2 * Kernel-based Virtual Machine driver for Linux
3 * cpuid support routines
5 * derived from arch/x86/kvm/x86.c
7 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
8 * Copyright IBM Corporation, 2008
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
15 #include <linux/kvm_host.h>
16 #include <linux/module.h>
17 #include <linux/vmalloc.h>
18 #include <linux/uaccess.h>
20 #include <asm/xsave.h>
26 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
28 struct kvm_cpuid_entry2 *best;
29 struct kvm_lapic *apic = vcpu->arch.apic;
31 best = kvm_find_cpuid_entry(vcpu, 1, 0);
35 /* Update OSXSAVE bit */
36 if (cpu_has_xsave && best->function == 0x1) {
37 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
38 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
39 best->ecx |= bit(X86_FEATURE_OSXSAVE);
43 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
44 apic->lapic_timer.timer_mode_mask = 3 << 17;
46 apic->lapic_timer.timer_mode_mask = 1 << 17;
49 kvm_pmu_cpuid_update(vcpu);
52 static int is_efer_nx(void)
54 unsigned long long efer = 0;
56 rdmsrl_safe(MSR_EFER, &efer);
57 return efer & EFER_NX;
60 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
63 struct kvm_cpuid_entry2 *e, *entry;
66 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
67 e = &vcpu->arch.cpuid_entries[i];
68 if (e->function == 0x80000001) {
73 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
74 entry->edx &= ~(1 << 20);
75 printk(KERN_INFO "kvm: guest NX capability removed\n");
79 /* when an old userspace process fills a new kernel module */
80 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
81 struct kvm_cpuid *cpuid,
82 struct kvm_cpuid_entry __user *entries)
85 struct kvm_cpuid_entry *cpuid_entries;
88 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
91 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
95 if (copy_from_user(cpuid_entries, entries,
96 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
98 for (i = 0; i < cpuid->nent; i++) {
99 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
100 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
101 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
102 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
103 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
104 vcpu->arch.cpuid_entries[i].index = 0;
105 vcpu->arch.cpuid_entries[i].flags = 0;
106 vcpu->arch.cpuid_entries[i].padding[0] = 0;
107 vcpu->arch.cpuid_entries[i].padding[1] = 0;
108 vcpu->arch.cpuid_entries[i].padding[2] = 0;
110 vcpu->arch.cpuid_nent = cpuid->nent;
111 cpuid_fix_nx_cap(vcpu);
113 kvm_apic_set_version(vcpu);
114 kvm_x86_ops->cpuid_update(vcpu);
115 kvm_update_cpuid(vcpu);
118 vfree(cpuid_entries);
123 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
124 struct kvm_cpuid2 *cpuid,
125 struct kvm_cpuid_entry2 __user *entries)
130 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
133 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
134 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
136 vcpu->arch.cpuid_nent = cpuid->nent;
137 kvm_apic_set_version(vcpu);
138 kvm_x86_ops->cpuid_update(vcpu);
139 kvm_update_cpuid(vcpu);
146 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
147 struct kvm_cpuid2 *cpuid,
148 struct kvm_cpuid_entry2 __user *entries)
153 if (cpuid->nent < vcpu->arch.cpuid_nent)
156 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
157 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
162 cpuid->nent = vcpu->arch.cpuid_nent;
166 static void cpuid_mask(u32 *word, int wordnum)
168 *word &= boot_cpu_data.x86_capability[wordnum];
171 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
174 entry->function = function;
175 entry->index = index;
176 cpuid_count(entry->function, entry->index,
177 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
181 static bool supported_xcr0_bit(unsigned bit)
183 u64 mask = ((u64)1 << bit);
185 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
188 #define F(x) bit(X86_FEATURE_##x)
190 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
191 u32 index, int *nent, int maxnent)
194 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
196 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
198 unsigned f_lm = F(LM);
200 unsigned f_gbpages = 0;
203 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
206 const u32 kvm_supported_word0_x86_features =
207 F(FPU) | F(VME) | F(DE) | F(PSE) |
208 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
209 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
210 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
211 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
212 0 /* Reserved, DS, ACPI */ | F(MMX) |
213 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
214 0 /* HTT, TM, Reserved, PBE */;
215 /* cpuid 0x80000001.edx */
216 const u32 kvm_supported_word1_x86_features =
217 F(FPU) | F(VME) | F(DE) | F(PSE) |
218 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
219 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
220 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
221 F(PAT) | F(PSE36) | 0 /* Reserved */ |
222 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
223 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
224 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
226 const u32 kvm_supported_word4_x86_features =
227 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
228 0 /* DS-CPL, VMX, SMX, EST */ |
229 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
230 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
231 0 /* Reserved, DCA */ | F(XMM4_1) |
232 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
233 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
235 /* cpuid 0x80000001.ecx */
236 const u32 kvm_supported_word6_x86_features =
237 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
238 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
239 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
240 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
242 /* cpuid 0xC0000001.edx */
243 const u32 kvm_supported_word5_x86_features =
244 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
245 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
249 const u32 kvm_supported_word9_x86_features =
250 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
251 F(BMI2) | F(ERMS) | F(RTM);
253 /* all calls to cpuid_count() should be made on the same cpu */
258 if (*nent >= maxnent)
261 do_cpuid_1_ent(entry, function, index);
266 entry->eax = min(entry->eax, (u32)0xd);
269 entry->edx &= kvm_supported_word0_x86_features;
270 cpuid_mask(&entry->edx, 0);
271 entry->ecx &= kvm_supported_word4_x86_features;
272 cpuid_mask(&entry->ecx, 4);
273 /* we support x2apic emulation even if host does not support
274 * it since we emulate x2apic in software */
275 entry->ecx |= F(X2APIC);
277 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
278 * may return different values. This forces us to get_cpu() before
279 * issuing the first command, and also to emulate this annoying behavior
280 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
282 int t, times = entry->eax & 0xff;
284 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
285 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
286 for (t = 1; t < times; ++t) {
287 if (*nent >= maxnent)
290 do_cpuid_1_ent(&entry[t], function, 0);
291 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
296 /* function 4 has additional index. */
300 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
301 /* read more entries until cache_type is zero */
303 if (*nent >= maxnent)
306 cache_type = entry[i - 1].eax & 0x1f;
309 do_cpuid_1_ent(&entry[i], function, i);
311 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
317 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
318 /* Mask ebx against host capbability word 9 */
320 entry->ebx &= kvm_supported_word9_x86_features;
321 cpuid_mask(&entry->ebx, 9);
331 case 0xa: { /* Architectural Performance Monitoring */
332 struct x86_pmu_capability cap;
333 union cpuid10_eax eax;
334 union cpuid10_edx edx;
336 perf_get_x86_pmu_capability(&cap);
339 * Only support guest architectural pmu on a host
340 * with architectural pmu.
343 memset(&cap, 0, sizeof(cap));
345 eax.split.version_id = min(cap.version, 2);
346 eax.split.num_counters = cap.num_counters_gp;
347 eax.split.bit_width = cap.bit_width_gp;
348 eax.split.mask_length = cap.events_mask_len;
350 edx.split.num_counters_fixed = cap.num_counters_fixed;
351 edx.split.bit_width_fixed = cap.bit_width_fixed;
352 edx.split.reserved = 0;
354 entry->eax = eax.full;
355 entry->ebx = cap.events_mask;
357 entry->edx = edx.full;
360 /* function 0xb has additional index. */
364 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
365 /* read more entries until level_type is zero */
367 if (*nent >= maxnent)
370 level_type = entry[i - 1].ecx & 0xff00;
373 do_cpuid_1_ent(&entry[i], function, i);
375 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
383 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
384 for (idx = 1, i = 1; idx < 64; ++idx) {
385 if (*nent >= maxnent)
388 do_cpuid_1_ent(&entry[i], function, idx);
389 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
392 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
398 case KVM_CPUID_SIGNATURE: {
399 char signature[12] = "KVMKVMKVM\0\0";
400 u32 *sigptr = (u32 *)signature;
401 entry->eax = KVM_CPUID_FEATURES;
402 entry->ebx = sigptr[0];
403 entry->ecx = sigptr[1];
404 entry->edx = sigptr[2];
407 case KVM_CPUID_FEATURES:
408 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
409 (1 << KVM_FEATURE_NOP_IO_DELAY) |
410 (1 << KVM_FEATURE_CLOCKSOURCE2) |
411 (1 << KVM_FEATURE_ASYNC_PF) |
412 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
415 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
422 entry->eax = min(entry->eax, 0x8000001a);
425 entry->edx &= kvm_supported_word1_x86_features;
426 cpuid_mask(&entry->edx, 1);
427 entry->ecx &= kvm_supported_word6_x86_features;
428 cpuid_mask(&entry->ecx, 6);
431 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
432 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
433 unsigned phys_as = entry->eax & 0xff;
437 entry->eax = g_phys_as | (virt_as << 8);
438 entry->ebx = entry->edx = 0;
442 entry->ecx = entry->edx = 0;
448 /*Add support for Centaur's CPUID instruction*/
450 /*Just support up to 0xC0000004 now*/
451 entry->eax = min(entry->eax, 0xC0000004);
454 entry->edx &= kvm_supported_word5_x86_features;
455 cpuid_mask(&entry->edx, 5);
457 case 3: /* Processor serial number */
458 case 5: /* MONITOR/MWAIT */
459 case 6: /* Thermal management */
460 case 0x80000007: /* Advanced power management */
465 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
469 kvm_x86_ops->set_supported_cpuid(function, entry);
481 struct kvm_cpuid_param {
485 bool (*qualifier)(struct kvm_cpuid_param *param);
488 static bool is_centaur_cpu(struct kvm_cpuid_param *param)
490 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
493 int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
494 struct kvm_cpuid_entry2 __user *entries)
496 struct kvm_cpuid_entry2 *cpuid_entries;
497 int limit, nent = 0, r = -E2BIG, i;
499 static struct kvm_cpuid_param param[] = {
500 { .func = 0, .has_leaf_count = true },
501 { .func = 0x80000000, .has_leaf_count = true },
502 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
503 { .func = KVM_CPUID_SIGNATURE },
504 { .func = KVM_CPUID_FEATURES },
509 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
510 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
512 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
517 for (i = 0; i < ARRAY_SIZE(param); i++) {
518 struct kvm_cpuid_param *ent = ¶m[i];
520 if (ent->qualifier && !ent->qualifier(ent))
523 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
529 if (!ent->has_leaf_count)
532 limit = cpuid_entries[nent - 1].eax;
533 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
534 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
542 if (copy_to_user(entries, cpuid_entries,
543 nent * sizeof(struct kvm_cpuid_entry2)))
549 vfree(cpuid_entries);
554 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
556 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
557 int j, nent = vcpu->arch.cpuid_nent;
559 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
560 /* when no next entry is found, the current entry[i] is reselected */
561 for (j = i + 1; ; j = (j + 1) % nent) {
562 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
563 if (ej->function == e->function) {
564 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
568 return 0; /* silence gcc, even though control never reaches here */
571 /* find an entry with matching function, matching index (if needed), and that
572 * should be read next (if it's stateful) */
573 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
574 u32 function, u32 index)
576 if (e->function != function)
578 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
580 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
581 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
586 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
587 u32 function, u32 index)
590 struct kvm_cpuid_entry2 *best = NULL;
592 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
593 struct kvm_cpuid_entry2 *e;
595 e = &vcpu->arch.cpuid_entries[i];
596 if (is_matching_cpuid_entry(e, function, index)) {
597 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
598 move_to_next_stateful_cpuid_entry(vcpu, i);
605 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
607 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
609 struct kvm_cpuid_entry2 *best;
611 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
612 if (!best || best->eax < 0x80000008)
614 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
616 return best->eax & 0xff;
622 * If no match is found, check whether we exceed the vCPU's limit
623 * and return the content of the highest valid _standard_ leaf instead.
624 * This is to satisfy the CPUID specification.
626 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
627 u32 function, u32 index)
629 struct kvm_cpuid_entry2 *maxlevel;
631 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
632 if (!maxlevel || maxlevel->eax >= function)
634 if (function & 0x80000000) {
635 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
639 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
642 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
645 struct kvm_cpuid_entry2 *best;
647 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
648 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
649 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
650 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
651 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
652 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
653 best = kvm_find_cpuid_entry(vcpu, function, index);
656 best = check_cpuid_limit(vcpu, function, index);
659 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
660 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
661 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
662 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
664 kvm_x86_ops->skip_emulated_instruction(vcpu);
665 trace_kvm_cpuid(function,
666 kvm_register_read(vcpu, VCPU_REGS_RAX),
667 kvm_register_read(vcpu, VCPU_REGS_RBX),
668 kvm_register_read(vcpu, VCPU_REGS_RCX),
669 kvm_register_read(vcpu, VCPU_REGS_RDX));
671 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);