]>
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> | |
4504b5c9 | 19 | #include <asm/processor.h> |
4e241557 | 20 | #include <asm/fpu/internal.h> /* For use_eager_fpu. Ugh! */ |
00b27a3e | 21 | #include <asm/user.h> |
669ebabb | 22 | #include <asm/fpu/xstate.h> |
00b27a3e AK |
23 | #include "cpuid.h" |
24 | #include "lapic.h" | |
25 | #include "mmu.h" | |
26 | #include "trace.h" | |
474a5bb9 | 27 | #include "pmu.h" |
00b27a3e | 28 | |
412a3c41 | 29 | static u32 xstate_required_size(u64 xstate_bv, bool compacted) |
4344ee98 PB |
30 | { |
31 | int feature_bit = 0; | |
32 | u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
33 | ||
d91cab78 | 34 | xstate_bv &= XFEATURE_MASK_EXTEND; |
4344ee98 PB |
35 | while (xstate_bv) { |
36 | if (xstate_bv & 0x1) { | |
412a3c41 | 37 | u32 eax, ebx, ecx, edx, offset; |
4344ee98 | 38 | cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); |
412a3c41 PB |
39 | offset = compacted ? ret : ebx; |
40 | ret = max(ret, offset + eax); | |
4344ee98 PB |
41 | } |
42 | ||
43 | xstate_bv >>= 1; | |
44 | feature_bit++; | |
45 | } | |
46 | ||
47 | return ret; | |
48 | } | |
49 | ||
a87036ad PB |
50 | bool kvm_mpx_supported(void) |
51 | { | |
52 | return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)) | |
53 | && kvm_x86_ops->mpx_supported()); | |
54 | } | |
55 | EXPORT_SYMBOL_GPL(kvm_mpx_supported); | |
56 | ||
4ff41732 PB |
57 | u64 kvm_supported_xcr0(void) |
58 | { | |
59 | u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; | |
60 | ||
a87036ad | 61 | if (!kvm_mpx_supported()) |
d91cab78 | 62 | xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); |
4ff41732 PB |
63 | |
64 | return xcr0; | |
65 | } | |
66 | ||
5c404cab PB |
67 | #define F(x) bit(X86_FEATURE_##x) |
68 | ||
4504b5c9 LK |
69 | /* These are scattered features in cpufeatures.h. */ |
70 | #define KVM_CPUID_BIT_AVX512_4VNNIW 2 | |
71 | #define KVM_CPUID_BIT_AVX512_4FMAPS 3 | |
72 | #define KF(x) bit(KVM_CPUID_BIT_##x) | |
73 | ||
dd598091 | 74 | int kvm_update_cpuid(struct kvm_vcpu *vcpu) |
00b27a3e AK |
75 | { |
76 | struct kvm_cpuid_entry2 *best; | |
77 | struct kvm_lapic *apic = vcpu->arch.apic; | |
78 | ||
79 | best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
80 | if (!best) | |
dd598091 | 81 | return 0; |
00b27a3e AK |
82 | |
83 | /* Update OSXSAVE bit */ | |
d366bf7e | 84 | if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) { |
5c404cab | 85 | best->ecx &= ~F(OSXSAVE); |
00b27a3e | 86 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) |
5c404cab | 87 | best->ecx |= F(OSXSAVE); |
00b27a3e AK |
88 | } |
89 | ||
c7dd15b3 JM |
90 | best->edx &= ~F(APIC); |
91 | if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE) | |
92 | best->edx |= F(APIC); | |
93 | ||
00b27a3e | 94 | if (apic) { |
5c404cab | 95 | if (best->ecx & F(TSC_DEADLINE_TIMER)) |
00b27a3e AK |
96 | apic->lapic_timer.timer_mode_mask = 3 << 17; |
97 | else | |
98 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
99 | } | |
f5132b01 | 100 | |
b9baba86 HH |
101 | best = kvm_find_cpuid_entry(vcpu, 7, 0); |
102 | if (best) { | |
103 | /* Update OSPKE bit */ | |
104 | if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) { | |
105 | best->ecx &= ~F(OSPKE); | |
106 | if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) | |
107 | best->ecx |= F(OSPKE); | |
108 | } | |
109 | } | |
110 | ||
d7876f1b | 111 | best = kvm_find_cpuid_entry(vcpu, 0xD, 0); |
4344ee98 | 112 | if (!best) { |
d7876f1b | 113 | vcpu->arch.guest_supported_xcr0 = 0; |
4344ee98 PB |
114 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
115 | } else { | |
d7876f1b PB |
116 | vcpu->arch.guest_supported_xcr0 = |
117 | (best->eax | ((u64)best->edx << 32)) & | |
4ff41732 | 118 | kvm_supported_xcr0(); |
56c103ec | 119 | vcpu->arch.guest_xstate_size = best->ebx = |
412a3c41 | 120 | xstate_required_size(vcpu->arch.xcr0, false); |
4344ee98 | 121 | } |
d7876f1b | 122 | |
412a3c41 PB |
123 | best = kvm_find_cpuid_entry(vcpu, 0xD, 1); |
124 | if (best && (best->eax & (F(XSAVES) | F(XSAVEC)))) | |
125 | best->ebx = xstate_required_size(vcpu->arch.xcr0, true); | |
126 | ||
5a5fbdc0 | 127 | if (use_eager_fpu()) |
370777da | 128 | kvm_x86_ops->fpu_activate(vcpu); |
c447e76b | 129 | |
dd598091 NA |
130 | /* |
131 | * The existing code assumes virtual address is 48-bit in the canonical | |
132 | * address checks; exit if it is ever changed. | |
133 | */ | |
134 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
135 | if (best && ((best->eax & 0xff00) >> 8) != 48 && | |
136 | ((best->eax & 0xff00) >> 8) != 0) | |
137 | return -EINVAL; | |
138 | ||
5a4f55cd EK |
139 | /* Update physical-address width */ |
140 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); | |
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) | |
612263b3 | 378 | F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) | |
8e3562f6 LK |
379 | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | |
380 | 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 HH |
386 | /* cpuid 7.0.ecx*/ |
387 | const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/; | |
388 | ||
4504b5c9 LK |
389 | /* cpuid 7.0.edx*/ |
390 | const u32 kvm_cpuid_7_0_edx_x86_features = | |
391 | KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS); | |
392 | ||
00b27a3e AK |
393 | /* all calls to cpuid_count() should be made on the same cpu */ |
394 | get_cpu(); | |
831bf664 SL |
395 | |
396 | r = -E2BIG; | |
397 | ||
398 | if (*nent >= maxnent) | |
399 | goto out; | |
400 | ||
00b27a3e AK |
401 | do_cpuid_1_ent(entry, function, index); |
402 | ++*nent; | |
403 | ||
404 | switch (function) { | |
405 | case 0: | |
406 | entry->eax = min(entry->eax, (u32)0xd); | |
407 | break; | |
408 | case 1: | |
e0b18ef7 HH |
409 | entry->edx &= kvm_cpuid_1_edx_x86_features; |
410 | cpuid_mask(&entry->edx, CPUID_1_EDX); | |
411 | entry->ecx &= kvm_cpuid_1_ecx_x86_features; | |
412 | cpuid_mask(&entry->ecx, CPUID_1_ECX); | |
00b27a3e AK |
413 | /* we support x2apic emulation even if host does not support |
414 | * it since we emulate x2apic in software */ | |
415 | entry->ecx |= F(X2APIC); | |
416 | break; | |
417 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
418 | * may return different values. This forces us to get_cpu() before | |
419 | * issuing the first command, and also to emulate this annoying behavior | |
420 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
421 | case 2: { | |
422 | int t, times = entry->eax & 0xff; | |
423 | ||
424 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
425 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
831bf664 SL |
426 | for (t = 1; t < times; ++t) { |
427 | if (*nent >= maxnent) | |
428 | goto out; | |
429 | ||
00b27a3e AK |
430 | do_cpuid_1_ent(&entry[t], function, 0); |
431 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
432 | ++*nent; | |
433 | } | |
434 | break; | |
435 | } | |
436 | /* function 4 has additional index. */ | |
437 | case 4: { | |
438 | int i, cache_type; | |
439 | ||
440 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
441 | /* read more entries until cache_type is zero */ | |
831bf664 SL |
442 | for (i = 1; ; ++i) { |
443 | if (*nent >= maxnent) | |
444 | goto out; | |
445 | ||
00b27a3e AK |
446 | cache_type = entry[i - 1].eax & 0x1f; |
447 | if (!cache_type) | |
448 | break; | |
449 | do_cpuid_1_ent(&entry[i], function, i); | |
450 | entry[i].flags |= | |
451 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
452 | ++*nent; | |
453 | } | |
454 | break; | |
455 | } | |
e453aa0f JK |
456 | case 6: /* Thermal management */ |
457 | entry->eax = 0x4; /* allow ARAT */ | |
458 | entry->ebx = 0; | |
459 | entry->ecx = 0; | |
460 | entry->edx = 0; | |
461 | break; | |
00b27a3e AK |
462 | case 7: { |
463 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
bbbda795 | 464 | /* Mask ebx against host capability word 9 */ |
00b27a3e | 465 | if (index == 0) { |
e0b18ef7 HH |
466 | entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; |
467 | cpuid_mask(&entry->ebx, CPUID_7_0_EBX); | |
ba904635 WA |
468 | // TSC_ADJUST is emulated |
469 | entry->ebx |= F(TSC_ADJUST); | |
b9baba86 HH |
470 | entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; |
471 | cpuid_mask(&entry->ecx, CPUID_7_ECX); | |
472 | /* PKU is not yet implemented for shadow paging. */ | |
473 | if (!tdp_enabled) | |
474 | entry->ecx &= ~F(PKU); | |
4504b5c9 LK |
475 | entry->edx &= kvm_cpuid_7_0_edx_x86_features; |
476 | entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX); | |
b9baba86 | 477 | } else { |
00b27a3e | 478 | entry->ebx = 0; |
b9baba86 | 479 | entry->ecx = 0; |
4504b5c9 | 480 | entry->edx = 0; |
b9baba86 | 481 | } |
00b27a3e | 482 | entry->eax = 0; |
00b27a3e AK |
483 | break; |
484 | } | |
485 | case 9: | |
486 | break; | |
a6c06ed1 GN |
487 | case 0xa: { /* Architectural Performance Monitoring */ |
488 | struct x86_pmu_capability cap; | |
489 | union cpuid10_eax eax; | |
490 | union cpuid10_edx edx; | |
491 | ||
492 | perf_get_x86_pmu_capability(&cap); | |
493 | ||
494 | /* | |
495 | * Only support guest architectural pmu on a host | |
496 | * with architectural pmu. | |
497 | */ | |
498 | if (!cap.version) | |
499 | memset(&cap, 0, sizeof(cap)); | |
500 | ||
501 | eax.split.version_id = min(cap.version, 2); | |
502 | eax.split.num_counters = cap.num_counters_gp; | |
503 | eax.split.bit_width = cap.bit_width_gp; | |
504 | eax.split.mask_length = cap.events_mask_len; | |
505 | ||
506 | edx.split.num_counters_fixed = cap.num_counters_fixed; | |
507 | edx.split.bit_width_fixed = cap.bit_width_fixed; | |
508 | edx.split.reserved = 0; | |
509 | ||
510 | entry->eax = eax.full; | |
511 | entry->ebx = cap.events_mask; | |
512 | entry->ecx = 0; | |
513 | entry->edx = edx.full; | |
514 | break; | |
515 | } | |
00b27a3e AK |
516 | /* function 0xb has additional index. */ |
517 | case 0xb: { | |
518 | int i, level_type; | |
519 | ||
520 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
521 | /* read more entries until level_type is zero */ | |
831bf664 SL |
522 | for (i = 1; ; ++i) { |
523 | if (*nent >= maxnent) | |
524 | goto out; | |
525 | ||
00b27a3e AK |
526 | level_type = entry[i - 1].ecx & 0xff00; |
527 | if (!level_type) | |
528 | break; | |
529 | do_cpuid_1_ent(&entry[i], function, i); | |
530 | entry[i].flags |= | |
531 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
532 | ++*nent; | |
533 | } | |
534 | break; | |
535 | } | |
536 | case 0xd: { | |
537 | int idx, i; | |
4ff41732 | 538 | u64 supported = kvm_supported_xcr0(); |
00b27a3e | 539 | |
4ff41732 | 540 | entry->eax &= supported; |
e08e8336 RK |
541 | entry->ebx = xstate_required_size(supported, false); |
542 | entry->ecx = entry->ebx; | |
4ff41732 | 543 | entry->edx &= supported >> 32; |
00b27a3e | 544 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
b65d6e17 PB |
545 | if (!supported) |
546 | break; | |
547 | ||
831bf664 | 548 | for (idx = 1, i = 1; idx < 64; ++idx) { |
4ff41732 | 549 | u64 mask = ((u64)1 << idx); |
831bf664 SL |
550 | if (*nent >= maxnent) |
551 | goto out; | |
552 | ||
00b27a3e | 553 | do_cpuid_1_ent(&entry[i], function, idx); |
412a3c41 | 554 | if (idx == 1) { |
e0b18ef7 | 555 | entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; |
316314ca | 556 | cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); |
412a3c41 PB |
557 | entry[i].ebx = 0; |
558 | if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) | |
559 | entry[i].ebx = | |
560 | xstate_required_size(supported, | |
561 | true); | |
404e0a19 PB |
562 | } else { |
563 | if (entry[i].eax == 0 || !(supported & mask)) | |
564 | continue; | |
565 | if (WARN_ON_ONCE(entry[i].ecx & 1)) | |
566 | continue; | |
567 | } | |
568 | entry[i].ecx = 0; | |
569 | entry[i].edx = 0; | |
00b27a3e AK |
570 | entry[i].flags |= |
571 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
572 | ++*nent; | |
573 | ++i; | |
574 | } | |
575 | break; | |
576 | } | |
577 | case KVM_CPUID_SIGNATURE: { | |
326d07cb MK |
578 | static const char signature[12] = "KVMKVMKVM\0\0"; |
579 | const u32 *sigptr = (const u32 *)signature; | |
57c22e5f | 580 | entry->eax = KVM_CPUID_FEATURES; |
00b27a3e AK |
581 | entry->ebx = sigptr[0]; |
582 | entry->ecx = sigptr[1]; | |
583 | entry->edx = sigptr[2]; | |
584 | break; | |
585 | } | |
586 | case KVM_CPUID_FEATURES: | |
587 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
588 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
589 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
590 | (1 << KVM_FEATURE_ASYNC_PF) | | |
ae7a2a3f | 591 | (1 << KVM_FEATURE_PV_EOI) | |
6aef266c SV |
592 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | |
593 | (1 << KVM_FEATURE_PV_UNHALT); | |
00b27a3e AK |
594 | |
595 | if (sched_info_on()) | |
596 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
597 | ||
598 | entry->ebx = 0; | |
599 | entry->ecx = 0; | |
600 | entry->edx = 0; | |
601 | break; | |
602 | case 0x80000000: | |
603 | entry->eax = min(entry->eax, 0x8000001a); | |
604 | break; | |
605 | case 0x80000001: | |
e0b18ef7 HH |
606 | entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; |
607 | cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); | |
608 | entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; | |
609 | cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); | |
00b27a3e | 610 | break; |
e4c9a5a1 MT |
611 | case 0x80000007: /* Advanced power management */ |
612 | /* invariant TSC is CPUID.80000007H:EDX[8] */ | |
613 | entry->edx &= (1 << 8); | |
614 | /* mask against host */ | |
615 | entry->edx &= boot_cpu_data.x86_power; | |
616 | entry->eax = entry->ebx = entry->ecx = 0; | |
617 | break; | |
00b27a3e AK |
618 | case 0x80000008: { |
619 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
620 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
621 | unsigned phys_as = entry->eax & 0xff; | |
622 | ||
623 | if (!g_phys_as) | |
624 | g_phys_as = phys_as; | |
625 | entry->eax = g_phys_as | (virt_as << 8); | |
626 | entry->ebx = entry->edx = 0; | |
627 | break; | |
628 | } | |
629 | case 0x80000019: | |
630 | entry->ecx = entry->edx = 0; | |
631 | break; | |
632 | case 0x8000001a: | |
633 | break; | |
634 | case 0x8000001d: | |
635 | break; | |
636 | /*Add support for Centaur's CPUID instruction*/ | |
637 | case 0xC0000000: | |
638 | /*Just support up to 0xC0000004 now*/ | |
639 | entry->eax = min(entry->eax, 0xC0000004); | |
640 | break; | |
641 | case 0xC0000001: | |
e0b18ef7 HH |
642 | entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; |
643 | cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); | |
00b27a3e AK |
644 | break; |
645 | case 3: /* Processor serial number */ | |
646 | case 5: /* MONITOR/MWAIT */ | |
00b27a3e AK |
647 | case 0xC0000002: |
648 | case 0xC0000003: | |
649 | case 0xC0000004: | |
650 | default: | |
651 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
652 | break; | |
653 | } | |
654 | ||
655 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
656 | ||
831bf664 SL |
657 | r = 0; |
658 | ||
659 | out: | |
00b27a3e | 660 | put_cpu(); |
831bf664 SL |
661 | |
662 | return r; | |
00b27a3e AK |
663 | } |
664 | ||
9c15bb1d BP |
665 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, |
666 | u32 idx, int *nent, int maxnent, unsigned int type) | |
667 | { | |
668 | if (type == KVM_GET_EMULATED_CPUID) | |
669 | return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); | |
670 | ||
671 | return __do_cpuid_ent(entry, func, idx, nent, maxnent); | |
672 | } | |
673 | ||
00b27a3e AK |
674 | #undef F |
675 | ||
831bf664 SL |
676 | struct kvm_cpuid_param { |
677 | u32 func; | |
678 | u32 idx; | |
679 | bool has_leaf_count; | |
326d07cb | 680 | bool (*qualifier)(const struct kvm_cpuid_param *param); |
831bf664 SL |
681 | }; |
682 | ||
326d07cb | 683 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) |
831bf664 SL |
684 | { |
685 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; | |
686 | } | |
687 | ||
9c15bb1d BP |
688 | static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, |
689 | __u32 num_entries, unsigned int ioctl_type) | |
690 | { | |
691 | int i; | |
1b2ca422 | 692 | __u32 pad[3]; |
9c15bb1d BP |
693 | |
694 | if (ioctl_type != KVM_GET_EMULATED_CPUID) | |
695 | return false; | |
696 | ||
697 | /* | |
698 | * We want to make sure that ->padding is being passed clean from | |
699 | * userspace in case we want to use it for something in the future. | |
700 | * | |
701 | * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we | |
702 | * have to give ourselves satisfied only with the emulated side. /me | |
703 | * sheds a tear. | |
704 | */ | |
705 | for (i = 0; i < num_entries; i++) { | |
1b2ca422 BP |
706 | if (copy_from_user(pad, entries[i].padding, sizeof(pad))) |
707 | return true; | |
708 | ||
709 | if (pad[0] || pad[1] || pad[2]) | |
9c15bb1d BP |
710 | return true; |
711 | } | |
712 | return false; | |
713 | } | |
714 | ||
715 | int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, | |
716 | struct kvm_cpuid_entry2 __user *entries, | |
717 | unsigned int type) | |
00b27a3e AK |
718 | { |
719 | struct kvm_cpuid_entry2 *cpuid_entries; | |
831bf664 | 720 | int limit, nent = 0, r = -E2BIG, i; |
00b27a3e | 721 | u32 func; |
326d07cb | 722 | static const struct kvm_cpuid_param param[] = { |
831bf664 SL |
723 | { .func = 0, .has_leaf_count = true }, |
724 | { .func = 0x80000000, .has_leaf_count = true }, | |
725 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, | |
726 | { .func = KVM_CPUID_SIGNATURE }, | |
727 | { .func = KVM_CPUID_FEATURES }, | |
728 | }; | |
00b27a3e AK |
729 | |
730 | if (cpuid->nent < 1) | |
731 | goto out; | |
732 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
733 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
9c15bb1d BP |
734 | |
735 | if (sanity_check_entries(entries, cpuid->nent, type)) | |
736 | return -EINVAL; | |
737 | ||
00b27a3e | 738 | r = -ENOMEM; |
84cffe49 | 739 | cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); |
00b27a3e AK |
740 | if (!cpuid_entries) |
741 | goto out; | |
742 | ||
831bf664 SL |
743 | r = 0; |
744 | for (i = 0; i < ARRAY_SIZE(param); i++) { | |
326d07cb | 745 | const struct kvm_cpuid_param *ent = ¶m[i]; |
00b27a3e | 746 | |
831bf664 SL |
747 | if (ent->qualifier && !ent->qualifier(ent)) |
748 | continue; | |
00b27a3e | 749 | |
831bf664 | 750 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, |
9c15bb1d | 751 | &nent, cpuid->nent, type); |
00b27a3e | 752 | |
831bf664 | 753 | if (r) |
00b27a3e AK |
754 | goto out_free; |
755 | ||
831bf664 SL |
756 | if (!ent->has_leaf_count) |
757 | continue; | |
758 | ||
00b27a3e | 759 | limit = cpuid_entries[nent - 1].eax; |
831bf664 SL |
760 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) |
761 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, | |
9c15bb1d | 762 | &nent, cpuid->nent, type); |
00b27a3e | 763 | |
831bf664 | 764 | if (r) |
00b27a3e AK |
765 | goto out_free; |
766 | } | |
767 | ||
00b27a3e AK |
768 | r = -EFAULT; |
769 | if (copy_to_user(entries, cpuid_entries, | |
770 | nent * sizeof(struct kvm_cpuid_entry2))) | |
771 | goto out_free; | |
772 | cpuid->nent = nent; | |
773 | r = 0; | |
774 | ||
775 | out_free: | |
776 | vfree(cpuid_entries); | |
777 | out: | |
778 | return r; | |
779 | } | |
780 | ||
781 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
782 | { | |
783 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
784 | int j, nent = vcpu->arch.cpuid_nent; | |
785 | ||
786 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
787 | /* when no next entry is found, the current entry[i] is reselected */ | |
788 | for (j = i + 1; ; j = (j + 1) % nent) { | |
789 | struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; | |
790 | if (ej->function == e->function) { | |
791 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
792 | return j; | |
793 | } | |
794 | } | |
795 | return 0; /* silence gcc, even though control never reaches here */ | |
796 | } | |
797 | ||
798 | /* find an entry with matching function, matching index (if needed), and that | |
799 | * should be read next (if it's stateful) */ | |
800 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
801 | u32 function, u32 index) | |
802 | { | |
803 | if (e->function != function) | |
804 | return 0; | |
805 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
806 | return 0; | |
807 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
808 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
809 | return 0; | |
810 | return 1; | |
811 | } | |
812 | ||
813 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
814 | u32 function, u32 index) | |
815 | { | |
816 | int i; | |
817 | struct kvm_cpuid_entry2 *best = NULL; | |
818 | ||
819 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
820 | struct kvm_cpuid_entry2 *e; | |
821 | ||
822 | e = &vcpu->arch.cpuid_entries[i]; | |
823 | if (is_matching_cpuid_entry(e, function, index)) { | |
824 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
825 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
826 | best = e; | |
827 | break; | |
828 | } | |
829 | } | |
830 | return best; | |
831 | } | |
832 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
833 | ||
00b27a3e AK |
834 | /* |
835 | * If no match is found, check whether we exceed the vCPU's limit | |
836 | * and return the content of the highest valid _standard_ leaf instead. | |
837 | * This is to satisfy the CPUID specification. | |
838 | */ | |
839 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
840 | u32 function, u32 index) | |
841 | { | |
842 | struct kvm_cpuid_entry2 *maxlevel; | |
843 | ||
844 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
845 | if (!maxlevel || maxlevel->eax >= function) | |
846 | return NULL; | |
847 | if (function & 0x80000000) { | |
848 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
849 | if (!maxlevel) | |
850 | return NULL; | |
851 | } | |
852 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
853 | } | |
854 | ||
62046e5a | 855 | void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
00b27a3e | 856 | { |
62046e5a | 857 | u32 function = *eax, index = *ecx; |
00b27a3e AK |
858 | struct kvm_cpuid_entry2 *best; |
859 | ||
00b27a3e AK |
860 | best = kvm_find_cpuid_entry(vcpu, function, index); |
861 | ||
862 | if (!best) | |
863 | best = check_cpuid_limit(vcpu, function, index); | |
864 | ||
bc613494 MT |
865 | /* |
866 | * Perfmon not yet supported for L2 guest. | |
867 | */ | |
868 | if (is_guest_mode(vcpu) && function == 0xa) | |
869 | best = NULL; | |
870 | ||
00b27a3e | 871 | if (best) { |
62046e5a AK |
872 | *eax = best->eax; |
873 | *ebx = best->ebx; | |
874 | *ecx = best->ecx; | |
875 | *edx = best->edx; | |
876 | } else | |
877 | *eax = *ebx = *ecx = *edx = 0; | |
a9d4e439 | 878 | trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx); |
62046e5a | 879 | } |
66f7b72e | 880 | EXPORT_SYMBOL_GPL(kvm_cpuid); |
62046e5a | 881 | |
6a908b62 | 882 | int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) |
62046e5a | 883 | { |
1e13175b | 884 | u32 eax, ebx, ecx, edx; |
62046e5a | 885 | |
1e13175b | 886 | eax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
62046e5a AK |
887 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); |
888 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx); | |
889 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); | |
890 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); | |
891 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); | |
892 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); | |
00b27a3e | 893 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
6a908b62 | 894 | return 1; |
00b27a3e AK |
895 | } |
896 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |