]>
Commit | Line | Data |
---|---|---|
00b27a3e AK |
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> | |
1767e931 | 16 | #include <linux/export.h> |
bb5a798a JK |
17 | #include <linux/vmalloc.h> |
18 | #include <linux/uaccess.h> | |
3905f9ad IM |
19 | #include <linux/sched/stat.h> |
20 | ||
4504b5c9 | 21 | #include <asm/processor.h> |
00b27a3e | 22 | #include <asm/user.h> |
669ebabb | 23 | #include <asm/fpu/xstate.h> |
00b27a3e AK |
24 | #include "cpuid.h" |
25 | #include "lapic.h" | |
26 | #include "mmu.h" | |
27 | #include "trace.h" | |
474a5bb9 | 28 | #include "pmu.h" |
00b27a3e | 29 | |
412a3c41 | 30 | static u32 xstate_required_size(u64 xstate_bv, bool compacted) |
4344ee98 PB |
31 | { |
32 | int feature_bit = 0; | |
33 | u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
34 | ||
d91cab78 | 35 | xstate_bv &= XFEATURE_MASK_EXTEND; |
4344ee98 PB |
36 | while (xstate_bv) { |
37 | if (xstate_bv & 0x1) { | |
412a3c41 | 38 | u32 eax, ebx, ecx, edx, offset; |
4344ee98 | 39 | cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); |
412a3c41 PB |
40 | offset = compacted ? ret : ebx; |
41 | ret = max(ret, offset + eax); | |
4344ee98 PB |
42 | } |
43 | ||
44 | xstate_bv >>= 1; | |
45 | feature_bit++; | |
46 | } | |
47 | ||
48 | return ret; | |
49 | } | |
50 | ||
a87036ad PB |
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 | ||
4ff41732 PB |
58 | u64 kvm_supported_xcr0(void) |
59 | { | |
60 | u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; | |
61 | ||
a87036ad | 62 | if (!kvm_mpx_supported()) |
d91cab78 | 63 | xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); |
4ff41732 PB |
64 | |
65 | return xcr0; | |
66 | } | |
67 | ||
5c404cab PB |
68 | #define F(x) bit(X86_FEATURE_##x) |
69 | ||
0fb3c779 | 70 | /* For scattered features from cpufeatures.h; we currently expose none */ |
4504b5c9 LK |
71 | #define KF(x) bit(KVM_CPUID_BIT_##x) |
72 | ||
dd598091 | 73 | int kvm_update_cpuid(struct kvm_vcpu *vcpu) |
00b27a3e AK |
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) | |
dd598091 | 80 | return 0; |
00b27a3e AK |
81 | |
82 | /* Update OSXSAVE bit */ | |
d366bf7e | 83 | if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) { |
5c404cab | 84 | best->ecx &= ~F(OSXSAVE); |
00b27a3e | 85 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) |
5c404cab | 86 | best->ecx |= F(OSXSAVE); |
00b27a3e AK |
87 | } |
88 | ||
c7dd15b3 JM |
89 | best->edx &= ~F(APIC); |
90 | if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE) | |
91 | best->edx |= F(APIC); | |
92 | ||
00b27a3e | 93 | if (apic) { |
5c404cab | 94 | if (best->ecx & F(TSC_DEADLINE_TIMER)) |
00b27a3e AK |
95 | apic->lapic_timer.timer_mode_mask = 3 << 17; |
96 | else | |
97 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
98 | } | |
f5132b01 | 99 | |
b9baba86 HH |
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 | ||
d7876f1b | 110 | best = kvm_find_cpuid_entry(vcpu, 0xD, 0); |
4344ee98 | 111 | if (!best) { |
d7876f1b | 112 | vcpu->arch.guest_supported_xcr0 = 0; |
4344ee98 PB |
113 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
114 | } else { | |
d7876f1b PB |
115 | vcpu->arch.guest_supported_xcr0 = |
116 | (best->eax | ((u64)best->edx << 32)) & | |
4ff41732 | 117 | kvm_supported_xcr0(); |
56c103ec | 118 | vcpu->arch.guest_xstate_size = best->ebx = |
412a3c41 | 119 | xstate_required_size(vcpu->arch.xcr0, false); |
4344ee98 | 120 | } |
d7876f1b | 121 | |
412a3c41 PB |
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 | ||
dd598091 | 126 | /* |
fd8cb433 YZ |
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. | |
dd598091 NA |
129 | */ |
130 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
fd8cb433 YZ |
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 | } | |
dd598091 | 137 | |
5a4f55cd EK |
138 | /* Update physical-address width */ |
139 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); | |
855feb67 | 140 | kvm_mmu_reset_context(vcpu); |
5a4f55cd | 141 | |
c6702c9d | 142 | kvm_pmu_refresh(vcpu); |
dd598091 | 143 | return 0; |
00b27a3e AK |
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 | } | |
5c404cab PB |
167 | if (entry && (entry->edx & F(NX)) && !is_efer_nx()) { |
168 | entry->edx &= ~F(NX); | |
00b27a3e AK |
169 | printk(KERN_INFO "kvm: guest NX capability removed\n"); |
170 | } | |
171 | } | |
172 | ||
5a4f55cd EK |
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 | ||
00b27a3e AK |
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; | |
83676e92 | 194 | struct kvm_cpuid_entry *cpuid_entries = NULL; |
00b27a3e AK |
195 | |
196 | r = -E2BIG; | |
197 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
198 | goto out; | |
199 | r = -ENOMEM; | |
83676e92 PB |
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 | } | |
00b27a3e AK |
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); | |
00b27a3e AK |
224 | kvm_apic_set_version(vcpu); |
225 | kvm_x86_ops->cpuid_update(vcpu); | |
dd598091 | 226 | r = kvm_update_cpuid(vcpu); |
00b27a3e | 227 | |
00b27a3e | 228 | out: |
83676e92 | 229 | vfree(cpuid_entries); |
00b27a3e AK |
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); | |
dd598091 | 249 | r = kvm_update_cpuid(vcpu); |
00b27a3e AK |
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 | ||
9c15bb1d BP |
289 | static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, |
290 | u32 func, u32 index, int *nent, int maxnent) | |
291 | { | |
84cffe49 BP |
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 | ||
9c15bb1d BP |
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) | |
00b27a3e | 313 | { |
831bf664 | 314 | int r; |
00b27a3e AK |
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; | |
ad756a16 | 325 | unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; |
a87036ad | 326 | unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0; |
55412b2e | 327 | unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; |
00b27a3e AK |
328 | |
329 | /* cpuid 1.edx */ | |
e0b18ef7 | 330 | const u32 kvm_cpuid_1_edx_x86_features = |
00b27a3e AK |
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) | | |
840d2830 | 335 | F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | |
00b27a3e AK |
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 */ | |
e0b18ef7 | 340 | const u32 kvm_cpuid_8000_0001_edx_x86_features = |
00b27a3e AK |
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 */ | |
e0b18ef7 | 350 | const u32 kvm_cpuid_1_ecx_x86_features = |
87c00572 GS |
351 | /* NOTE: MONITOR (and MWAIT) are emulated as NOP, |
352 | * but *not* advertised to guests via CPUID ! */ | |
00b27a3e AK |
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 */ | | |
fb215366 | 356 | F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | |
ad756a16 | 357 | F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | |
00b27a3e AK |
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 */ | |
e0b18ef7 | 362 | const u32 kvm_cpuid_8000_0001_ecx_x86_features = |
00b27a3e AK |
363 | F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | |
364 | F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | | |
2b036c6b | 365 | F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | |
00b27a3e AK |
366 | 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); |
367 | ||
368 | /* cpuid 0xC0000001.edx */ | |
e0b18ef7 | 369 | const u32 kvm_cpuid_C000_0001_edx_x86_features = |
00b27a3e AK |
370 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | |
371 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | | |
372 | F(PMM) | F(PMM_EN); | |
373 | ||
374 | /* cpuid 7.0.ebx */ | |
e0b18ef7 | 375 | const u32 kvm_cpuid_7_0_ebx_x86_features = |
83c52915 | 376 | F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | |
390bd528 | 377 | F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | |
83781d18 YS |
378 | F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) | |
379 | F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | | |
380 | F(SHA_NI) | F(AVX512BW) | F(AVX512VL); | |
00b27a3e | 381 | |
b65d6e17 | 382 | /* cpuid 0xD.1.eax */ |
e0b18ef7 | 383 | const u32 kvm_cpuid_D_1_eax_x86_features = |
55412b2e | 384 | F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; |
b65d6e17 | 385 | |
b9baba86 | 386 | /* cpuid 7.0.ecx*/ |
83781d18 | 387 | const u32 kvm_cpuid_7_0_ecx_x86_features = |
fd8cb433 YZ |
388 | F(AVX512VBMI) | F(LA57) | F(PKU) | |
389 | 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ); | |
b9baba86 | 390 | |
4504b5c9 LK |
391 | /* cpuid 7.0.edx*/ |
392 | const u32 kvm_cpuid_7_0_edx_x86_features = | |
0fb3c779 | 393 | F(AVX512_4VNNIW) | F(AVX512_4FMAPS); |
4504b5c9 | 394 | |
00b27a3e AK |
395 | /* all calls to cpuid_count() should be made on the same cpu */ |
396 | get_cpu(); | |
831bf664 SL |
397 | |
398 | r = -E2BIG; | |
399 | ||
400 | if (*nent >= maxnent) | |
401 | goto out; | |
402 | ||
00b27a3e AK |
403 | do_cpuid_1_ent(entry, function, index); |
404 | ++*nent; | |
405 | ||
406 | switch (function) { | |
407 | case 0: | |
408 | entry->eax = min(entry->eax, (u32)0xd); | |
409 | break; | |
410 | case 1: | |
e0b18ef7 HH |
411 | entry->edx &= kvm_cpuid_1_edx_x86_features; |
412 | cpuid_mask(&entry->edx, CPUID_1_EDX); | |
413 | entry->ecx &= kvm_cpuid_1_ecx_x86_features; | |
414 | cpuid_mask(&entry->ecx, CPUID_1_ECX); | |
00b27a3e AK |
415 | /* we support x2apic emulation even if host does not support |
416 | * it since we emulate x2apic in software */ | |
417 | entry->ecx |= F(X2APIC); | |
418 | break; | |
419 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
420 | * may return different values. This forces us to get_cpu() before | |
421 | * issuing the first command, and also to emulate this annoying behavior | |
422 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
423 | case 2: { | |
424 | int t, times = entry->eax & 0xff; | |
425 | ||
426 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
427 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
831bf664 SL |
428 | for (t = 1; t < times; ++t) { |
429 | if (*nent >= maxnent) | |
430 | goto out; | |
431 | ||
00b27a3e AK |
432 | do_cpuid_1_ent(&entry[t], function, 0); |
433 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
434 | ++*nent; | |
435 | } | |
436 | break; | |
437 | } | |
438 | /* function 4 has additional index. */ | |
439 | case 4: { | |
440 | int i, cache_type; | |
441 | ||
442 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
443 | /* read more entries until cache_type is zero */ | |
831bf664 SL |
444 | for (i = 1; ; ++i) { |
445 | if (*nent >= maxnent) | |
446 | goto out; | |
447 | ||
00b27a3e AK |
448 | cache_type = entry[i - 1].eax & 0x1f; |
449 | if (!cache_type) | |
450 | break; | |
451 | do_cpuid_1_ent(&entry[i], function, i); | |
452 | entry[i].flags |= | |
453 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
454 | ++*nent; | |
455 | } | |
456 | break; | |
457 | } | |
e453aa0f JK |
458 | case 6: /* Thermal management */ |
459 | entry->eax = 0x4; /* allow ARAT */ | |
460 | entry->ebx = 0; | |
461 | entry->ecx = 0; | |
462 | entry->edx = 0; | |
463 | break; | |
00b27a3e AK |
464 | case 7: { |
465 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
bbbda795 | 466 | /* Mask ebx against host capability word 9 */ |
00b27a3e | 467 | if (index == 0) { |
e0b18ef7 HH |
468 | entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; |
469 | cpuid_mask(&entry->ebx, CPUID_7_0_EBX); | |
ba904635 WA |
470 | // TSC_ADJUST is emulated |
471 | entry->ebx |= F(TSC_ADJUST); | |
b9baba86 HH |
472 | entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; |
473 | cpuid_mask(&entry->ecx, CPUID_7_ECX); | |
474 | /* PKU is not yet implemented for shadow paging. */ | |
c469268c | 475 | if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE)) |
b9baba86 | 476 | entry->ecx &= ~F(PKU); |
4504b5c9 | 477 | entry->edx &= kvm_cpuid_7_0_edx_x86_features; |
0fb3c779 | 478 | cpuid_mask(&entry->edx, CPUID_7_EDX); |
b9baba86 | 479 | } else { |
00b27a3e | 480 | entry->ebx = 0; |
b9baba86 | 481 | entry->ecx = 0; |
4504b5c9 | 482 | entry->edx = 0; |
b9baba86 | 483 | } |
00b27a3e | 484 | entry->eax = 0; |
00b27a3e AK |
485 | break; |
486 | } | |
487 | case 9: | |
488 | break; | |
a6c06ed1 GN |
489 | case 0xa: { /* Architectural Performance Monitoring */ |
490 | struct x86_pmu_capability cap; | |
491 | union cpuid10_eax eax; | |
492 | union cpuid10_edx edx; | |
493 | ||
494 | perf_get_x86_pmu_capability(&cap); | |
495 | ||
496 | /* | |
497 | * Only support guest architectural pmu on a host | |
498 | * with architectural pmu. | |
499 | */ | |
500 | if (!cap.version) | |
501 | memset(&cap, 0, sizeof(cap)); | |
502 | ||
503 | eax.split.version_id = min(cap.version, 2); | |
504 | eax.split.num_counters = cap.num_counters_gp; | |
505 | eax.split.bit_width = cap.bit_width_gp; | |
506 | eax.split.mask_length = cap.events_mask_len; | |
507 | ||
508 | edx.split.num_counters_fixed = cap.num_counters_fixed; | |
509 | edx.split.bit_width_fixed = cap.bit_width_fixed; | |
510 | edx.split.reserved = 0; | |
511 | ||
512 | entry->eax = eax.full; | |
513 | entry->ebx = cap.events_mask; | |
514 | entry->ecx = 0; | |
515 | entry->edx = edx.full; | |
516 | break; | |
517 | } | |
00b27a3e AK |
518 | /* function 0xb has additional index. */ |
519 | case 0xb: { | |
520 | int i, level_type; | |
521 | ||
522 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
523 | /* read more entries until level_type is zero */ | |
831bf664 SL |
524 | for (i = 1; ; ++i) { |
525 | if (*nent >= maxnent) | |
526 | goto out; | |
527 | ||
00b27a3e AK |
528 | level_type = entry[i - 1].ecx & 0xff00; |
529 | if (!level_type) | |
530 | break; | |
531 | do_cpuid_1_ent(&entry[i], function, i); | |
532 | entry[i].flags |= | |
533 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
534 | ++*nent; | |
535 | } | |
536 | break; | |
537 | } | |
538 | case 0xd: { | |
539 | int idx, i; | |
4ff41732 | 540 | u64 supported = kvm_supported_xcr0(); |
00b27a3e | 541 | |
4ff41732 | 542 | entry->eax &= supported; |
e08e8336 RK |
543 | entry->ebx = xstate_required_size(supported, false); |
544 | entry->ecx = entry->ebx; | |
4ff41732 | 545 | entry->edx &= supported >> 32; |
00b27a3e | 546 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
b65d6e17 PB |
547 | if (!supported) |
548 | break; | |
549 | ||
831bf664 | 550 | for (idx = 1, i = 1; idx < 64; ++idx) { |
4ff41732 | 551 | u64 mask = ((u64)1 << idx); |
831bf664 SL |
552 | if (*nent >= maxnent) |
553 | goto out; | |
554 | ||
00b27a3e | 555 | do_cpuid_1_ent(&entry[i], function, idx); |
412a3c41 | 556 | if (idx == 1) { |
e0b18ef7 | 557 | entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; |
316314ca | 558 | cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); |
412a3c41 PB |
559 | entry[i].ebx = 0; |
560 | if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) | |
561 | entry[i].ebx = | |
562 | xstate_required_size(supported, | |
563 | true); | |
404e0a19 PB |
564 | } else { |
565 | if (entry[i].eax == 0 || !(supported & mask)) | |
566 | continue; | |
567 | if (WARN_ON_ONCE(entry[i].ecx & 1)) | |
568 | continue; | |
569 | } | |
570 | entry[i].ecx = 0; | |
571 | entry[i].edx = 0; | |
00b27a3e AK |
572 | entry[i].flags |= |
573 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
574 | ++*nent; | |
575 | ++i; | |
576 | } | |
577 | break; | |
578 | } | |
579 | case KVM_CPUID_SIGNATURE: { | |
326d07cb MK |
580 | static const char signature[12] = "KVMKVMKVM\0\0"; |
581 | const u32 *sigptr = (const u32 *)signature; | |
57c22e5f | 582 | entry->eax = KVM_CPUID_FEATURES; |
00b27a3e AK |
583 | entry->ebx = sigptr[0]; |
584 | entry->ecx = sigptr[1]; | |
585 | entry->edx = sigptr[2]; | |
586 | break; | |
587 | } | |
588 | case KVM_CPUID_FEATURES: | |
589 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
590 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
591 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
592 | (1 << KVM_FEATURE_ASYNC_PF) | | |
ae7a2a3f | 593 | (1 << KVM_FEATURE_PV_EOI) | |
6aef266c SV |
594 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | |
595 | (1 << KVM_FEATURE_PV_UNHALT); | |
00b27a3e AK |
596 | |
597 | if (sched_info_on()) | |
598 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
599 | ||
600 | entry->ebx = 0; | |
601 | entry->ecx = 0; | |
602 | entry->edx = 0; | |
603 | break; | |
604 | case 0x80000000: | |
605 | entry->eax = min(entry->eax, 0x8000001a); | |
606 | break; | |
607 | case 0x80000001: | |
e0b18ef7 HH |
608 | entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; |
609 | cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); | |
610 | entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; | |
611 | cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); | |
00b27a3e | 612 | break; |
e4c9a5a1 MT |
613 | case 0x80000007: /* Advanced power management */ |
614 | /* invariant TSC is CPUID.80000007H:EDX[8] */ | |
615 | entry->edx &= (1 << 8); | |
616 | /* mask against host */ | |
617 | entry->edx &= boot_cpu_data.x86_power; | |
618 | entry->eax = entry->ebx = entry->ecx = 0; | |
619 | break; | |
00b27a3e AK |
620 | case 0x80000008: { |
621 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
622 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
623 | unsigned phys_as = entry->eax & 0xff; | |
624 | ||
625 | if (!g_phys_as) | |
626 | g_phys_as = phys_as; | |
627 | entry->eax = g_phys_as | (virt_as << 8); | |
628 | entry->ebx = entry->edx = 0; | |
629 | break; | |
630 | } | |
631 | case 0x80000019: | |
632 | entry->ecx = entry->edx = 0; | |
633 | break; | |
634 | case 0x8000001a: | |
635 | break; | |
636 | case 0x8000001d: | |
637 | break; | |
638 | /*Add support for Centaur's CPUID instruction*/ | |
639 | case 0xC0000000: | |
640 | /*Just support up to 0xC0000004 now*/ | |
641 | entry->eax = min(entry->eax, 0xC0000004); | |
642 | break; | |
643 | case 0xC0000001: | |
e0b18ef7 HH |
644 | entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; |
645 | cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); | |
00b27a3e AK |
646 | break; |
647 | case 3: /* Processor serial number */ | |
648 | case 5: /* MONITOR/MWAIT */ | |
00b27a3e AK |
649 | case 0xC0000002: |
650 | case 0xC0000003: | |
651 | case 0xC0000004: | |
652 | default: | |
653 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
654 | break; | |
655 | } | |
656 | ||
657 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
658 | ||
831bf664 SL |
659 | r = 0; |
660 | ||
661 | out: | |
00b27a3e | 662 | put_cpu(); |
831bf664 SL |
663 | |
664 | return r; | |
00b27a3e AK |
665 | } |
666 | ||
9c15bb1d BP |
667 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, |
668 | u32 idx, int *nent, int maxnent, unsigned int type) | |
669 | { | |
670 | if (type == KVM_GET_EMULATED_CPUID) | |
671 | return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); | |
672 | ||
673 | return __do_cpuid_ent(entry, func, idx, nent, maxnent); | |
674 | } | |
675 | ||
00b27a3e AK |
676 | #undef F |
677 | ||
831bf664 SL |
678 | struct kvm_cpuid_param { |
679 | u32 func; | |
680 | u32 idx; | |
681 | bool has_leaf_count; | |
326d07cb | 682 | bool (*qualifier)(const struct kvm_cpuid_param *param); |
831bf664 SL |
683 | }; |
684 | ||
326d07cb | 685 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) |
831bf664 SL |
686 | { |
687 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; | |
688 | } | |
689 | ||
9c15bb1d BP |
690 | static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, |
691 | __u32 num_entries, unsigned int ioctl_type) | |
692 | { | |
693 | int i; | |
1b2ca422 | 694 | __u32 pad[3]; |
9c15bb1d BP |
695 | |
696 | if (ioctl_type != KVM_GET_EMULATED_CPUID) | |
697 | return false; | |
698 | ||
699 | /* | |
700 | * We want to make sure that ->padding is being passed clean from | |
701 | * userspace in case we want to use it for something in the future. | |
702 | * | |
703 | * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we | |
704 | * have to give ourselves satisfied only with the emulated side. /me | |
705 | * sheds a tear. | |
706 | */ | |
707 | for (i = 0; i < num_entries; i++) { | |
1b2ca422 BP |
708 | if (copy_from_user(pad, entries[i].padding, sizeof(pad))) |
709 | return true; | |
710 | ||
711 | if (pad[0] || pad[1] || pad[2]) | |
9c15bb1d BP |
712 | return true; |
713 | } | |
714 | return false; | |
715 | } | |
716 | ||
717 | int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, | |
718 | struct kvm_cpuid_entry2 __user *entries, | |
719 | unsigned int type) | |
00b27a3e AK |
720 | { |
721 | struct kvm_cpuid_entry2 *cpuid_entries; | |
831bf664 | 722 | int limit, nent = 0, r = -E2BIG, i; |
00b27a3e | 723 | u32 func; |
326d07cb | 724 | static const struct kvm_cpuid_param param[] = { |
831bf664 SL |
725 | { .func = 0, .has_leaf_count = true }, |
726 | { .func = 0x80000000, .has_leaf_count = true }, | |
727 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, | |
728 | { .func = KVM_CPUID_SIGNATURE }, | |
729 | { .func = KVM_CPUID_FEATURES }, | |
730 | }; | |
00b27a3e AK |
731 | |
732 | if (cpuid->nent < 1) | |
733 | goto out; | |
734 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
735 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
9c15bb1d BP |
736 | |
737 | if (sanity_check_entries(entries, cpuid->nent, type)) | |
738 | return -EINVAL; | |
739 | ||
00b27a3e | 740 | r = -ENOMEM; |
84cffe49 | 741 | cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); |
00b27a3e AK |
742 | if (!cpuid_entries) |
743 | goto out; | |
744 | ||
831bf664 SL |
745 | r = 0; |
746 | for (i = 0; i < ARRAY_SIZE(param); i++) { | |
326d07cb | 747 | const struct kvm_cpuid_param *ent = ¶m[i]; |
00b27a3e | 748 | |
831bf664 SL |
749 | if (ent->qualifier && !ent->qualifier(ent)) |
750 | continue; | |
00b27a3e | 751 | |
831bf664 | 752 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, |
9c15bb1d | 753 | &nent, cpuid->nent, type); |
00b27a3e | 754 | |
831bf664 | 755 | if (r) |
00b27a3e AK |
756 | goto out_free; |
757 | ||
831bf664 SL |
758 | if (!ent->has_leaf_count) |
759 | continue; | |
760 | ||
00b27a3e | 761 | limit = cpuid_entries[nent - 1].eax; |
831bf664 SL |
762 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) |
763 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, | |
9c15bb1d | 764 | &nent, cpuid->nent, type); |
00b27a3e | 765 | |
831bf664 | 766 | if (r) |
00b27a3e AK |
767 | goto out_free; |
768 | } | |
769 | ||
00b27a3e AK |
770 | r = -EFAULT; |
771 | if (copy_to_user(entries, cpuid_entries, | |
772 | nent * sizeof(struct kvm_cpuid_entry2))) | |
773 | goto out_free; | |
774 | cpuid->nent = nent; | |
775 | r = 0; | |
776 | ||
777 | out_free: | |
778 | vfree(cpuid_entries); | |
779 | out: | |
780 | return r; | |
781 | } | |
782 | ||
783 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
784 | { | |
785 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
a3641631 WL |
786 | struct kvm_cpuid_entry2 *ej; |
787 | int j = i; | |
788 | int nent = vcpu->arch.cpuid_nent; | |
00b27a3e AK |
789 | |
790 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
791 | /* when no next entry is found, the current entry[i] is reselected */ | |
a3641631 WL |
792 | do { |
793 | j = (j + 1) % nent; | |
794 | ej = &vcpu->arch.cpuid_entries[j]; | |
795 | } while (ej->function != e->function); | |
796 | ||
797 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
798 | ||
799 | return j; | |
00b27a3e AK |
800 | } |
801 | ||
802 | /* find an entry with matching function, matching index (if needed), and that | |
803 | * should be read next (if it's stateful) */ | |
804 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
805 | u32 function, u32 index) | |
806 | { | |
807 | if (e->function != function) | |
808 | return 0; | |
809 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
810 | return 0; | |
811 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
812 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
813 | return 0; | |
814 | return 1; | |
815 | } | |
816 | ||
817 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
818 | u32 function, u32 index) | |
819 | { | |
820 | int i; | |
821 | struct kvm_cpuid_entry2 *best = NULL; | |
822 | ||
823 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
824 | struct kvm_cpuid_entry2 *e; | |
825 | ||
826 | e = &vcpu->arch.cpuid_entries[i]; | |
827 | if (is_matching_cpuid_entry(e, function, index)) { | |
828 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
829 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
830 | best = e; | |
831 | break; | |
832 | } | |
833 | } | |
834 | return best; | |
835 | } | |
836 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
837 | ||
00b27a3e AK |
838 | /* |
839 | * If no match is found, check whether we exceed the vCPU's limit | |
840 | * and return the content of the highest valid _standard_ leaf instead. | |
841 | * This is to satisfy the CPUID specification. | |
842 | */ | |
843 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
844 | u32 function, u32 index) | |
845 | { | |
846 | struct kvm_cpuid_entry2 *maxlevel; | |
847 | ||
848 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
849 | if (!maxlevel || maxlevel->eax >= function) | |
850 | return NULL; | |
851 | if (function & 0x80000000) { | |
852 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
853 | if (!maxlevel) | |
854 | return NULL; | |
855 | } | |
856 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
857 | } | |
858 | ||
e911eb3b YZ |
859 | bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, |
860 | u32 *ecx, u32 *edx, bool check_limit) | |
00b27a3e | 861 | { |
62046e5a | 862 | u32 function = *eax, index = *ecx; |
00b27a3e | 863 | struct kvm_cpuid_entry2 *best; |
e911eb3b | 864 | bool entry_found = true; |
00b27a3e | 865 | |
00b27a3e AK |
866 | best = kvm_find_cpuid_entry(vcpu, function, index); |
867 | ||
e911eb3b YZ |
868 | if (!best) { |
869 | entry_found = false; | |
870 | if (!check_limit) | |
871 | goto out; | |
872 | ||
00b27a3e | 873 | best = check_cpuid_limit(vcpu, function, index); |
e911eb3b | 874 | } |
00b27a3e | 875 | |
e911eb3b | 876 | out: |
00b27a3e | 877 | if (best) { |
62046e5a AK |
878 | *eax = best->eax; |
879 | *ebx = best->ebx; | |
880 | *ecx = best->ecx; | |
881 | *edx = best->edx; | |
882 | } else | |
883 | *eax = *ebx = *ecx = *edx = 0; | |
e911eb3b YZ |
884 | trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found); |
885 | return entry_found; | |
62046e5a | 886 | } |
66f7b72e | 887 | EXPORT_SYMBOL_GPL(kvm_cpuid); |
62046e5a | 888 | |
6a908b62 | 889 | int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) |
62046e5a | 890 | { |
1e13175b | 891 | u32 eax, ebx, ecx, edx; |
62046e5a | 892 | |
db2336a8 KH |
893 | if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0)) |
894 | return 1; | |
895 | ||
1e13175b | 896 | eax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
62046e5a | 897 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); |
e911eb3b | 898 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true); |
62046e5a AK |
899 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); |
900 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); | |
901 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); | |
902 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); | |
6affcbed | 903 | return kvm_skip_emulated_instruction(vcpu); |
00b27a3e AK |
904 | } |
905 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |