]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * cpuid support routines | |
4 | * | |
5 | * derived from arch/x86/kvm/x86.c | |
6 | * | |
7 | * Copyright 2011 Red Hat, Inc. and/or its affiliates. | |
8 | * Copyright IBM Corporation, 2008 | |
9 | * | |
10 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
11 | * the COPYING file in the top-level directory. | |
12 | * | |
13 | */ | |
14 | ||
15 | #include <linux/kvm_host.h> | |
16 | #include <linux/export.h> | |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/uaccess.h> | |
19 | #include <linux/sched/stat.h> | |
20 | ||
21 | #include <asm/processor.h> | |
22 | #include <asm/user.h> | |
23 | #include <asm/fpu/xstate.h> | |
24 | #include "cpuid.h" | |
25 | #include "lapic.h" | |
26 | #include "mmu.h" | |
27 | #include "trace.h" | |
28 | #include "pmu.h" | |
29 | ||
30 | static u32 xstate_required_size(u64 xstate_bv, bool compacted) | |
31 | { | |
32 | int feature_bit = 0; | |
33 | u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
34 | ||
35 | xstate_bv &= XFEATURE_MASK_EXTEND; | |
36 | while (xstate_bv) { | |
37 | if (xstate_bv & 0x1) { | |
38 | u32 eax, ebx, ecx, edx, offset; | |
39 | cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); | |
40 | offset = compacted ? ret : ebx; | |
41 | ret = max(ret, offset + eax); | |
42 | } | |
43 | ||
44 | xstate_bv >>= 1; | |
45 | feature_bit++; | |
46 | } | |
47 | ||
48 | return ret; | |
49 | } | |
50 | ||
51 | bool kvm_mpx_supported(void) | |
52 | { | |
53 | return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)) | |
54 | && kvm_x86_ops->mpx_supported()); | |
55 | } | |
56 | EXPORT_SYMBOL_GPL(kvm_mpx_supported); | |
57 | ||
58 | u64 kvm_supported_xcr0(void) | |
59 | { | |
60 | u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; | |
61 | ||
62 | if (!kvm_mpx_supported()) | |
63 | xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); | |
64 | ||
65 | return xcr0; | |
66 | } | |
67 | ||
68 | #define F(x) bit(X86_FEATURE_##x) | |
69 | ||
70 | /* For scattered features from cpufeatures.h; we currently expose none */ | |
71 | #define KF(x) bit(KVM_CPUID_BIT_##x) | |
72 | ||
73 | int kvm_update_cpuid(struct kvm_vcpu *vcpu) | |
74 | { | |
75 | struct kvm_cpuid_entry2 *best; | |
76 | struct kvm_lapic *apic = vcpu->arch.apic; | |
77 | ||
78 | best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
79 | if (!best) | |
80 | return 0; | |
81 | ||
82 | /* Update OSXSAVE bit */ | |
83 | if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) { | |
84 | best->ecx &= ~F(OSXSAVE); | |
85 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) | |
86 | best->ecx |= F(OSXSAVE); | |
87 | } | |
88 | ||
89 | best->edx &= ~F(APIC); | |
90 | if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE) | |
91 | best->edx |= F(APIC); | |
92 | ||
93 | if (apic) { | |
94 | if (best->ecx & F(TSC_DEADLINE_TIMER)) | |
95 | apic->lapic_timer.timer_mode_mask = 3 << 17; | |
96 | else | |
97 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
98 | } | |
99 | ||
100 | best = kvm_find_cpuid_entry(vcpu, 7, 0); | |
101 | if (best) { | |
102 | /* Update OSPKE bit */ | |
103 | if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) { | |
104 | best->ecx &= ~F(OSPKE); | |
105 | if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) | |
106 | best->ecx |= F(OSPKE); | |
107 | } | |
108 | } | |
109 | ||
110 | best = kvm_find_cpuid_entry(vcpu, 0xD, 0); | |
111 | if (!best) { | |
112 | vcpu->arch.guest_supported_xcr0 = 0; | |
113 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
114 | } else { | |
115 | vcpu->arch.guest_supported_xcr0 = | |
116 | (best->eax | ((u64)best->edx << 32)) & | |
117 | kvm_supported_xcr0(); | |
118 | vcpu->arch.guest_xstate_size = best->ebx = | |
119 | xstate_required_size(vcpu->arch.xcr0, false); | |
120 | } | |
121 | ||
122 | best = kvm_find_cpuid_entry(vcpu, 0xD, 1); | |
123 | if (best && (best->eax & (F(XSAVES) | F(XSAVEC)))) | |
124 | best->ebx = xstate_required_size(vcpu->arch.xcr0, true); | |
125 | ||
126 | /* | |
127 | * The existing code assumes virtual address is 48-bit or 57-bit in the | |
128 | * canonical address checks; exit if it is ever changed. | |
129 | */ | |
130 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
131 | if (best) { | |
132 | int vaddr_bits = (best->eax & 0xff00) >> 8; | |
133 | ||
134 | if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0) | |
135 | return -EINVAL; | |
136 | } | |
137 | ||
138 | /* Update physical-address width */ | |
139 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); | |
140 | kvm_mmu_reset_context(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 0x80000008.ebx */ | |
369 | const u32 kvm_cpuid_8000_0008_ebx_x86_features = | |
370 | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) | | |
371 | F(AMD_SSB_NO) | F(AMD_STIBP); | |
372 | ||
373 | /* cpuid 0xC0000001.edx */ | |
374 | const u32 kvm_cpuid_C000_0001_edx_x86_features = | |
375 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | | |
376 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | | |
377 | F(PMM) | F(PMM_EN); | |
378 | ||
379 | /* cpuid 7.0.ebx */ | |
380 | const u32 kvm_cpuid_7_0_ebx_x86_features = | |
381 | F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | | |
382 | F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | | |
383 | F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) | | |
384 | F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | | |
385 | F(SHA_NI) | F(AVX512BW) | F(AVX512VL); | |
386 | ||
387 | /* cpuid 0xD.1.eax */ | |
388 | const u32 kvm_cpuid_D_1_eax_x86_features = | |
389 | F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; | |
390 | ||
391 | /* cpuid 7.0.ecx*/ | |
392 | const u32 kvm_cpuid_7_0_ecx_x86_features = | |
393 | F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | | |
394 | F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) | | |
395 | F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG); | |
396 | ||
397 | /* cpuid 7.0.edx*/ | |
398 | const u32 kvm_cpuid_7_0_edx_x86_features = | |
399 | F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) | | |
400 | F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP); | |
401 | ||
402 | /* all calls to cpuid_count() should be made on the same cpu */ | |
403 | get_cpu(); | |
404 | ||
405 | r = -E2BIG; | |
406 | ||
407 | if (*nent >= maxnent) | |
408 | goto out; | |
409 | ||
410 | do_cpuid_1_ent(entry, function, index); | |
411 | ++*nent; | |
412 | ||
413 | switch (function) { | |
414 | case 0: | |
415 | entry->eax = min(entry->eax, (u32)0xd); | |
416 | break; | |
417 | case 1: | |
418 | entry->edx &= kvm_cpuid_1_edx_x86_features; | |
419 | cpuid_mask(&entry->edx, CPUID_1_EDX); | |
420 | entry->ecx &= kvm_cpuid_1_ecx_x86_features; | |
421 | cpuid_mask(&entry->ecx, CPUID_1_ECX); | |
422 | /* we support x2apic emulation even if host does not support | |
423 | * it since we emulate x2apic in software */ | |
424 | entry->ecx |= F(X2APIC); | |
425 | break; | |
426 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
427 | * may return different values. This forces us to get_cpu() before | |
428 | * issuing the first command, and also to emulate this annoying behavior | |
429 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
430 | case 2: { | |
431 | int t, times = entry->eax & 0xff; | |
432 | ||
433 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
434 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
435 | for (t = 1; t < times; ++t) { | |
436 | if (*nent >= maxnent) | |
437 | goto out; | |
438 | ||
439 | do_cpuid_1_ent(&entry[t], function, 0); | |
440 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
441 | ++*nent; | |
442 | } | |
443 | break; | |
444 | } | |
445 | /* function 4 has additional index. */ | |
446 | case 4: { | |
447 | int i, cache_type; | |
448 | ||
449 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
450 | /* read more entries until cache_type is zero */ | |
451 | for (i = 1; ; ++i) { | |
452 | if (*nent >= maxnent) | |
453 | goto out; | |
454 | ||
455 | cache_type = entry[i - 1].eax & 0x1f; | |
456 | if (!cache_type) | |
457 | break; | |
458 | do_cpuid_1_ent(&entry[i], function, i); | |
459 | entry[i].flags |= | |
460 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
461 | ++*nent; | |
462 | } | |
463 | break; | |
464 | } | |
465 | case 6: /* Thermal management */ | |
466 | entry->eax = 0x4; /* allow ARAT */ | |
467 | entry->ebx = 0; | |
468 | entry->ecx = 0; | |
469 | entry->edx = 0; | |
470 | break; | |
471 | case 7: { | |
472 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
473 | /* Mask ebx against host capability word 9 */ | |
474 | if (index == 0) { | |
475 | entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; | |
476 | cpuid_mask(&entry->ebx, CPUID_7_0_EBX); | |
477 | // TSC_ADJUST is emulated | |
478 | entry->ebx |= F(TSC_ADJUST); | |
479 | entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; | |
480 | cpuid_mask(&entry->ecx, CPUID_7_ECX); | |
481 | /* PKU is not yet implemented for shadow paging. */ | |
482 | if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE)) | |
483 | entry->ecx &= ~F(PKU); | |
484 | entry->edx &= kvm_cpuid_7_0_edx_x86_features; | |
485 | cpuid_mask(&entry->edx, CPUID_7_EDX); | |
486 | /* | |
487 | * We emulate ARCH_CAPABILITIES in software even | |
488 | * if the host doesn't support it. | |
489 | */ | |
490 | entry->edx |= F(ARCH_CAPABILITIES); | |
491 | } else { | |
492 | entry->ebx = 0; | |
493 | entry->ecx = 0; | |
494 | entry->edx = 0; | |
495 | } | |
496 | entry->eax = 0; | |
497 | break; | |
498 | } | |
499 | case 9: | |
500 | break; | |
501 | case 0xa: { /* Architectural Performance Monitoring */ | |
502 | struct x86_pmu_capability cap; | |
503 | union cpuid10_eax eax; | |
504 | union cpuid10_edx edx; | |
505 | ||
506 | perf_get_x86_pmu_capability(&cap); | |
507 | ||
508 | /* | |
509 | * Only support guest architectural pmu on a host | |
510 | * with architectural pmu. | |
511 | */ | |
512 | if (!cap.version) | |
513 | memset(&cap, 0, sizeof(cap)); | |
514 | ||
515 | eax.split.version_id = min(cap.version, 2); | |
516 | eax.split.num_counters = cap.num_counters_gp; | |
517 | eax.split.bit_width = cap.bit_width_gp; | |
518 | eax.split.mask_length = cap.events_mask_len; | |
519 | ||
520 | edx.split.num_counters_fixed = cap.num_counters_fixed; | |
521 | edx.split.bit_width_fixed = cap.bit_width_fixed; | |
522 | edx.split.reserved = 0; | |
523 | ||
524 | entry->eax = eax.full; | |
525 | entry->ebx = cap.events_mask; | |
526 | entry->ecx = 0; | |
527 | entry->edx = edx.full; | |
528 | break; | |
529 | } | |
530 | /* function 0xb has additional index. */ | |
531 | case 0xb: { | |
532 | int i, level_type; | |
533 | ||
534 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
535 | /* read more entries until level_type is zero */ | |
536 | for (i = 1; ; ++i) { | |
537 | if (*nent >= maxnent) | |
538 | goto out; | |
539 | ||
540 | level_type = entry[i - 1].ecx & 0xff00; | |
541 | if (!level_type) | |
542 | break; | |
543 | do_cpuid_1_ent(&entry[i], function, i); | |
544 | entry[i].flags |= | |
545 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
546 | ++*nent; | |
547 | } | |
548 | break; | |
549 | } | |
550 | case 0xd: { | |
551 | int idx, i; | |
552 | u64 supported = kvm_supported_xcr0(); | |
553 | ||
554 | entry->eax &= supported; | |
555 | entry->ebx = xstate_required_size(supported, false); | |
556 | entry->ecx = entry->ebx; | |
557 | entry->edx &= supported >> 32; | |
558 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
559 | if (!supported) | |
560 | break; | |
561 | ||
562 | for (idx = 1, i = 1; idx < 64; ++idx) { | |
563 | u64 mask = ((u64)1 << idx); | |
564 | if (*nent >= maxnent) | |
565 | goto out; | |
566 | ||
567 | do_cpuid_1_ent(&entry[i], function, idx); | |
568 | if (idx == 1) { | |
569 | entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; | |
570 | cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); | |
571 | entry[i].ebx = 0; | |
572 | if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) | |
573 | entry[i].ebx = | |
574 | xstate_required_size(supported, | |
575 | true); | |
576 | } else { | |
577 | if (entry[i].eax == 0 || !(supported & mask)) | |
578 | continue; | |
579 | if (WARN_ON_ONCE(entry[i].ecx & 1)) | |
580 | continue; | |
581 | } | |
582 | entry[i].ecx = 0; | |
583 | entry[i].edx = 0; | |
584 | entry[i].flags |= | |
585 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
586 | ++*nent; | |
587 | ++i; | |
588 | } | |
589 | break; | |
590 | } | |
591 | case KVM_CPUID_SIGNATURE: { | |
592 | static const char signature[12] = "KVMKVMKVM\0\0"; | |
593 | const u32 *sigptr = (const u32 *)signature; | |
594 | entry->eax = KVM_CPUID_FEATURES; | |
595 | entry->ebx = sigptr[0]; | |
596 | entry->ecx = sigptr[1]; | |
597 | entry->edx = sigptr[2]; | |
598 | break; | |
599 | } | |
600 | case KVM_CPUID_FEATURES: | |
601 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
602 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
603 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
604 | (1 << KVM_FEATURE_ASYNC_PF) | | |
605 | (1 << KVM_FEATURE_PV_EOI) | | |
606 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | | |
607 | (1 << KVM_FEATURE_PV_UNHALT) | | |
608 | (1 << KVM_FEATURE_ASYNC_PF_VMEXIT); | |
609 | ||
610 | if (sched_info_on()) | |
611 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
612 | ||
613 | entry->ebx = 0; | |
614 | entry->ecx = 0; | |
615 | entry->edx = 0; | |
616 | break; | |
617 | case 0x80000000: | |
618 | entry->eax = min(entry->eax, 0x8000001a); | |
619 | break; | |
620 | case 0x80000001: | |
621 | entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; | |
622 | cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); | |
623 | entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; | |
624 | cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); | |
625 | break; | |
626 | case 0x80000007: /* Advanced power management */ | |
627 | /* invariant TSC is CPUID.80000007H:EDX[8] */ | |
628 | entry->edx &= (1 << 8); | |
629 | /* mask against host */ | |
630 | entry->edx &= boot_cpu_data.x86_power; | |
631 | entry->eax = entry->ebx = entry->ecx = 0; | |
632 | break; | |
633 | case 0x80000008: { | |
634 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
635 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
636 | unsigned phys_as = entry->eax & 0xff; | |
637 | ||
638 | if (!g_phys_as) | |
639 | g_phys_as = phys_as; | |
640 | entry->eax = g_phys_as | (virt_as << 8); | |
641 | entry->edx = 0; | |
642 | /* | |
643 | * IBRS, IBPB and VIRT_SSBD aren't necessarily present in | |
644 | * hardware cpuid | |
645 | */ | |
646 | if (boot_cpu_has(X86_FEATURE_AMD_IBPB)) | |
647 | entry->ebx |= F(AMD_IBPB); | |
648 | if (boot_cpu_has(X86_FEATURE_AMD_IBRS)) | |
649 | entry->ebx |= F(AMD_IBRS); | |
650 | if (boot_cpu_has(X86_FEATURE_VIRT_SSBD)) | |
651 | entry->ebx |= F(VIRT_SSBD); | |
652 | entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features; | |
653 | cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX); | |
654 | /* | |
655 | * The preference is to use SPEC CTRL MSR instead of the | |
656 | * VIRT_SPEC MSR. | |
657 | */ | |
658 | if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) && | |
659 | !boot_cpu_has(X86_FEATURE_AMD_SSBD)) | |
660 | entry->ebx |= F(VIRT_SSBD); | |
661 | break; | |
662 | } | |
663 | case 0x80000019: | |
664 | entry->ecx = entry->edx = 0; | |
665 | break; | |
666 | case 0x8000001a: | |
667 | break; | |
668 | case 0x8000001d: | |
669 | break; | |
670 | /*Add support for Centaur's CPUID instruction*/ | |
671 | case 0xC0000000: | |
672 | /*Just support up to 0xC0000004 now*/ | |
673 | entry->eax = min(entry->eax, 0xC0000004); | |
674 | break; | |
675 | case 0xC0000001: | |
676 | entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; | |
677 | cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); | |
678 | break; | |
679 | case 3: /* Processor serial number */ | |
680 | case 5: /* MONITOR/MWAIT */ | |
681 | case 0xC0000002: | |
682 | case 0xC0000003: | |
683 | case 0xC0000004: | |
684 | default: | |
685 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
686 | break; | |
687 | } | |
688 | ||
689 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
690 | ||
691 | r = 0; | |
692 | ||
693 | out: | |
694 | put_cpu(); | |
695 | ||
696 | return r; | |
697 | } | |
698 | ||
699 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, | |
700 | u32 idx, int *nent, int maxnent, unsigned int type) | |
701 | { | |
702 | if (type == KVM_GET_EMULATED_CPUID) | |
703 | return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); | |
704 | ||
705 | return __do_cpuid_ent(entry, func, idx, nent, maxnent); | |
706 | } | |
707 | ||
708 | #undef F | |
709 | ||
710 | struct kvm_cpuid_param { | |
711 | u32 func; | |
712 | u32 idx; | |
713 | bool has_leaf_count; | |
714 | bool (*qualifier)(const struct kvm_cpuid_param *param); | |
715 | }; | |
716 | ||
717 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) | |
718 | { | |
719 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; | |
720 | } | |
721 | ||
722 | static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, | |
723 | __u32 num_entries, unsigned int ioctl_type) | |
724 | { | |
725 | int i; | |
726 | __u32 pad[3]; | |
727 | ||
728 | if (ioctl_type != KVM_GET_EMULATED_CPUID) | |
729 | return false; | |
730 | ||
731 | /* | |
732 | * We want to make sure that ->padding is being passed clean from | |
733 | * userspace in case we want to use it for something in the future. | |
734 | * | |
735 | * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we | |
736 | * have to give ourselves satisfied only with the emulated side. /me | |
737 | * sheds a tear. | |
738 | */ | |
739 | for (i = 0; i < num_entries; i++) { | |
740 | if (copy_from_user(pad, entries[i].padding, sizeof(pad))) | |
741 | return true; | |
742 | ||
743 | if (pad[0] || pad[1] || pad[2]) | |
744 | return true; | |
745 | } | |
746 | return false; | |
747 | } | |
748 | ||
749 | int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, | |
750 | struct kvm_cpuid_entry2 __user *entries, | |
751 | unsigned int type) | |
752 | { | |
753 | struct kvm_cpuid_entry2 *cpuid_entries; | |
754 | int limit, nent = 0, r = -E2BIG, i; | |
755 | u32 func; | |
756 | static const struct kvm_cpuid_param param[] = { | |
757 | { .func = 0, .has_leaf_count = true }, | |
758 | { .func = 0x80000000, .has_leaf_count = true }, | |
759 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, | |
760 | { .func = KVM_CPUID_SIGNATURE }, | |
761 | { .func = KVM_CPUID_FEATURES }, | |
762 | }; | |
763 | ||
764 | if (cpuid->nent < 1) | |
765 | goto out; | |
766 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
767 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
768 | ||
769 | if (sanity_check_entries(entries, cpuid->nent, type)) | |
770 | return -EINVAL; | |
771 | ||
772 | r = -ENOMEM; | |
773 | cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); | |
774 | if (!cpuid_entries) | |
775 | goto out; | |
776 | ||
777 | r = 0; | |
778 | for (i = 0; i < ARRAY_SIZE(param); i++) { | |
779 | const struct kvm_cpuid_param *ent = ¶m[i]; | |
780 | ||
781 | if (ent->qualifier && !ent->qualifier(ent)) | |
782 | continue; | |
783 | ||
784 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, | |
785 | &nent, cpuid->nent, type); | |
786 | ||
787 | if (r) | |
788 | goto out_free; | |
789 | ||
790 | if (!ent->has_leaf_count) | |
791 | continue; | |
792 | ||
793 | limit = cpuid_entries[nent - 1].eax; | |
794 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) | |
795 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, | |
796 | &nent, cpuid->nent, type); | |
797 | ||
798 | if (r) | |
799 | goto out_free; | |
800 | } | |
801 | ||
802 | r = -EFAULT; | |
803 | if (copy_to_user(entries, cpuid_entries, | |
804 | nent * sizeof(struct kvm_cpuid_entry2))) | |
805 | goto out_free; | |
806 | cpuid->nent = nent; | |
807 | r = 0; | |
808 | ||
809 | out_free: | |
810 | vfree(cpuid_entries); | |
811 | out: | |
812 | return r; | |
813 | } | |
814 | ||
815 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
816 | { | |
817 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
818 | struct kvm_cpuid_entry2 *ej; | |
819 | int j = i; | |
820 | int nent = vcpu->arch.cpuid_nent; | |
821 | ||
822 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
823 | /* when no next entry is found, the current entry[i] is reselected */ | |
824 | do { | |
825 | j = (j + 1) % nent; | |
826 | ej = &vcpu->arch.cpuid_entries[j]; | |
827 | } while (ej->function != e->function); | |
828 | ||
829 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
830 | ||
831 | return j; | |
832 | } | |
833 | ||
834 | /* find an entry with matching function, matching index (if needed), and that | |
835 | * should be read next (if it's stateful) */ | |
836 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
837 | u32 function, u32 index) | |
838 | { | |
839 | if (e->function != function) | |
840 | return 0; | |
841 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
842 | return 0; | |
843 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
844 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
845 | return 0; | |
846 | return 1; | |
847 | } | |
848 | ||
849 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
850 | u32 function, u32 index) | |
851 | { | |
852 | int i; | |
853 | struct kvm_cpuid_entry2 *best = NULL; | |
854 | ||
855 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
856 | struct kvm_cpuid_entry2 *e; | |
857 | ||
858 | e = &vcpu->arch.cpuid_entries[i]; | |
859 | if (is_matching_cpuid_entry(e, function, index)) { | |
860 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
861 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
862 | best = e; | |
863 | break; | |
864 | } | |
865 | } | |
866 | return best; | |
867 | } | |
868 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
869 | ||
870 | /* | |
871 | * If no match is found, check whether we exceed the vCPU's limit | |
872 | * and return the content of the highest valid _standard_ leaf instead. | |
873 | * This is to satisfy the CPUID specification. | |
874 | */ | |
875 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
876 | u32 function, u32 index) | |
877 | { | |
878 | struct kvm_cpuid_entry2 *maxlevel; | |
879 | ||
880 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
881 | if (!maxlevel || maxlevel->eax >= function) | |
882 | return NULL; | |
883 | if (function & 0x80000000) { | |
884 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
885 | if (!maxlevel) | |
886 | return NULL; | |
887 | } | |
888 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
889 | } | |
890 | ||
891 | bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, | |
892 | u32 *ecx, u32 *edx, bool check_limit) | |
893 | { | |
894 | u32 function = *eax, index = *ecx; | |
895 | struct kvm_cpuid_entry2 *best; | |
896 | bool entry_found = true; | |
897 | ||
898 | best = kvm_find_cpuid_entry(vcpu, function, index); | |
899 | ||
900 | if (!best) { | |
901 | entry_found = false; | |
902 | if (!check_limit) | |
903 | goto out; | |
904 | ||
905 | best = check_cpuid_limit(vcpu, function, index); | |
906 | } | |
907 | ||
908 | out: | |
909 | if (best) { | |
910 | *eax = best->eax; | |
911 | *ebx = best->ebx; | |
912 | *ecx = best->ecx; | |
913 | *edx = best->edx; | |
914 | } else | |
915 | *eax = *ebx = *ecx = *edx = 0; | |
916 | trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found); | |
917 | return entry_found; | |
918 | } | |
919 | EXPORT_SYMBOL_GPL(kvm_cpuid); | |
920 | ||
921 | int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) | |
922 | { | |
923 | u32 eax, ebx, ecx, edx; | |
924 | ||
925 | if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0)) | |
926 | return 1; | |
927 | ||
928 | eax = kvm_register_read(vcpu, VCPU_REGS_RAX); | |
929 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
930 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true); | |
931 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); | |
932 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); | |
933 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); | |
934 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); | |
935 | return kvm_skip_emulated_instruction(vcpu); | |
936 | } | |
937 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |