]>
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> | |
16 | #include <linux/module.h> | |
bb5a798a JK |
17 | #include <linux/vmalloc.h> |
18 | #include <linux/uaccess.h> | |
00b27a3e AK |
19 | #include <asm/user.h> |
20 | #include <asm/xsave.h> | |
21 | #include "cpuid.h" | |
22 | #include "lapic.h" | |
23 | #include "mmu.h" | |
24 | #include "trace.h" | |
25 | ||
4344ee98 PB |
26 | static u32 xstate_required_size(u64 xstate_bv) |
27 | { | |
28 | int feature_bit = 0; | |
29 | u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
30 | ||
56c103ec | 31 | xstate_bv &= XSTATE_EXTEND_MASK; |
4344ee98 PB |
32 | while (xstate_bv) { |
33 | if (xstate_bv & 0x1) { | |
34 | u32 eax, ebx, ecx, edx; | |
35 | cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); | |
36 | ret = max(ret, eax + ebx); | |
37 | } | |
38 | ||
39 | xstate_bv >>= 1; | |
40 | feature_bit++; | |
41 | } | |
42 | ||
43 | return ret; | |
44 | } | |
45 | ||
4ff41732 PB |
46 | u64 kvm_supported_xcr0(void) |
47 | { | |
48 | u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; | |
49 | ||
93c4adc7 | 50 | if (!kvm_x86_ops->mpx_supported()) |
4ff41732 PB |
51 | xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR); |
52 | ||
53 | return xcr0; | |
54 | } | |
55 | ||
dd598091 | 56 | int kvm_update_cpuid(struct kvm_vcpu *vcpu) |
00b27a3e AK |
57 | { |
58 | struct kvm_cpuid_entry2 *best; | |
59 | struct kvm_lapic *apic = vcpu->arch.apic; | |
60 | ||
61 | best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
62 | if (!best) | |
dd598091 | 63 | return 0; |
00b27a3e AK |
64 | |
65 | /* Update OSXSAVE bit */ | |
66 | if (cpu_has_xsave && best->function == 0x1) { | |
67 | best->ecx &= ~(bit(X86_FEATURE_OSXSAVE)); | |
68 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) | |
69 | best->ecx |= bit(X86_FEATURE_OSXSAVE); | |
70 | } | |
71 | ||
72 | if (apic) { | |
73 | if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER)) | |
74 | apic->lapic_timer.timer_mode_mask = 3 << 17; | |
75 | else | |
76 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
77 | } | |
f5132b01 | 78 | |
d7876f1b | 79 | best = kvm_find_cpuid_entry(vcpu, 0xD, 0); |
4344ee98 | 80 | if (!best) { |
d7876f1b | 81 | vcpu->arch.guest_supported_xcr0 = 0; |
4344ee98 PB |
82 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
83 | } else { | |
d7876f1b PB |
84 | vcpu->arch.guest_supported_xcr0 = |
85 | (best->eax | ((u64)best->edx << 32)) & | |
4ff41732 | 86 | kvm_supported_xcr0(); |
56c103ec LJ |
87 | vcpu->arch.guest_xstate_size = best->ebx = |
88 | xstate_required_size(vcpu->arch.xcr0); | |
4344ee98 | 89 | } |
d7876f1b | 90 | |
dd598091 NA |
91 | /* |
92 | * The existing code assumes virtual address is 48-bit in the canonical | |
93 | * address checks; exit if it is ever changed. | |
94 | */ | |
95 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
96 | if (best && ((best->eax & 0xff00) >> 8) != 48 && | |
97 | ((best->eax & 0xff00) >> 8) != 0) | |
98 | return -EINVAL; | |
99 | ||
f5132b01 | 100 | kvm_pmu_cpuid_update(vcpu); |
dd598091 | 101 | return 0; |
00b27a3e AK |
102 | } |
103 | ||
104 | static int is_efer_nx(void) | |
105 | { | |
106 | unsigned long long efer = 0; | |
107 | ||
108 | rdmsrl_safe(MSR_EFER, &efer); | |
109 | return efer & EFER_NX; | |
110 | } | |
111 | ||
112 | static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) | |
113 | { | |
114 | int i; | |
115 | struct kvm_cpuid_entry2 *e, *entry; | |
116 | ||
117 | entry = NULL; | |
118 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
119 | e = &vcpu->arch.cpuid_entries[i]; | |
120 | if (e->function == 0x80000001) { | |
121 | entry = e; | |
122 | break; | |
123 | } | |
124 | } | |
6689fbe3 NA |
125 | if (entry && (entry->edx & bit(X86_FEATURE_NX)) && !is_efer_nx()) { |
126 | entry->edx &= ~bit(X86_FEATURE_NX); | |
00b27a3e AK |
127 | printk(KERN_INFO "kvm: guest NX capability removed\n"); |
128 | } | |
129 | } | |
130 | ||
131 | /* when an old userspace process fills a new kernel module */ | |
132 | int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, | |
133 | struct kvm_cpuid *cpuid, | |
134 | struct kvm_cpuid_entry __user *entries) | |
135 | { | |
136 | int r, i; | |
137 | struct kvm_cpuid_entry *cpuid_entries; | |
138 | ||
139 | r = -E2BIG; | |
140 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
141 | goto out; | |
142 | r = -ENOMEM; | |
143 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent); | |
144 | if (!cpuid_entries) | |
145 | goto out; | |
146 | r = -EFAULT; | |
147 | if (copy_from_user(cpuid_entries, entries, | |
148 | cpuid->nent * sizeof(struct kvm_cpuid_entry))) | |
149 | goto out_free; | |
150 | for (i = 0; i < cpuid->nent; i++) { | |
151 | vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; | |
152 | vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; | |
153 | vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; | |
154 | vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; | |
155 | vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; | |
156 | vcpu->arch.cpuid_entries[i].index = 0; | |
157 | vcpu->arch.cpuid_entries[i].flags = 0; | |
158 | vcpu->arch.cpuid_entries[i].padding[0] = 0; | |
159 | vcpu->arch.cpuid_entries[i].padding[1] = 0; | |
160 | vcpu->arch.cpuid_entries[i].padding[2] = 0; | |
161 | } | |
162 | vcpu->arch.cpuid_nent = cpuid->nent; | |
163 | cpuid_fix_nx_cap(vcpu); | |
00b27a3e AK |
164 | kvm_apic_set_version(vcpu); |
165 | kvm_x86_ops->cpuid_update(vcpu); | |
dd598091 | 166 | r = kvm_update_cpuid(vcpu); |
00b27a3e AK |
167 | |
168 | out_free: | |
169 | vfree(cpuid_entries); | |
170 | out: | |
171 | return r; | |
172 | } | |
173 | ||
174 | int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, | |
175 | struct kvm_cpuid2 *cpuid, | |
176 | struct kvm_cpuid_entry2 __user *entries) | |
177 | { | |
178 | int r; | |
179 | ||
180 | r = -E2BIG; | |
181 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
182 | goto out; | |
183 | r = -EFAULT; | |
184 | if (copy_from_user(&vcpu->arch.cpuid_entries, entries, | |
185 | cpuid->nent * sizeof(struct kvm_cpuid_entry2))) | |
186 | goto out; | |
187 | vcpu->arch.cpuid_nent = cpuid->nent; | |
188 | kvm_apic_set_version(vcpu); | |
189 | kvm_x86_ops->cpuid_update(vcpu); | |
dd598091 | 190 | r = kvm_update_cpuid(vcpu); |
00b27a3e AK |
191 | out: |
192 | return r; | |
193 | } | |
194 | ||
195 | int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, | |
196 | struct kvm_cpuid2 *cpuid, | |
197 | struct kvm_cpuid_entry2 __user *entries) | |
198 | { | |
199 | int r; | |
200 | ||
201 | r = -E2BIG; | |
202 | if (cpuid->nent < vcpu->arch.cpuid_nent) | |
203 | goto out; | |
204 | r = -EFAULT; | |
205 | if (copy_to_user(entries, &vcpu->arch.cpuid_entries, | |
206 | vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) | |
207 | goto out; | |
208 | return 0; | |
209 | ||
210 | out: | |
211 | cpuid->nent = vcpu->arch.cpuid_nent; | |
212 | return r; | |
213 | } | |
214 | ||
215 | static void cpuid_mask(u32 *word, int wordnum) | |
216 | { | |
217 | *word &= boot_cpu_data.x86_capability[wordnum]; | |
218 | } | |
219 | ||
220 | static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
221 | u32 index) | |
222 | { | |
223 | entry->function = function; | |
224 | entry->index = index; | |
225 | cpuid_count(entry->function, entry->index, | |
226 | &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); | |
227 | entry->flags = 0; | |
228 | } | |
229 | ||
00b27a3e AK |
230 | #define F(x) bit(X86_FEATURE_##x) |
231 | ||
9c15bb1d BP |
232 | static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, |
233 | u32 func, u32 index, int *nent, int maxnent) | |
234 | { | |
84cffe49 BP |
235 | switch (func) { |
236 | case 0: | |
237 | entry->eax = 1; /* only one leaf currently */ | |
238 | ++*nent; | |
239 | break; | |
240 | case 1: | |
241 | entry->ecx = F(MOVBE); | |
242 | ++*nent; | |
243 | break; | |
244 | default: | |
245 | break; | |
246 | } | |
247 | ||
248 | entry->function = func; | |
249 | entry->index = index; | |
250 | ||
9c15bb1d BP |
251 | return 0; |
252 | } | |
253 | ||
254 | static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
255 | u32 index, int *nent, int maxnent) | |
00b27a3e | 256 | { |
831bf664 | 257 | int r; |
00b27a3e AK |
258 | unsigned f_nx = is_efer_nx() ? F(NX) : 0; |
259 | #ifdef CONFIG_X86_64 | |
260 | unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) | |
261 | ? F(GBPAGES) : 0; | |
262 | unsigned f_lm = F(LM); | |
263 | #else | |
264 | unsigned f_gbpages = 0; | |
265 | unsigned f_lm = 0; | |
266 | #endif | |
267 | unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; | |
ad756a16 | 268 | unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; |
93c4adc7 | 269 | unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0; |
00b27a3e AK |
270 | |
271 | /* cpuid 1.edx */ | |
272 | const u32 kvm_supported_word0_x86_features = | |
273 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
274 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
275 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | | |
276 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
840d2830 | 277 | F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | |
00b27a3e AK |
278 | 0 /* Reserved, DS, ACPI */ | F(MMX) | |
279 | F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | | |
280 | 0 /* HTT, TM, Reserved, PBE */; | |
281 | /* cpuid 0x80000001.edx */ | |
282 | const u32 kvm_supported_word1_x86_features = | |
283 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
284 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
285 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | | |
286 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
287 | F(PAT) | F(PSE36) | 0 /* Reserved */ | | |
288 | f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | | |
289 | F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | | |
290 | 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); | |
291 | /* cpuid 1.ecx */ | |
292 | const u32 kvm_supported_word4_x86_features = | |
87c00572 GS |
293 | /* NOTE: MONITOR (and MWAIT) are emulated as NOP, |
294 | * but *not* advertised to guests via CPUID ! */ | |
00b27a3e AK |
295 | F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | |
296 | 0 /* DS-CPL, VMX, SMX, EST */ | | |
297 | 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | | |
fb215366 | 298 | F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | |
ad756a16 | 299 | F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | |
00b27a3e AK |
300 | F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | |
301 | 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | | |
302 | F(F16C) | F(RDRAND); | |
303 | /* cpuid 0x80000001.ecx */ | |
304 | const u32 kvm_supported_word6_x86_features = | |
305 | F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | | |
306 | F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | | |
2b036c6b | 307 | F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | |
00b27a3e AK |
308 | 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); |
309 | ||
310 | /* cpuid 0xC0000001.edx */ | |
311 | const u32 kvm_supported_word5_x86_features = | |
312 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | | |
313 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | | |
314 | F(PMM) | F(PMM_EN); | |
315 | ||
316 | /* cpuid 7.0.ebx */ | |
317 | const u32 kvm_supported_word9_x86_features = | |
83c52915 | 318 | F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | |
390bd528 | 319 | F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | |
612263b3 CP |
320 | F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) | |
321 | F(AVX512CD); | |
00b27a3e | 322 | |
b65d6e17 PB |
323 | /* cpuid 0xD.1.eax */ |
324 | const u32 kvm_supported_word10_x86_features = | |
325 | F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1); | |
326 | ||
00b27a3e AK |
327 | /* all calls to cpuid_count() should be made on the same cpu */ |
328 | get_cpu(); | |
831bf664 SL |
329 | |
330 | r = -E2BIG; | |
331 | ||
332 | if (*nent >= maxnent) | |
333 | goto out; | |
334 | ||
00b27a3e AK |
335 | do_cpuid_1_ent(entry, function, index); |
336 | ++*nent; | |
337 | ||
338 | switch (function) { | |
339 | case 0: | |
340 | entry->eax = min(entry->eax, (u32)0xd); | |
341 | break; | |
342 | case 1: | |
343 | entry->edx &= kvm_supported_word0_x86_features; | |
344 | cpuid_mask(&entry->edx, 0); | |
345 | entry->ecx &= kvm_supported_word4_x86_features; | |
346 | cpuid_mask(&entry->ecx, 4); | |
347 | /* we support x2apic emulation even if host does not support | |
348 | * it since we emulate x2apic in software */ | |
349 | entry->ecx |= F(X2APIC); | |
350 | break; | |
351 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
352 | * may return different values. This forces us to get_cpu() before | |
353 | * issuing the first command, and also to emulate this annoying behavior | |
354 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
355 | case 2: { | |
356 | int t, times = entry->eax & 0xff; | |
357 | ||
358 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
359 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
831bf664 SL |
360 | for (t = 1; t < times; ++t) { |
361 | if (*nent >= maxnent) | |
362 | goto out; | |
363 | ||
00b27a3e AK |
364 | do_cpuid_1_ent(&entry[t], function, 0); |
365 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
366 | ++*nent; | |
367 | } | |
368 | break; | |
369 | } | |
370 | /* function 4 has additional index. */ | |
371 | case 4: { | |
372 | int i, cache_type; | |
373 | ||
374 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
375 | /* read more entries until cache_type is zero */ | |
831bf664 SL |
376 | for (i = 1; ; ++i) { |
377 | if (*nent >= maxnent) | |
378 | goto out; | |
379 | ||
00b27a3e AK |
380 | cache_type = entry[i - 1].eax & 0x1f; |
381 | if (!cache_type) | |
382 | break; | |
383 | do_cpuid_1_ent(&entry[i], function, i); | |
384 | entry[i].flags |= | |
385 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
386 | ++*nent; | |
387 | } | |
388 | break; | |
389 | } | |
390 | case 7: { | |
391 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
bbbda795 | 392 | /* Mask ebx against host capability word 9 */ |
00b27a3e AK |
393 | if (index == 0) { |
394 | entry->ebx &= kvm_supported_word9_x86_features; | |
395 | cpuid_mask(&entry->ebx, 9); | |
ba904635 WA |
396 | // TSC_ADJUST is emulated |
397 | entry->ebx |= F(TSC_ADJUST); | |
00b27a3e AK |
398 | } else |
399 | entry->ebx = 0; | |
400 | entry->eax = 0; | |
401 | entry->ecx = 0; | |
402 | entry->edx = 0; | |
403 | break; | |
404 | } | |
405 | case 9: | |
406 | break; | |
a6c06ed1 GN |
407 | case 0xa: { /* Architectural Performance Monitoring */ |
408 | struct x86_pmu_capability cap; | |
409 | union cpuid10_eax eax; | |
410 | union cpuid10_edx edx; | |
411 | ||
412 | perf_get_x86_pmu_capability(&cap); | |
413 | ||
414 | /* | |
415 | * Only support guest architectural pmu on a host | |
416 | * with architectural pmu. | |
417 | */ | |
418 | if (!cap.version) | |
419 | memset(&cap, 0, sizeof(cap)); | |
420 | ||
421 | eax.split.version_id = min(cap.version, 2); | |
422 | eax.split.num_counters = cap.num_counters_gp; | |
423 | eax.split.bit_width = cap.bit_width_gp; | |
424 | eax.split.mask_length = cap.events_mask_len; | |
425 | ||
426 | edx.split.num_counters_fixed = cap.num_counters_fixed; | |
427 | edx.split.bit_width_fixed = cap.bit_width_fixed; | |
428 | edx.split.reserved = 0; | |
429 | ||
430 | entry->eax = eax.full; | |
431 | entry->ebx = cap.events_mask; | |
432 | entry->ecx = 0; | |
433 | entry->edx = edx.full; | |
434 | break; | |
435 | } | |
00b27a3e AK |
436 | /* function 0xb has additional index. */ |
437 | case 0xb: { | |
438 | int i, level_type; | |
439 | ||
440 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
441 | /* read more entries until level_type is zero */ | |
831bf664 SL |
442 | for (i = 1; ; ++i) { |
443 | if (*nent >= maxnent) | |
444 | goto out; | |
445 | ||
00b27a3e AK |
446 | level_type = entry[i - 1].ecx & 0xff00; |
447 | if (!level_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 | } | |
456 | case 0xd: { | |
457 | int idx, i; | |
4ff41732 | 458 | u64 supported = kvm_supported_xcr0(); |
00b27a3e | 459 | |
4ff41732 PB |
460 | entry->eax &= supported; |
461 | entry->edx &= supported >> 32; | |
00b27a3e | 462 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
b65d6e17 PB |
463 | if (!supported) |
464 | break; | |
465 | ||
831bf664 | 466 | for (idx = 1, i = 1; idx < 64; ++idx) { |
4ff41732 | 467 | u64 mask = ((u64)1 << idx); |
831bf664 SL |
468 | if (*nent >= maxnent) |
469 | goto out; | |
470 | ||
00b27a3e | 471 | do_cpuid_1_ent(&entry[i], function, idx); |
b65d6e17 PB |
472 | if (idx == 1) |
473 | entry[i].eax &= kvm_supported_word10_x86_features; | |
474 | else if (entry[i].eax == 0 || !(supported & mask)) | |
00b27a3e AK |
475 | continue; |
476 | entry[i].flags |= | |
477 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
478 | ++*nent; | |
479 | ++i; | |
480 | } | |
481 | break; | |
482 | } | |
483 | case KVM_CPUID_SIGNATURE: { | |
326d07cb MK |
484 | static const char signature[12] = "KVMKVMKVM\0\0"; |
485 | const u32 *sigptr = (const u32 *)signature; | |
57c22e5f | 486 | entry->eax = KVM_CPUID_FEATURES; |
00b27a3e AK |
487 | entry->ebx = sigptr[0]; |
488 | entry->ecx = sigptr[1]; | |
489 | entry->edx = sigptr[2]; | |
490 | break; | |
491 | } | |
492 | case KVM_CPUID_FEATURES: | |
493 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
494 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
495 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
496 | (1 << KVM_FEATURE_ASYNC_PF) | | |
ae7a2a3f | 497 | (1 << KVM_FEATURE_PV_EOI) | |
6aef266c SV |
498 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | |
499 | (1 << KVM_FEATURE_PV_UNHALT); | |
00b27a3e AK |
500 | |
501 | if (sched_info_on()) | |
502 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
503 | ||
504 | entry->ebx = 0; | |
505 | entry->ecx = 0; | |
506 | entry->edx = 0; | |
507 | break; | |
508 | case 0x80000000: | |
509 | entry->eax = min(entry->eax, 0x8000001a); | |
510 | break; | |
511 | case 0x80000001: | |
512 | entry->edx &= kvm_supported_word1_x86_features; | |
513 | cpuid_mask(&entry->edx, 1); | |
514 | entry->ecx &= kvm_supported_word6_x86_features; | |
515 | cpuid_mask(&entry->ecx, 6); | |
516 | break; | |
e4c9a5a1 MT |
517 | case 0x80000007: /* Advanced power management */ |
518 | /* invariant TSC is CPUID.80000007H:EDX[8] */ | |
519 | entry->edx &= (1 << 8); | |
520 | /* mask against host */ | |
521 | entry->edx &= boot_cpu_data.x86_power; | |
522 | entry->eax = entry->ebx = entry->ecx = 0; | |
523 | break; | |
00b27a3e AK |
524 | case 0x80000008: { |
525 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
526 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
527 | unsigned phys_as = entry->eax & 0xff; | |
528 | ||
529 | if (!g_phys_as) | |
530 | g_phys_as = phys_as; | |
531 | entry->eax = g_phys_as | (virt_as << 8); | |
532 | entry->ebx = entry->edx = 0; | |
533 | break; | |
534 | } | |
535 | case 0x80000019: | |
536 | entry->ecx = entry->edx = 0; | |
537 | break; | |
538 | case 0x8000001a: | |
539 | break; | |
540 | case 0x8000001d: | |
541 | break; | |
542 | /*Add support for Centaur's CPUID instruction*/ | |
543 | case 0xC0000000: | |
544 | /*Just support up to 0xC0000004 now*/ | |
545 | entry->eax = min(entry->eax, 0xC0000004); | |
546 | break; | |
547 | case 0xC0000001: | |
548 | entry->edx &= kvm_supported_word5_x86_features; | |
549 | cpuid_mask(&entry->edx, 5); | |
550 | break; | |
551 | case 3: /* Processor serial number */ | |
552 | case 5: /* MONITOR/MWAIT */ | |
553 | case 6: /* Thermal management */ | |
00b27a3e AK |
554 | case 0xC0000002: |
555 | case 0xC0000003: | |
556 | case 0xC0000004: | |
557 | default: | |
558 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
559 | break; | |
560 | } | |
561 | ||
562 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
563 | ||
831bf664 SL |
564 | r = 0; |
565 | ||
566 | out: | |
00b27a3e | 567 | put_cpu(); |
831bf664 SL |
568 | |
569 | return r; | |
00b27a3e AK |
570 | } |
571 | ||
9c15bb1d BP |
572 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, |
573 | u32 idx, int *nent, int maxnent, unsigned int type) | |
574 | { | |
575 | if (type == KVM_GET_EMULATED_CPUID) | |
576 | return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); | |
577 | ||
578 | return __do_cpuid_ent(entry, func, idx, nent, maxnent); | |
579 | } | |
580 | ||
00b27a3e AK |
581 | #undef F |
582 | ||
831bf664 SL |
583 | struct kvm_cpuid_param { |
584 | u32 func; | |
585 | u32 idx; | |
586 | bool has_leaf_count; | |
326d07cb | 587 | bool (*qualifier)(const struct kvm_cpuid_param *param); |
831bf664 SL |
588 | }; |
589 | ||
326d07cb | 590 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) |
831bf664 SL |
591 | { |
592 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; | |
593 | } | |
594 | ||
9c15bb1d BP |
595 | static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, |
596 | __u32 num_entries, unsigned int ioctl_type) | |
597 | { | |
598 | int i; | |
1b2ca422 | 599 | __u32 pad[3]; |
9c15bb1d BP |
600 | |
601 | if (ioctl_type != KVM_GET_EMULATED_CPUID) | |
602 | return false; | |
603 | ||
604 | /* | |
605 | * We want to make sure that ->padding is being passed clean from | |
606 | * userspace in case we want to use it for something in the future. | |
607 | * | |
608 | * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we | |
609 | * have to give ourselves satisfied only with the emulated side. /me | |
610 | * sheds a tear. | |
611 | */ | |
612 | for (i = 0; i < num_entries; i++) { | |
1b2ca422 BP |
613 | if (copy_from_user(pad, entries[i].padding, sizeof(pad))) |
614 | return true; | |
615 | ||
616 | if (pad[0] || pad[1] || pad[2]) | |
9c15bb1d BP |
617 | return true; |
618 | } | |
619 | return false; | |
620 | } | |
621 | ||
622 | int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, | |
623 | struct kvm_cpuid_entry2 __user *entries, | |
624 | unsigned int type) | |
00b27a3e AK |
625 | { |
626 | struct kvm_cpuid_entry2 *cpuid_entries; | |
831bf664 | 627 | int limit, nent = 0, r = -E2BIG, i; |
00b27a3e | 628 | u32 func; |
326d07cb | 629 | static const struct kvm_cpuid_param param[] = { |
831bf664 SL |
630 | { .func = 0, .has_leaf_count = true }, |
631 | { .func = 0x80000000, .has_leaf_count = true }, | |
632 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, | |
633 | { .func = KVM_CPUID_SIGNATURE }, | |
634 | { .func = KVM_CPUID_FEATURES }, | |
635 | }; | |
00b27a3e AK |
636 | |
637 | if (cpuid->nent < 1) | |
638 | goto out; | |
639 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
640 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
9c15bb1d BP |
641 | |
642 | if (sanity_check_entries(entries, cpuid->nent, type)) | |
643 | return -EINVAL; | |
644 | ||
00b27a3e | 645 | r = -ENOMEM; |
84cffe49 | 646 | cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); |
00b27a3e AK |
647 | if (!cpuid_entries) |
648 | goto out; | |
649 | ||
831bf664 SL |
650 | r = 0; |
651 | for (i = 0; i < ARRAY_SIZE(param); i++) { | |
326d07cb | 652 | const struct kvm_cpuid_param *ent = ¶m[i]; |
00b27a3e | 653 | |
831bf664 SL |
654 | if (ent->qualifier && !ent->qualifier(ent)) |
655 | continue; | |
00b27a3e | 656 | |
831bf664 | 657 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, |
9c15bb1d | 658 | &nent, cpuid->nent, type); |
00b27a3e | 659 | |
831bf664 | 660 | if (r) |
00b27a3e AK |
661 | goto out_free; |
662 | ||
831bf664 SL |
663 | if (!ent->has_leaf_count) |
664 | continue; | |
665 | ||
00b27a3e | 666 | limit = cpuid_entries[nent - 1].eax; |
831bf664 SL |
667 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) |
668 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, | |
9c15bb1d | 669 | &nent, cpuid->nent, type); |
00b27a3e | 670 | |
831bf664 | 671 | if (r) |
00b27a3e AK |
672 | goto out_free; |
673 | } | |
674 | ||
00b27a3e AK |
675 | r = -EFAULT; |
676 | if (copy_to_user(entries, cpuid_entries, | |
677 | nent * sizeof(struct kvm_cpuid_entry2))) | |
678 | goto out_free; | |
679 | cpuid->nent = nent; | |
680 | r = 0; | |
681 | ||
682 | out_free: | |
683 | vfree(cpuid_entries); | |
684 | out: | |
685 | return r; | |
686 | } | |
687 | ||
688 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
689 | { | |
690 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
691 | int j, nent = vcpu->arch.cpuid_nent; | |
692 | ||
693 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
694 | /* when no next entry is found, the current entry[i] is reselected */ | |
695 | for (j = i + 1; ; j = (j + 1) % nent) { | |
696 | struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; | |
697 | if (ej->function == e->function) { | |
698 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
699 | return j; | |
700 | } | |
701 | } | |
702 | return 0; /* silence gcc, even though control never reaches here */ | |
703 | } | |
704 | ||
705 | /* find an entry with matching function, matching index (if needed), and that | |
706 | * should be read next (if it's stateful) */ | |
707 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
708 | u32 function, u32 index) | |
709 | { | |
710 | if (e->function != function) | |
711 | return 0; | |
712 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
713 | return 0; | |
714 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
715 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
716 | return 0; | |
717 | return 1; | |
718 | } | |
719 | ||
720 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
721 | u32 function, u32 index) | |
722 | { | |
723 | int i; | |
724 | struct kvm_cpuid_entry2 *best = NULL; | |
725 | ||
726 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
727 | struct kvm_cpuid_entry2 *e; | |
728 | ||
729 | e = &vcpu->arch.cpuid_entries[i]; | |
730 | if (is_matching_cpuid_entry(e, function, index)) { | |
731 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
732 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
733 | best = e; | |
734 | break; | |
735 | } | |
736 | } | |
737 | return best; | |
738 | } | |
739 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
740 | ||
741 | int cpuid_maxphyaddr(struct kvm_vcpu *vcpu) | |
742 | { | |
743 | struct kvm_cpuid_entry2 *best; | |
744 | ||
745 | best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); | |
746 | if (!best || best->eax < 0x80000008) | |
747 | goto not_found; | |
748 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
749 | if (best) | |
750 | return best->eax & 0xff; | |
751 | not_found: | |
752 | return 36; | |
753 | } | |
3573e22c | 754 | EXPORT_SYMBOL_GPL(cpuid_maxphyaddr); |
00b27a3e AK |
755 | |
756 | /* | |
757 | * If no match is found, check whether we exceed the vCPU's limit | |
758 | * and return the content of the highest valid _standard_ leaf instead. | |
759 | * This is to satisfy the CPUID specification. | |
760 | */ | |
761 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
762 | u32 function, u32 index) | |
763 | { | |
764 | struct kvm_cpuid_entry2 *maxlevel; | |
765 | ||
766 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
767 | if (!maxlevel || maxlevel->eax >= function) | |
768 | return NULL; | |
769 | if (function & 0x80000000) { | |
770 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
771 | if (!maxlevel) | |
772 | return NULL; | |
773 | } | |
774 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
775 | } | |
776 | ||
62046e5a | 777 | void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
00b27a3e | 778 | { |
62046e5a | 779 | u32 function = *eax, index = *ecx; |
00b27a3e AK |
780 | struct kvm_cpuid_entry2 *best; |
781 | ||
00b27a3e AK |
782 | best = kvm_find_cpuid_entry(vcpu, function, index); |
783 | ||
784 | if (!best) | |
785 | best = check_cpuid_limit(vcpu, function, index); | |
786 | ||
bc613494 MT |
787 | /* |
788 | * Perfmon not yet supported for L2 guest. | |
789 | */ | |
790 | if (is_guest_mode(vcpu) && function == 0xa) | |
791 | best = NULL; | |
792 | ||
00b27a3e | 793 | if (best) { |
62046e5a AK |
794 | *eax = best->eax; |
795 | *ebx = best->ebx; | |
796 | *ecx = best->ecx; | |
797 | *edx = best->edx; | |
798 | } else | |
799 | *eax = *ebx = *ecx = *edx = 0; | |
a9d4e439 | 800 | trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx); |
62046e5a | 801 | } |
66f7b72e | 802 | EXPORT_SYMBOL_GPL(kvm_cpuid); |
62046e5a AK |
803 | |
804 | void kvm_emulate_cpuid(struct kvm_vcpu *vcpu) | |
805 | { | |
806 | u32 function, eax, ebx, ecx, edx; | |
807 | ||
808 | function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX); | |
809 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
810 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx); | |
811 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); | |
812 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); | |
813 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); | |
814 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); | |
00b27a3e | 815 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
00b27a3e AK |
816 | } |
817 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |