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> | |
17 | #include <asm/user.h> | |
18 | #include <asm/xsave.h> | |
19 | #include "cpuid.h" | |
20 | #include "lapic.h" | |
21 | #include "mmu.h" | |
22 | #include "trace.h" | |
23 | ||
24 | void kvm_update_cpuid(struct kvm_vcpu *vcpu) | |
25 | { | |
26 | struct kvm_cpuid_entry2 *best; | |
27 | struct kvm_lapic *apic = vcpu->arch.apic; | |
28 | ||
29 | best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
30 | if (!best) | |
31 | return; | |
32 | ||
33 | /* Update OSXSAVE bit */ | |
34 | if (cpu_has_xsave && best->function == 0x1) { | |
35 | best->ecx &= ~(bit(X86_FEATURE_OSXSAVE)); | |
36 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) | |
37 | best->ecx |= bit(X86_FEATURE_OSXSAVE); | |
38 | } | |
39 | ||
40 | if (apic) { | |
41 | if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER)) | |
42 | apic->lapic_timer.timer_mode_mask = 3 << 17; | |
43 | else | |
44 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
45 | } | |
46 | } | |
47 | ||
48 | static int is_efer_nx(void) | |
49 | { | |
50 | unsigned long long efer = 0; | |
51 | ||
52 | rdmsrl_safe(MSR_EFER, &efer); | |
53 | return efer & EFER_NX; | |
54 | } | |
55 | ||
56 | static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) | |
57 | { | |
58 | int i; | |
59 | struct kvm_cpuid_entry2 *e, *entry; | |
60 | ||
61 | entry = NULL; | |
62 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
63 | e = &vcpu->arch.cpuid_entries[i]; | |
64 | if (e->function == 0x80000001) { | |
65 | entry = e; | |
66 | break; | |
67 | } | |
68 | } | |
69 | if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) { | |
70 | entry->edx &= ~(1 << 20); | |
71 | printk(KERN_INFO "kvm: guest NX capability removed\n"); | |
72 | } | |
73 | } | |
74 | ||
75 | /* when an old userspace process fills a new kernel module */ | |
76 | int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, | |
77 | struct kvm_cpuid *cpuid, | |
78 | struct kvm_cpuid_entry __user *entries) | |
79 | { | |
80 | int r, i; | |
81 | struct kvm_cpuid_entry *cpuid_entries; | |
82 | ||
83 | r = -E2BIG; | |
84 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
85 | goto out; | |
86 | r = -ENOMEM; | |
87 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent); | |
88 | if (!cpuid_entries) | |
89 | goto out; | |
90 | r = -EFAULT; | |
91 | if (copy_from_user(cpuid_entries, entries, | |
92 | cpuid->nent * sizeof(struct kvm_cpuid_entry))) | |
93 | goto out_free; | |
94 | for (i = 0; i < cpuid->nent; i++) { | |
95 | vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; | |
96 | vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; | |
97 | vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; | |
98 | vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; | |
99 | vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; | |
100 | vcpu->arch.cpuid_entries[i].index = 0; | |
101 | vcpu->arch.cpuid_entries[i].flags = 0; | |
102 | vcpu->arch.cpuid_entries[i].padding[0] = 0; | |
103 | vcpu->arch.cpuid_entries[i].padding[1] = 0; | |
104 | vcpu->arch.cpuid_entries[i].padding[2] = 0; | |
105 | } | |
106 | vcpu->arch.cpuid_nent = cpuid->nent; | |
107 | cpuid_fix_nx_cap(vcpu); | |
108 | r = 0; | |
109 | kvm_apic_set_version(vcpu); | |
110 | kvm_x86_ops->cpuid_update(vcpu); | |
111 | kvm_update_cpuid(vcpu); | |
112 | ||
113 | out_free: | |
114 | vfree(cpuid_entries); | |
115 | out: | |
116 | return r; | |
117 | } | |
118 | ||
119 | int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, | |
120 | struct kvm_cpuid2 *cpuid, | |
121 | struct kvm_cpuid_entry2 __user *entries) | |
122 | { | |
123 | int r; | |
124 | ||
125 | r = -E2BIG; | |
126 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
127 | goto out; | |
128 | r = -EFAULT; | |
129 | if (copy_from_user(&vcpu->arch.cpuid_entries, entries, | |
130 | cpuid->nent * sizeof(struct kvm_cpuid_entry2))) | |
131 | goto out; | |
132 | vcpu->arch.cpuid_nent = cpuid->nent; | |
133 | kvm_apic_set_version(vcpu); | |
134 | kvm_x86_ops->cpuid_update(vcpu); | |
135 | kvm_update_cpuid(vcpu); | |
136 | return 0; | |
137 | ||
138 | out: | |
139 | return r; | |
140 | } | |
141 | ||
142 | int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, | |
143 | struct kvm_cpuid2 *cpuid, | |
144 | struct kvm_cpuid_entry2 __user *entries) | |
145 | { | |
146 | int r; | |
147 | ||
148 | r = -E2BIG; | |
149 | if (cpuid->nent < vcpu->arch.cpuid_nent) | |
150 | goto out; | |
151 | r = -EFAULT; | |
152 | if (copy_to_user(entries, &vcpu->arch.cpuid_entries, | |
153 | vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) | |
154 | goto out; | |
155 | return 0; | |
156 | ||
157 | out: | |
158 | cpuid->nent = vcpu->arch.cpuid_nent; | |
159 | return r; | |
160 | } | |
161 | ||
162 | static void cpuid_mask(u32 *word, int wordnum) | |
163 | { | |
164 | *word &= boot_cpu_data.x86_capability[wordnum]; | |
165 | } | |
166 | ||
167 | static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
168 | u32 index) | |
169 | { | |
170 | entry->function = function; | |
171 | entry->index = index; | |
172 | cpuid_count(entry->function, entry->index, | |
173 | &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); | |
174 | entry->flags = 0; | |
175 | } | |
176 | ||
177 | static bool supported_xcr0_bit(unsigned bit) | |
178 | { | |
179 | u64 mask = ((u64)1 << bit); | |
180 | ||
181 | return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0; | |
182 | } | |
183 | ||
184 | #define F(x) bit(X86_FEATURE_##x) | |
185 | ||
186 | static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
187 | u32 index, int *nent, int maxnent) | |
188 | { | |
189 | unsigned f_nx = is_efer_nx() ? F(NX) : 0; | |
190 | #ifdef CONFIG_X86_64 | |
191 | unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) | |
192 | ? F(GBPAGES) : 0; | |
193 | unsigned f_lm = F(LM); | |
194 | #else | |
195 | unsigned f_gbpages = 0; | |
196 | unsigned f_lm = 0; | |
197 | #endif | |
198 | unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; | |
199 | ||
200 | /* cpuid 1.edx */ | |
201 | const u32 kvm_supported_word0_x86_features = | |
202 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
203 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
204 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | | |
205 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
206 | F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) | | |
207 | 0 /* Reserved, DS, ACPI */ | F(MMX) | | |
208 | F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | | |
209 | 0 /* HTT, TM, Reserved, PBE */; | |
210 | /* cpuid 0x80000001.edx */ | |
211 | const u32 kvm_supported_word1_x86_features = | |
212 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
213 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
214 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | | |
215 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
216 | F(PAT) | F(PSE36) | 0 /* Reserved */ | | |
217 | f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | | |
218 | F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | | |
219 | 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); | |
220 | /* cpuid 1.ecx */ | |
221 | const u32 kvm_supported_word4_x86_features = | |
222 | F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | | |
223 | 0 /* DS-CPL, VMX, SMX, EST */ | | |
224 | 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | | |
225 | 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ | | |
226 | 0 /* Reserved, DCA */ | F(XMM4_1) | | |
227 | F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | | |
228 | 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | | |
229 | F(F16C) | F(RDRAND); | |
230 | /* cpuid 0x80000001.ecx */ | |
231 | const u32 kvm_supported_word6_x86_features = | |
232 | F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | | |
233 | F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | | |
234 | F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) | | |
235 | 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); | |
236 | ||
237 | /* cpuid 0xC0000001.edx */ | |
238 | const u32 kvm_supported_word5_x86_features = | |
239 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | | |
240 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | | |
241 | F(PMM) | F(PMM_EN); | |
242 | ||
243 | /* cpuid 7.0.ebx */ | |
244 | const u32 kvm_supported_word9_x86_features = | |
245 | F(SMEP) | F(FSGSBASE) | F(ERMS); | |
246 | ||
247 | /* all calls to cpuid_count() should be made on the same cpu */ | |
248 | get_cpu(); | |
249 | do_cpuid_1_ent(entry, function, index); | |
250 | ++*nent; | |
251 | ||
252 | switch (function) { | |
253 | case 0: | |
254 | entry->eax = min(entry->eax, (u32)0xd); | |
255 | break; | |
256 | case 1: | |
257 | entry->edx &= kvm_supported_word0_x86_features; | |
258 | cpuid_mask(&entry->edx, 0); | |
259 | entry->ecx &= kvm_supported_word4_x86_features; | |
260 | cpuid_mask(&entry->ecx, 4); | |
261 | /* we support x2apic emulation even if host does not support | |
262 | * it since we emulate x2apic in software */ | |
263 | entry->ecx |= F(X2APIC); | |
264 | break; | |
265 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
266 | * may return different values. This forces us to get_cpu() before | |
267 | * issuing the first command, and also to emulate this annoying behavior | |
268 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
269 | case 2: { | |
270 | int t, times = entry->eax & 0xff; | |
271 | ||
272 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
273 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
274 | for (t = 1; t < times && *nent < maxnent; ++t) { | |
275 | do_cpuid_1_ent(&entry[t], function, 0); | |
276 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
277 | ++*nent; | |
278 | } | |
279 | break; | |
280 | } | |
281 | /* function 4 has additional index. */ | |
282 | case 4: { | |
283 | int i, cache_type; | |
284 | ||
285 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
286 | /* read more entries until cache_type is zero */ | |
287 | for (i = 1; *nent < maxnent; ++i) { | |
288 | cache_type = entry[i - 1].eax & 0x1f; | |
289 | if (!cache_type) | |
290 | break; | |
291 | do_cpuid_1_ent(&entry[i], function, i); | |
292 | entry[i].flags |= | |
293 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
294 | ++*nent; | |
295 | } | |
296 | break; | |
297 | } | |
298 | case 7: { | |
299 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
300 | /* Mask ebx against host capbability word 9 */ | |
301 | if (index == 0) { | |
302 | entry->ebx &= kvm_supported_word9_x86_features; | |
303 | cpuid_mask(&entry->ebx, 9); | |
304 | } else | |
305 | entry->ebx = 0; | |
306 | entry->eax = 0; | |
307 | entry->ecx = 0; | |
308 | entry->edx = 0; | |
309 | break; | |
310 | } | |
311 | case 9: | |
312 | break; | |
313 | /* function 0xb has additional index. */ | |
314 | case 0xb: { | |
315 | int i, level_type; | |
316 | ||
317 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
318 | /* read more entries until level_type is zero */ | |
319 | for (i = 1; *nent < maxnent; ++i) { | |
320 | level_type = entry[i - 1].ecx & 0xff00; | |
321 | if (!level_type) | |
322 | break; | |
323 | do_cpuid_1_ent(&entry[i], function, i); | |
324 | entry[i].flags |= | |
325 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
326 | ++*nent; | |
327 | } | |
328 | break; | |
329 | } | |
330 | case 0xd: { | |
331 | int idx, i; | |
332 | ||
333 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
334 | for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) { | |
335 | do_cpuid_1_ent(&entry[i], function, idx); | |
336 | if (entry[i].eax == 0 || !supported_xcr0_bit(idx)) | |
337 | continue; | |
338 | entry[i].flags |= | |
339 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
340 | ++*nent; | |
341 | ++i; | |
342 | } | |
343 | break; | |
344 | } | |
345 | case KVM_CPUID_SIGNATURE: { | |
346 | char signature[12] = "KVMKVMKVM\0\0"; | |
347 | u32 *sigptr = (u32 *)signature; | |
348 | entry->eax = 0; | |
349 | entry->ebx = sigptr[0]; | |
350 | entry->ecx = sigptr[1]; | |
351 | entry->edx = sigptr[2]; | |
352 | break; | |
353 | } | |
354 | case KVM_CPUID_FEATURES: | |
355 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
356 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
357 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
358 | (1 << KVM_FEATURE_ASYNC_PF) | | |
359 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT); | |
360 | ||
361 | if (sched_info_on()) | |
362 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
363 | ||
364 | entry->ebx = 0; | |
365 | entry->ecx = 0; | |
366 | entry->edx = 0; | |
367 | break; | |
368 | case 0x80000000: | |
369 | entry->eax = min(entry->eax, 0x8000001a); | |
370 | break; | |
371 | case 0x80000001: | |
372 | entry->edx &= kvm_supported_word1_x86_features; | |
373 | cpuid_mask(&entry->edx, 1); | |
374 | entry->ecx &= kvm_supported_word6_x86_features; | |
375 | cpuid_mask(&entry->ecx, 6); | |
376 | break; | |
377 | case 0x80000008: { | |
378 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
379 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
380 | unsigned phys_as = entry->eax & 0xff; | |
381 | ||
382 | if (!g_phys_as) | |
383 | g_phys_as = phys_as; | |
384 | entry->eax = g_phys_as | (virt_as << 8); | |
385 | entry->ebx = entry->edx = 0; | |
386 | break; | |
387 | } | |
388 | case 0x80000019: | |
389 | entry->ecx = entry->edx = 0; | |
390 | break; | |
391 | case 0x8000001a: | |
392 | break; | |
393 | case 0x8000001d: | |
394 | break; | |
395 | /*Add support for Centaur's CPUID instruction*/ | |
396 | case 0xC0000000: | |
397 | /*Just support up to 0xC0000004 now*/ | |
398 | entry->eax = min(entry->eax, 0xC0000004); | |
399 | break; | |
400 | case 0xC0000001: | |
401 | entry->edx &= kvm_supported_word5_x86_features; | |
402 | cpuid_mask(&entry->edx, 5); | |
403 | break; | |
404 | case 3: /* Processor serial number */ | |
405 | case 5: /* MONITOR/MWAIT */ | |
406 | case 6: /* Thermal management */ | |
407 | case 0xA: /* Architectural Performance Monitoring */ | |
408 | case 0x80000007: /* Advanced power management */ | |
409 | case 0xC0000002: | |
410 | case 0xC0000003: | |
411 | case 0xC0000004: | |
412 | default: | |
413 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
414 | break; | |
415 | } | |
416 | ||
417 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
418 | ||
419 | put_cpu(); | |
420 | } | |
421 | ||
422 | #undef F | |
423 | ||
424 | int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid, | |
425 | struct kvm_cpuid_entry2 __user *entries) | |
426 | { | |
427 | struct kvm_cpuid_entry2 *cpuid_entries; | |
428 | int limit, nent = 0, r = -E2BIG; | |
429 | u32 func; | |
430 | ||
431 | if (cpuid->nent < 1) | |
432 | goto out; | |
433 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
434 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
435 | r = -ENOMEM; | |
436 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); | |
437 | if (!cpuid_entries) | |
438 | goto out; | |
439 | ||
440 | do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent); | |
441 | limit = cpuid_entries[0].eax; | |
442 | for (func = 1; func <= limit && nent < cpuid->nent; ++func) | |
443 | do_cpuid_ent(&cpuid_entries[nent], func, 0, | |
444 | &nent, cpuid->nent); | |
445 | r = -E2BIG; | |
446 | if (nent >= cpuid->nent) | |
447 | goto out_free; | |
448 | ||
449 | do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent); | |
450 | limit = cpuid_entries[nent - 1].eax; | |
451 | for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func) | |
452 | do_cpuid_ent(&cpuid_entries[nent], func, 0, | |
453 | &nent, cpuid->nent); | |
454 | ||
455 | ||
456 | ||
457 | r = -E2BIG; | |
458 | if (nent >= cpuid->nent) | |
459 | goto out_free; | |
460 | ||
461 | /* Add support for Centaur's CPUID instruction. */ | |
462 | if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) { | |
463 | do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0, | |
464 | &nent, cpuid->nent); | |
465 | ||
466 | r = -E2BIG; | |
467 | if (nent >= cpuid->nent) | |
468 | goto out_free; | |
469 | ||
470 | limit = cpuid_entries[nent - 1].eax; | |
471 | for (func = 0xC0000001; | |
472 | func <= limit && nent < cpuid->nent; ++func) | |
473 | do_cpuid_ent(&cpuid_entries[nent], func, 0, | |
474 | &nent, cpuid->nent); | |
475 | ||
476 | r = -E2BIG; | |
477 | if (nent >= cpuid->nent) | |
478 | goto out_free; | |
479 | } | |
480 | ||
481 | do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent, | |
482 | cpuid->nent); | |
483 | ||
484 | r = -E2BIG; | |
485 | if (nent >= cpuid->nent) | |
486 | goto out_free; | |
487 | ||
488 | do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent, | |
489 | cpuid->nent); | |
490 | ||
491 | r = -E2BIG; | |
492 | if (nent >= cpuid->nent) | |
493 | goto out_free; | |
494 | ||
495 | r = -EFAULT; | |
496 | if (copy_to_user(entries, cpuid_entries, | |
497 | nent * sizeof(struct kvm_cpuid_entry2))) | |
498 | goto out_free; | |
499 | cpuid->nent = nent; | |
500 | r = 0; | |
501 | ||
502 | out_free: | |
503 | vfree(cpuid_entries); | |
504 | out: | |
505 | return r; | |
506 | } | |
507 | ||
508 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
509 | { | |
510 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
511 | int j, nent = vcpu->arch.cpuid_nent; | |
512 | ||
513 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
514 | /* when no next entry is found, the current entry[i] is reselected */ | |
515 | for (j = i + 1; ; j = (j + 1) % nent) { | |
516 | struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; | |
517 | if (ej->function == e->function) { | |
518 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
519 | return j; | |
520 | } | |
521 | } | |
522 | return 0; /* silence gcc, even though control never reaches here */ | |
523 | } | |
524 | ||
525 | /* find an entry with matching function, matching index (if needed), and that | |
526 | * should be read next (if it's stateful) */ | |
527 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
528 | u32 function, u32 index) | |
529 | { | |
530 | if (e->function != function) | |
531 | return 0; | |
532 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
533 | return 0; | |
534 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
535 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
536 | return 0; | |
537 | return 1; | |
538 | } | |
539 | ||
540 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
541 | u32 function, u32 index) | |
542 | { | |
543 | int i; | |
544 | struct kvm_cpuid_entry2 *best = NULL; | |
545 | ||
546 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
547 | struct kvm_cpuid_entry2 *e; | |
548 | ||
549 | e = &vcpu->arch.cpuid_entries[i]; | |
550 | if (is_matching_cpuid_entry(e, function, index)) { | |
551 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
552 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
553 | best = e; | |
554 | break; | |
555 | } | |
556 | } | |
557 | return best; | |
558 | } | |
559 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
560 | ||
561 | int cpuid_maxphyaddr(struct kvm_vcpu *vcpu) | |
562 | { | |
563 | struct kvm_cpuid_entry2 *best; | |
564 | ||
565 | best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); | |
566 | if (!best || best->eax < 0x80000008) | |
567 | goto not_found; | |
568 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
569 | if (best) | |
570 | return best->eax & 0xff; | |
571 | not_found: | |
572 | return 36; | |
573 | } | |
574 | ||
575 | /* | |
576 | * If no match is found, check whether we exceed the vCPU's limit | |
577 | * and return the content of the highest valid _standard_ leaf instead. | |
578 | * This is to satisfy the CPUID specification. | |
579 | */ | |
580 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
581 | u32 function, u32 index) | |
582 | { | |
583 | struct kvm_cpuid_entry2 *maxlevel; | |
584 | ||
585 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
586 | if (!maxlevel || maxlevel->eax >= function) | |
587 | return NULL; | |
588 | if (function & 0x80000000) { | |
589 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
590 | if (!maxlevel) | |
591 | return NULL; | |
592 | } | |
593 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
594 | } | |
595 | ||
596 | void kvm_emulate_cpuid(struct kvm_vcpu *vcpu) | |
597 | { | |
598 | u32 function, index; | |
599 | struct kvm_cpuid_entry2 *best; | |
600 | ||
601 | function = kvm_register_read(vcpu, VCPU_REGS_RAX); | |
602 | index = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
603 | kvm_register_write(vcpu, VCPU_REGS_RAX, 0); | |
604 | kvm_register_write(vcpu, VCPU_REGS_RBX, 0); | |
605 | kvm_register_write(vcpu, VCPU_REGS_RCX, 0); | |
606 | kvm_register_write(vcpu, VCPU_REGS_RDX, 0); | |
607 | best = kvm_find_cpuid_entry(vcpu, function, index); | |
608 | ||
609 | if (!best) | |
610 | best = check_cpuid_limit(vcpu, function, index); | |
611 | ||
612 | if (best) { | |
613 | kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax); | |
614 | kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx); | |
615 | kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx); | |
616 | kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx); | |
617 | } | |
618 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
619 | trace_kvm_cpuid(function, | |
620 | kvm_register_read(vcpu, VCPU_REGS_RAX), | |
621 | kvm_register_read(vcpu, VCPU_REGS_RBX), | |
622 | kvm_register_read(vcpu, VCPU_REGS_RCX), | |
623 | kvm_register_read(vcpu, VCPU_REGS_RDX)); | |
624 | } | |
625 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |