 }
453 break;
454 }
455 case 6: /* Thermal management */
456 entry->eax = 0x4; /* allow ARAT */
457 entry->ebx = 0;
458 entry->ecx = 0;
459 entry->edx = 0;
460 break;
461 case 7: {
462 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
463 /* Mask ebx against host capability word 9 */
464 if (index == 0) {
465 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
466 cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
467 // TSC_ADJUST is emulated
468 entry->ebx |= F(TSC_ADJUST);
469 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
470 cpuid_mask(&entry->ecx, CPUID_7_ECX);
471 /* PKU is not yet implemented for shadow paging. */
472 if (!tdp_enabled)
473 entry->ecx &= ~F(PKU);
474 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
475 entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
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->ebx = entry->edx = 0;
626 break;
627 }
628 case 0x80000019:
629 entry->ecx = entry->edx = 0;
630 break;
631 case 0x8000001a:
632 break;
633 case 0x8000001d:
634 break;
635 /*Add support for Centaur's CPUID instruction*/
636 case 0xC0000000:
637 /*Just support up to 0xC0000004 now*/
638 entry->eax = min(entry->eax, 0xC0000004);
639 break;
640 case 0xC0000001:
641 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
642 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
643 break;
644 case 3: /* Processor serial number */
645 case 5: /* MONITOR/MWAIT */
646 case 0xC0000002:
647 case 0xC0000003:
648 case 0xC0000004:
649 default:
650 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
651 break;
652 }
653
654 kvm_x86_ops->set_supported_cpuid(function, entry);
655
656 r = 0;
657
658out:
659 put_cpu();
660
661 return r;
662}
663
664static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
665 u32 idx, int *nent, int maxnent, unsigned int type)
666{
667 if (type == KVM_GET_EMULATED_CPUID)
668 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
669
670 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
671}
672
673#undef F
674
675struct kvm_cpuid_param {
676 u32 func;
677 u32 idx;
678 bool has_leaf_count;
679 bool (*qualifier)(const struct kvm_cpuid_param *param);
680};
681
682static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
683{
684 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
685}
686
687static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
688 __u32 num_entries, unsigned int ioctl_type)
689{
690 int i;
691 __u32 pad[3];
692
693 if (ioctl_type != KVM_GET_EMULATED_CPUID)
694 return false;
695
696 /*
697 * We want to make sure that ->padding is being passed clean from
698 * userspace in case we want to use it for something in the future.
699 *
700 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
701 * have to give ourselves satisfied only with the emulated side. /me
702 * sheds a tear.
703 */
704 for (i = 0; i < num_entries; i++) {
705 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
706 return true;
707
708 if (pad[0] || pad[1] || pad[2])
709 return true;
710 }
711 return false;
712}
713
714int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
715 struct kvm_cpuid_entry2 __user *entries,
716 unsigned int type)
717{
718 struct kvm_cpuid_entry2 *cpuid_entries;
719 int limit, nent = 0, r = -E2BIG, i;
720 u32 func;
721 static const struct kvm_cpuid_param param[] = {
722 { .func = 0, .has_leaf_count = true },
723 { .func = 0x80000000, .has_leaf_count = true },
724 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
725 { .func = KVM_CPUID_SIGNATURE },
726 { .func = KVM_CPUID_FEATURES },
727 };
728
729 if (cpuid->nent < 1)
730 goto out;
731 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
732 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
733
734 if (sanity_check_entries(entries, cpuid->nent, type))
735 return -EINVAL;
736
737 r = -ENOMEM;
738 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
739 if (!cpuid_entries)
740 goto out;
741
742 r = 0;
743 for (i = 0; i < ARRAY_SIZE(param); i++) {
744 const struct kvm_cpuid_param *ent = ¶m[i];
745
746 if (ent->qualifier && !ent->qualifier(ent))
747 continue;
748
749 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
750 &nent, cpuid->nent, type);
751
752 if (r)
753 goto out_free;
754
755 if (!ent->has_leaf_count)
756 continue;
757
758 limit = cpuid_entries[nent - 1].eax;
759 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
760 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
761 &nent, cpuid->nent, type);
762
763 if (r)
764 goto out_free;
765 }
766
767 r = -EFAULT;
768 if (copy_to_user(entries, cpuid_entries,
769 nent * sizeof(struct kvm_cpuid_entry2)))
770 goto out_free;
771 cpuid->nent = nent;
772 r = 0;
773
774out_free:
775 vfree(cpuid_entries);
776out:
777 return r;
778}
779
780static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
781{
782 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
783 int j, nent = vcpu->arch.cpuid_nent;
784
785 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
786 /* when no next entry is found, the current entry[i] is reselected */
787 for (j = i + 1; ; j = (j + 1) % nent) {
788 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
789 if (ej->function == e->function) {
790 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
791 return j;
792 }
793 }
794 return 0; /* silence gcc, even though control never reaches here */
795}
796
797/* find an entry with matching function, matching index (if needed), and that
798 * should be read next (if it's stateful) */
799static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
800 u32 function, u32 index)
801{
802 if (e->function != function)
803 return 0;
804 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
805 return 0;
806 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
807 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
808 return 0;
809 return 1;
810}
811
812struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
813 u32 function, u32 index)
814{
815 int i;
816 struct kvm_cpuid_entry2 *best = NULL;
817
818 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
819 struct kvm_cpuid_entry2 *e;
820
821 e = &vcpu->arch.cpuid_entries[i];
822 if (is_matching_cpuid_entry(e, function, index)) {
823 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
824 move_to_next_stateful_cpuid_entry(vcpu, i);
825 best = e;
826 break;
827 }
828 }
829 return best;
830}
831EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
832
833/*
834 * If no match is found, check whether we exceed the vCPU's limit
835 * and return the content of the highest valid _standard_ leaf instead.
836 * This is to satisfy the CPUID specification.
837 */
838static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
839 u32 function, u32 index)
840{
841 struct kvm_cpuid_entry2 *maxlevel;
842
843 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
844 if (!maxlevel || maxlevel->eax >= function)
845 return NULL;
846 if (function & 0x80000000) {
847 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
848 if (!maxlevel)
849 return NULL;
850 }
851 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
852}
853
854void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
855{
856 u32 function = *eax, index = *ecx;
857 struct kvm_cpuid_entry2 *best;
858
859 best = kvm_find_cpuid_entry(vcpu, function, index);
860
861 if (!best)
862 best = check_cpuid_limit(vcpu, function, index);
863
864 /*
865 * Perfmon not yet supported for L2 guest.
866 */
867 if (is_guest_mode(vcpu) && function == 0xa)
868 best = NULL;
869
870 if (best) {
871 *eax = best->eax;
872 *ebx = best->ebx;
873 *ecx = best->ecx;
874 *edx = best->edx;
875 } else
876 *eax = *ebx = *ecx = *edx = 0;
877 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
878}
879EXPORT_SYMBOL_GPL(kvm_cpuid);
880
881int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
882{
883 u32 eax, ebx, ecx, edx;
884
885 eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
886 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
887 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
888 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
889 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
890 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
891 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
892 return kvm_skip_emulated_instruction(vcpu);
893}
894EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);