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 | ||
412a3c41 | 26 | static u32 xstate_required_size(u64 xstate_bv, bool compacted) |
4344ee98 PB |
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) { | |
412a3c41 | 34 | u32 eax, ebx, ecx, edx, offset; |
4344ee98 | 35 | cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); |
412a3c41 PB |
36 | offset = compacted ? ret : ebx; |
37 | ret = max(ret, offset + eax); | |
4344ee98 PB |
38 | } |
39 | ||
40 | xstate_bv >>= 1; | |
41 | feature_bit++; | |
42 | } | |
43 | ||
44 | return ret; | |
45 | } | |
46 | ||
4ff41732 PB |
47 | u64 kvm_supported_xcr0(void) |
48 | { | |
49 | u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; | |
50 | ||
93c4adc7 | 51 | if (!kvm_x86_ops->mpx_supported()) |
4ff41732 PB |
52 | xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR); |
53 | ||
54 | return xcr0; | |
55 | } | |
56 | ||
5c404cab PB |
57 | #define F(x) bit(X86_FEATURE_##x) |
58 | ||
dd598091 | 59 | int kvm_update_cpuid(struct kvm_vcpu *vcpu) |
00b27a3e AK |
60 | { |
61 | struct kvm_cpuid_entry2 *best; | |
62 | struct kvm_lapic *apic = vcpu->arch.apic; | |
63 | ||
64 | best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
65 | if (!best) | |
dd598091 | 66 | return 0; |
00b27a3e AK |
67 | |
68 | /* Update OSXSAVE bit */ | |
69 | if (cpu_has_xsave && best->function == 0x1) { | |
5c404cab | 70 | best->ecx &= ~F(OSXSAVE); |
00b27a3e | 71 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) |
5c404cab | 72 | best->ecx |= F(OSXSAVE); |
00b27a3e AK |
73 | } |
74 | ||
75 | if (apic) { | |
5c404cab | 76 | if (best->ecx & F(TSC_DEADLINE_TIMER)) |
00b27a3e AK |
77 | apic->lapic_timer.timer_mode_mask = 3 << 17; |
78 | else | |
79 | apic->lapic_timer.timer_mode_mask = 1 << 17; | |
80 | } | |
f5132b01 | 81 | |
d7876f1b | 82 | best = kvm_find_cpuid_entry(vcpu, 0xD, 0); |
4344ee98 | 83 | if (!best) { |
d7876f1b | 84 | vcpu->arch.guest_supported_xcr0 = 0; |
4344ee98 PB |
85 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
86 | } else { | |
d7876f1b PB |
87 | vcpu->arch.guest_supported_xcr0 = |
88 | (best->eax | ((u64)best->edx << 32)) & | |
4ff41732 | 89 | kvm_supported_xcr0(); |
56c103ec | 90 | vcpu->arch.guest_xstate_size = best->ebx = |
412a3c41 | 91 | xstate_required_size(vcpu->arch.xcr0, false); |
4344ee98 | 92 | } |
d7876f1b | 93 | |
412a3c41 PB |
94 | best = kvm_find_cpuid_entry(vcpu, 0xD, 1); |
95 | if (best && (best->eax & (F(XSAVES) | F(XSAVEC)))) | |
96 | best->ebx = xstate_required_size(vcpu->arch.xcr0, true); | |
97 | ||
dd598091 NA |
98 | /* |
99 | * The existing code assumes virtual address is 48-bit in the canonical | |
100 | * address checks; exit if it is ever changed. | |
101 | */ | |
102 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
103 | if (best && ((best->eax & 0xff00) >> 8) != 48 && | |
104 | ((best->eax & 0xff00) >> 8) != 0) | |
105 | return -EINVAL; | |
106 | ||
f5132b01 | 107 | kvm_pmu_cpuid_update(vcpu); |
dd598091 | 108 | return 0; |
00b27a3e AK |
109 | } |
110 | ||
111 | static int is_efer_nx(void) | |
112 | { | |
113 | unsigned long long efer = 0; | |
114 | ||
115 | rdmsrl_safe(MSR_EFER, &efer); | |
116 | return efer & EFER_NX; | |
117 | } | |
118 | ||
119 | static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) | |
120 | { | |
121 | int i; | |
122 | struct kvm_cpuid_entry2 *e, *entry; | |
123 | ||
124 | entry = NULL; | |
125 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
126 | e = &vcpu->arch.cpuid_entries[i]; | |
127 | if (e->function == 0x80000001) { | |
128 | entry = e; | |
129 | break; | |
130 | } | |
131 | } | |
5c404cab PB |
132 | if (entry && (entry->edx & F(NX)) && !is_efer_nx()) { |
133 | entry->edx &= ~F(NX); | |
00b27a3e AK |
134 | printk(KERN_INFO "kvm: guest NX capability removed\n"); |
135 | } | |
136 | } | |
137 | ||
138 | /* when an old userspace process fills a new kernel module */ | |
139 | int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, | |
140 | struct kvm_cpuid *cpuid, | |
141 | struct kvm_cpuid_entry __user *entries) | |
142 | { | |
143 | int r, i; | |
144 | struct kvm_cpuid_entry *cpuid_entries; | |
145 | ||
146 | r = -E2BIG; | |
147 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
148 | goto out; | |
149 | r = -ENOMEM; | |
150 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent); | |
151 | if (!cpuid_entries) | |
152 | goto out; | |
153 | r = -EFAULT; | |
154 | if (copy_from_user(cpuid_entries, entries, | |
155 | cpuid->nent * sizeof(struct kvm_cpuid_entry))) | |
156 | goto out_free; | |
157 | for (i = 0; i < cpuid->nent; i++) { | |
158 | vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; | |
159 | vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; | |
160 | vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; | |
161 | vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; | |
162 | vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; | |
163 | vcpu->arch.cpuid_entries[i].index = 0; | |
164 | vcpu->arch.cpuid_entries[i].flags = 0; | |
165 | vcpu->arch.cpuid_entries[i].padding[0] = 0; | |
166 | vcpu->arch.cpuid_entries[i].padding[1] = 0; | |
167 | vcpu->arch.cpuid_entries[i].padding[2] = 0; | |
168 | } | |
169 | vcpu->arch.cpuid_nent = cpuid->nent; | |
170 | cpuid_fix_nx_cap(vcpu); | |
00b27a3e AK |
171 | kvm_apic_set_version(vcpu); |
172 | kvm_x86_ops->cpuid_update(vcpu); | |
dd598091 | 173 | r = kvm_update_cpuid(vcpu); |
00b27a3e AK |
174 | |
175 | out_free: | |
176 | vfree(cpuid_entries); | |
177 | out: | |
178 | return r; | |
179 | } | |
180 | ||
181 | int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, | |
182 | struct kvm_cpuid2 *cpuid, | |
183 | struct kvm_cpuid_entry2 __user *entries) | |
184 | { | |
185 | int r; | |
186 | ||
187 | r = -E2BIG; | |
188 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
189 | goto out; | |
190 | r = -EFAULT; | |
191 | if (copy_from_user(&vcpu->arch.cpuid_entries, entries, | |
192 | cpuid->nent * sizeof(struct kvm_cpuid_entry2))) | |
193 | goto out; | |
194 | vcpu->arch.cpuid_nent = cpuid->nent; | |
195 | kvm_apic_set_version(vcpu); | |
196 | kvm_x86_ops->cpuid_update(vcpu); | |
dd598091 | 197 | r = kvm_update_cpuid(vcpu); |
00b27a3e AK |
198 | out: |
199 | return r; | |
200 | } | |
201 | ||
202 | int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, | |
203 | struct kvm_cpuid2 *cpuid, | |
204 | struct kvm_cpuid_entry2 __user *entries) | |
205 | { | |
206 | int r; | |
207 | ||
208 | r = -E2BIG; | |
209 | if (cpuid->nent < vcpu->arch.cpuid_nent) | |
210 | goto out; | |
211 | r = -EFAULT; | |
212 | if (copy_to_user(entries, &vcpu->arch.cpuid_entries, | |
213 | vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) | |
214 | goto out; | |
215 | return 0; | |
216 | ||
217 | out: | |
218 | cpuid->nent = vcpu->arch.cpuid_nent; | |
219 | return r; | |
220 | } | |
221 | ||
222 | static void cpuid_mask(u32 *word, int wordnum) | |
223 | { | |
224 | *word &= boot_cpu_data.x86_capability[wordnum]; | |
225 | } | |
226 | ||
227 | static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
228 | u32 index) | |
229 | { | |
230 | entry->function = function; | |
231 | entry->index = index; | |
232 | cpuid_count(entry->function, entry->index, | |
233 | &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); | |
234 | entry->flags = 0; | |
235 | } | |
236 | ||
9c15bb1d BP |
237 | static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, |
238 | u32 func, u32 index, int *nent, int maxnent) | |
239 | { | |
84cffe49 BP |
240 | switch (func) { |
241 | case 0: | |
242 | entry->eax = 1; /* only one leaf currently */ | |
243 | ++*nent; | |
244 | break; | |
245 | case 1: | |
246 | entry->ecx = F(MOVBE); | |
247 | ++*nent; | |
248 | break; | |
249 | default: | |
250 | break; | |
251 | } | |
252 | ||
253 | entry->function = func; | |
254 | entry->index = index; | |
255 | ||
9c15bb1d BP |
256 | return 0; |
257 | } | |
258 | ||
259 | static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, | |
260 | u32 index, int *nent, int maxnent) | |
00b27a3e | 261 | { |
831bf664 | 262 | int r; |
00b27a3e AK |
263 | unsigned f_nx = is_efer_nx() ? F(NX) : 0; |
264 | #ifdef CONFIG_X86_64 | |
265 | unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) | |
266 | ? F(GBPAGES) : 0; | |
267 | unsigned f_lm = F(LM); | |
268 | #else | |
269 | unsigned f_gbpages = 0; | |
270 | unsigned f_lm = 0; | |
271 | #endif | |
272 | unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; | |
ad756a16 | 273 | unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; |
93c4adc7 | 274 | unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0; |
55412b2e | 275 | unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; |
00b27a3e AK |
276 | |
277 | /* cpuid 1.edx */ | |
278 | const u32 kvm_supported_word0_x86_features = | |
279 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
280 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
281 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | | |
282 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
840d2830 | 283 | F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | |
00b27a3e AK |
284 | 0 /* Reserved, DS, ACPI */ | F(MMX) | |
285 | F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | | |
286 | 0 /* HTT, TM, Reserved, PBE */; | |
287 | /* cpuid 0x80000001.edx */ | |
288 | const u32 kvm_supported_word1_x86_features = | |
289 | F(FPU) | F(VME) | F(DE) | F(PSE) | | |
290 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | | |
291 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | | |
292 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | | |
293 | F(PAT) | F(PSE36) | 0 /* Reserved */ | | |
294 | f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | | |
295 | F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | | |
296 | 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); | |
297 | /* cpuid 1.ecx */ | |
298 | const u32 kvm_supported_word4_x86_features = | |
87c00572 GS |
299 | /* NOTE: MONITOR (and MWAIT) are emulated as NOP, |
300 | * but *not* advertised to guests via CPUID ! */ | |
00b27a3e AK |
301 | F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | |
302 | 0 /* DS-CPL, VMX, SMX, EST */ | | |
303 | 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | | |
fb215366 | 304 | F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | |
ad756a16 | 305 | F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | |
00b27a3e AK |
306 | F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | |
307 | 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | | |
308 | F(F16C) | F(RDRAND); | |
309 | /* cpuid 0x80000001.ecx */ | |
310 | const u32 kvm_supported_word6_x86_features = | |
311 | F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | | |
312 | F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | | |
2b036c6b | 313 | F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | |
00b27a3e AK |
314 | 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); |
315 | ||
316 | /* cpuid 0xC0000001.edx */ | |
317 | const u32 kvm_supported_word5_x86_features = | |
318 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | | |
319 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | | |
320 | F(PMM) | F(PMM_EN); | |
321 | ||
322 | /* cpuid 7.0.ebx */ | |
323 | const u32 kvm_supported_word9_x86_features = | |
83c52915 | 324 | F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | |
390bd528 | 325 | F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | |
612263b3 CP |
326 | F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) | |
327 | F(AVX512CD); | |
00b27a3e | 328 | |
b65d6e17 PB |
329 | /* cpuid 0xD.1.eax */ |
330 | const u32 kvm_supported_word10_x86_features = | |
55412b2e | 331 | F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; |
b65d6e17 | 332 | |
00b27a3e AK |
333 | /* all calls to cpuid_count() should be made on the same cpu */ |
334 | get_cpu(); | |
831bf664 SL |
335 | |
336 | r = -E2BIG; | |
337 | ||
338 | if (*nent >= maxnent) | |
339 | goto out; | |
340 | ||
00b27a3e AK |
341 | do_cpuid_1_ent(entry, function, index); |
342 | ++*nent; | |
343 | ||
344 | switch (function) { | |
345 | case 0: | |
346 | entry->eax = min(entry->eax, (u32)0xd); | |
347 | break; | |
348 | case 1: | |
349 | entry->edx &= kvm_supported_word0_x86_features; | |
350 | cpuid_mask(&entry->edx, 0); | |
351 | entry->ecx &= kvm_supported_word4_x86_features; | |
352 | cpuid_mask(&entry->ecx, 4); | |
353 | /* we support x2apic emulation even if host does not support | |
354 | * it since we emulate x2apic in software */ | |
355 | entry->ecx |= F(X2APIC); | |
356 | break; | |
357 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands | |
358 | * may return different values. This forces us to get_cpu() before | |
359 | * issuing the first command, and also to emulate this annoying behavior | |
360 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ | |
361 | case 2: { | |
362 | int t, times = entry->eax & 0xff; | |
363 | ||
364 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
365 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
831bf664 SL |
366 | for (t = 1; t < times; ++t) { |
367 | if (*nent >= maxnent) | |
368 | goto out; | |
369 | ||
00b27a3e AK |
370 | do_cpuid_1_ent(&entry[t], function, 0); |
371 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; | |
372 | ++*nent; | |
373 | } | |
374 | break; | |
375 | } | |
376 | /* function 4 has additional index. */ | |
377 | case 4: { | |
378 | int i, cache_type; | |
379 | ||
380 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
381 | /* read more entries until cache_type is zero */ | |
831bf664 SL |
382 | for (i = 1; ; ++i) { |
383 | if (*nent >= maxnent) | |
384 | goto out; | |
385 | ||
00b27a3e AK |
386 | cache_type = entry[i - 1].eax & 0x1f; |
387 | if (!cache_type) | |
388 | break; | |
389 | do_cpuid_1_ent(&entry[i], function, i); | |
390 | entry[i].flags |= | |
391 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
392 | ++*nent; | |
393 | } | |
394 | break; | |
395 | } | |
396 | case 7: { | |
397 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
bbbda795 | 398 | /* Mask ebx against host capability word 9 */ |
00b27a3e AK |
399 | if (index == 0) { |
400 | entry->ebx &= kvm_supported_word9_x86_features; | |
401 | cpuid_mask(&entry->ebx, 9); | |
ba904635 WA |
402 | // TSC_ADJUST is emulated |
403 | entry->ebx |= F(TSC_ADJUST); | |
00b27a3e AK |
404 | } else |
405 | entry->ebx = 0; | |
406 | entry->eax = 0; | |
407 | entry->ecx = 0; | |
408 | entry->edx = 0; | |
409 | break; | |
410 | } | |
411 | case 9: | |
412 | break; | |
a6c06ed1 GN |
413 | case 0xa: { /* Architectural Performance Monitoring */ |
414 | struct x86_pmu_capability cap; | |
415 | union cpuid10_eax eax; | |
416 | union cpuid10_edx edx; | |
417 | ||
418 | perf_get_x86_pmu_capability(&cap); | |
419 | ||
420 | /* | |
421 | * Only support guest architectural pmu on a host | |
422 | * with architectural pmu. | |
423 | */ | |
424 | if (!cap.version) | |
425 | memset(&cap, 0, sizeof(cap)); | |
426 | ||
427 | eax.split.version_id = min(cap.version, 2); | |
428 | eax.split.num_counters = cap.num_counters_gp; | |
429 | eax.split.bit_width = cap.bit_width_gp; | |
430 | eax.split.mask_length = cap.events_mask_len; | |
431 | ||
432 | edx.split.num_counters_fixed = cap.num_counters_fixed; | |
433 | edx.split.bit_width_fixed = cap.bit_width_fixed; | |
434 | edx.split.reserved = 0; | |
435 | ||
436 | entry->eax = eax.full; | |
437 | entry->ebx = cap.events_mask; | |
438 | entry->ecx = 0; | |
439 | entry->edx = edx.full; | |
440 | break; | |
441 | } | |
00b27a3e AK |
442 | /* function 0xb has additional index. */ |
443 | case 0xb: { | |
444 | int i, level_type; | |
445 | ||
446 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
447 | /* read more entries until level_type is zero */ | |
831bf664 SL |
448 | for (i = 1; ; ++i) { |
449 | if (*nent >= maxnent) | |
450 | goto out; | |
451 | ||
00b27a3e AK |
452 | level_type = entry[i - 1].ecx & 0xff00; |
453 | if (!level_type) | |
454 | break; | |
455 | do_cpuid_1_ent(&entry[i], function, i); | |
456 | entry[i].flags |= | |
457 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
458 | ++*nent; | |
459 | } | |
460 | break; | |
461 | } | |
462 | case 0xd: { | |
463 | int idx, i; | |
4ff41732 | 464 | u64 supported = kvm_supported_xcr0(); |
00b27a3e | 465 | |
4ff41732 PB |
466 | entry->eax &= supported; |
467 | entry->edx &= supported >> 32; | |
00b27a3e | 468 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
b65d6e17 PB |
469 | if (!supported) |
470 | break; | |
471 | ||
831bf664 | 472 | for (idx = 1, i = 1; idx < 64; ++idx) { |
4ff41732 | 473 | u64 mask = ((u64)1 << idx); |
831bf664 SL |
474 | if (*nent >= maxnent) |
475 | goto out; | |
476 | ||
00b27a3e | 477 | do_cpuid_1_ent(&entry[i], function, idx); |
412a3c41 | 478 | if (idx == 1) { |
b65d6e17 | 479 | entry[i].eax &= kvm_supported_word10_x86_features; |
412a3c41 PB |
480 | entry[i].ebx = 0; |
481 | if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) | |
482 | entry[i].ebx = | |
483 | xstate_required_size(supported, | |
484 | true); | |
485 | } else if (entry[i].eax == 0 || !(supported & mask)) | |
00b27a3e AK |
486 | continue; |
487 | entry[i].flags |= | |
488 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; | |
489 | ++*nent; | |
490 | ++i; | |
491 | } | |
492 | break; | |
493 | } | |
494 | case KVM_CPUID_SIGNATURE: { | |
326d07cb MK |
495 | static const char signature[12] = "KVMKVMKVM\0\0"; |
496 | const u32 *sigptr = (const u32 *)signature; | |
57c22e5f | 497 | entry->eax = KVM_CPUID_FEATURES; |
00b27a3e AK |
498 | entry->ebx = sigptr[0]; |
499 | entry->ecx = sigptr[1]; | |
500 | entry->edx = sigptr[2]; | |
501 | break; | |
502 | } | |
503 | case KVM_CPUID_FEATURES: | |
504 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | | |
505 | (1 << KVM_FEATURE_NOP_IO_DELAY) | | |
506 | (1 << KVM_FEATURE_CLOCKSOURCE2) | | |
507 | (1 << KVM_FEATURE_ASYNC_PF) | | |
ae7a2a3f | 508 | (1 << KVM_FEATURE_PV_EOI) | |
6aef266c SV |
509 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | |
510 | (1 << KVM_FEATURE_PV_UNHALT); | |
00b27a3e AK |
511 | |
512 | if (sched_info_on()) | |
513 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); | |
514 | ||
515 | entry->ebx = 0; | |
516 | entry->ecx = 0; | |
517 | entry->edx = 0; | |
518 | break; | |
519 | case 0x80000000: | |
520 | entry->eax = min(entry->eax, 0x8000001a); | |
521 | break; | |
522 | case 0x80000001: | |
523 | entry->edx &= kvm_supported_word1_x86_features; | |
524 | cpuid_mask(&entry->edx, 1); | |
525 | entry->ecx &= kvm_supported_word6_x86_features; | |
526 | cpuid_mask(&entry->ecx, 6); | |
527 | break; | |
e4c9a5a1 MT |
528 | case 0x80000007: /* Advanced power management */ |
529 | /* invariant TSC is CPUID.80000007H:EDX[8] */ | |
530 | entry->edx &= (1 << 8); | |
531 | /* mask against host */ | |
532 | entry->edx &= boot_cpu_data.x86_power; | |
533 | entry->eax = entry->ebx = entry->ecx = 0; | |
534 | break; | |
00b27a3e AK |
535 | case 0x80000008: { |
536 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; | |
537 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); | |
538 | unsigned phys_as = entry->eax & 0xff; | |
539 | ||
540 | if (!g_phys_as) | |
541 | g_phys_as = phys_as; | |
542 | entry->eax = g_phys_as | (virt_as << 8); | |
543 | entry->ebx = entry->edx = 0; | |
544 | break; | |
545 | } | |
546 | case 0x80000019: | |
547 | entry->ecx = entry->edx = 0; | |
548 | break; | |
549 | case 0x8000001a: | |
550 | break; | |
551 | case 0x8000001d: | |
552 | break; | |
553 | /*Add support for Centaur's CPUID instruction*/ | |
554 | case 0xC0000000: | |
555 | /*Just support up to 0xC0000004 now*/ | |
556 | entry->eax = min(entry->eax, 0xC0000004); | |
557 | break; | |
558 | case 0xC0000001: | |
559 | entry->edx &= kvm_supported_word5_x86_features; | |
560 | cpuid_mask(&entry->edx, 5); | |
561 | break; | |
562 | case 3: /* Processor serial number */ | |
563 | case 5: /* MONITOR/MWAIT */ | |
564 | case 6: /* Thermal management */ | |
00b27a3e AK |
565 | case 0xC0000002: |
566 | case 0xC0000003: | |
567 | case 0xC0000004: | |
568 | default: | |
569 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; | |
570 | break; | |
571 | } | |
572 | ||
573 | kvm_x86_ops->set_supported_cpuid(function, entry); | |
574 | ||
831bf664 SL |
575 | r = 0; |
576 | ||
577 | out: | |
00b27a3e | 578 | put_cpu(); |
831bf664 SL |
579 | |
580 | return r; | |
00b27a3e AK |
581 | } |
582 | ||
9c15bb1d BP |
583 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, |
584 | u32 idx, int *nent, int maxnent, unsigned int type) | |
585 | { | |
586 | if (type == KVM_GET_EMULATED_CPUID) | |
587 | return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); | |
588 | ||
589 | return __do_cpuid_ent(entry, func, idx, nent, maxnent); | |
590 | } | |
591 | ||
00b27a3e AK |
592 | #undef F |
593 | ||
831bf664 SL |
594 | struct kvm_cpuid_param { |
595 | u32 func; | |
596 | u32 idx; | |
597 | bool has_leaf_count; | |
326d07cb | 598 | bool (*qualifier)(const struct kvm_cpuid_param *param); |
831bf664 SL |
599 | }; |
600 | ||
326d07cb | 601 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) |
831bf664 SL |
602 | { |
603 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; | |
604 | } | |
605 | ||
9c15bb1d BP |
606 | static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, |
607 | __u32 num_entries, unsigned int ioctl_type) | |
608 | { | |
609 | int i; | |
1b2ca422 | 610 | __u32 pad[3]; |
9c15bb1d BP |
611 | |
612 | if (ioctl_type != KVM_GET_EMULATED_CPUID) | |
613 | return false; | |
614 | ||
615 | /* | |
616 | * We want to make sure that ->padding is being passed clean from | |
617 | * userspace in case we want to use it for something in the future. | |
618 | * | |
619 | * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we | |
620 | * have to give ourselves satisfied only with the emulated side. /me | |
621 | * sheds a tear. | |
622 | */ | |
623 | for (i = 0; i < num_entries; i++) { | |
1b2ca422 BP |
624 | if (copy_from_user(pad, entries[i].padding, sizeof(pad))) |
625 | return true; | |
626 | ||
627 | if (pad[0] || pad[1] || pad[2]) | |
9c15bb1d BP |
628 | return true; |
629 | } | |
630 | return false; | |
631 | } | |
632 | ||
633 | int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, | |
634 | struct kvm_cpuid_entry2 __user *entries, | |
635 | unsigned int type) | |
00b27a3e AK |
636 | { |
637 | struct kvm_cpuid_entry2 *cpuid_entries; | |
831bf664 | 638 | int limit, nent = 0, r = -E2BIG, i; |
00b27a3e | 639 | u32 func; |
326d07cb | 640 | static const struct kvm_cpuid_param param[] = { |
831bf664 SL |
641 | { .func = 0, .has_leaf_count = true }, |
642 | { .func = 0x80000000, .has_leaf_count = true }, | |
643 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, | |
644 | { .func = KVM_CPUID_SIGNATURE }, | |
645 | { .func = KVM_CPUID_FEATURES }, | |
646 | }; | |
00b27a3e AK |
647 | |
648 | if (cpuid->nent < 1) | |
649 | goto out; | |
650 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) | |
651 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; | |
9c15bb1d BP |
652 | |
653 | if (sanity_check_entries(entries, cpuid->nent, type)) | |
654 | return -EINVAL; | |
655 | ||
00b27a3e | 656 | r = -ENOMEM; |
84cffe49 | 657 | cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); |
00b27a3e AK |
658 | if (!cpuid_entries) |
659 | goto out; | |
660 | ||
831bf664 SL |
661 | r = 0; |
662 | for (i = 0; i < ARRAY_SIZE(param); i++) { | |
326d07cb | 663 | const struct kvm_cpuid_param *ent = ¶m[i]; |
00b27a3e | 664 | |
831bf664 SL |
665 | if (ent->qualifier && !ent->qualifier(ent)) |
666 | continue; | |
00b27a3e | 667 | |
831bf664 | 668 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, |
9c15bb1d | 669 | &nent, cpuid->nent, type); |
00b27a3e | 670 | |
831bf664 | 671 | if (r) |
00b27a3e AK |
672 | goto out_free; |
673 | ||
831bf664 SL |
674 | if (!ent->has_leaf_count) |
675 | continue; | |
676 | ||
00b27a3e | 677 | limit = cpuid_entries[nent - 1].eax; |
831bf664 SL |
678 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) |
679 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, | |
9c15bb1d | 680 | &nent, cpuid->nent, type); |
00b27a3e | 681 | |
831bf664 | 682 | if (r) |
00b27a3e AK |
683 | goto out_free; |
684 | } | |
685 | ||
00b27a3e AK |
686 | r = -EFAULT; |
687 | if (copy_to_user(entries, cpuid_entries, | |
688 | nent * sizeof(struct kvm_cpuid_entry2))) | |
689 | goto out_free; | |
690 | cpuid->nent = nent; | |
691 | r = 0; | |
692 | ||
693 | out_free: | |
694 | vfree(cpuid_entries); | |
695 | out: | |
696 | return r; | |
697 | } | |
698 | ||
699 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) | |
700 | { | |
701 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; | |
702 | int j, nent = vcpu->arch.cpuid_nent; | |
703 | ||
704 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; | |
705 | /* when no next entry is found, the current entry[i] is reselected */ | |
706 | for (j = i + 1; ; j = (j + 1) % nent) { | |
707 | struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; | |
708 | if (ej->function == e->function) { | |
709 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; | |
710 | return j; | |
711 | } | |
712 | } | |
713 | return 0; /* silence gcc, even though control never reaches here */ | |
714 | } | |
715 | ||
716 | /* find an entry with matching function, matching index (if needed), and that | |
717 | * should be read next (if it's stateful) */ | |
718 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, | |
719 | u32 function, u32 index) | |
720 | { | |
721 | if (e->function != function) | |
722 | return 0; | |
723 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) | |
724 | return 0; | |
725 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && | |
726 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) | |
727 | return 0; | |
728 | return 1; | |
729 | } | |
730 | ||
731 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, | |
732 | u32 function, u32 index) | |
733 | { | |
734 | int i; | |
735 | struct kvm_cpuid_entry2 *best = NULL; | |
736 | ||
737 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { | |
738 | struct kvm_cpuid_entry2 *e; | |
739 | ||
740 | e = &vcpu->arch.cpuid_entries[i]; | |
741 | if (is_matching_cpuid_entry(e, function, index)) { | |
742 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) | |
743 | move_to_next_stateful_cpuid_entry(vcpu, i); | |
744 | best = e; | |
745 | break; | |
746 | } | |
747 | } | |
748 | return best; | |
749 | } | |
750 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); | |
751 | ||
752 | int cpuid_maxphyaddr(struct kvm_vcpu *vcpu) | |
753 | { | |
754 | struct kvm_cpuid_entry2 *best; | |
755 | ||
756 | best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); | |
757 | if (!best || best->eax < 0x80000008) | |
758 | goto not_found; | |
759 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); | |
760 | if (best) | |
761 | return best->eax & 0xff; | |
762 | not_found: | |
763 | return 36; | |
764 | } | |
3573e22c | 765 | EXPORT_SYMBOL_GPL(cpuid_maxphyaddr); |
00b27a3e AK |
766 | |
767 | /* | |
768 | * If no match is found, check whether we exceed the vCPU's limit | |
769 | * and return the content of the highest valid _standard_ leaf instead. | |
770 | * This is to satisfy the CPUID specification. | |
771 | */ | |
772 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, | |
773 | u32 function, u32 index) | |
774 | { | |
775 | struct kvm_cpuid_entry2 *maxlevel; | |
776 | ||
777 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); | |
778 | if (!maxlevel || maxlevel->eax >= function) | |
779 | return NULL; | |
780 | if (function & 0x80000000) { | |
781 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); | |
782 | if (!maxlevel) | |
783 | return NULL; | |
784 | } | |
785 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); | |
786 | } | |
787 | ||
62046e5a | 788 | void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
00b27a3e | 789 | { |
62046e5a | 790 | u32 function = *eax, index = *ecx; |
00b27a3e AK |
791 | struct kvm_cpuid_entry2 *best; |
792 | ||
00b27a3e AK |
793 | best = kvm_find_cpuid_entry(vcpu, function, index); |
794 | ||
795 | if (!best) | |
796 | best = check_cpuid_limit(vcpu, function, index); | |
797 | ||
bc613494 MT |
798 | /* |
799 | * Perfmon not yet supported for L2 guest. | |
800 | */ | |
801 | if (is_guest_mode(vcpu) && function == 0xa) | |
802 | best = NULL; | |
803 | ||
00b27a3e | 804 | if (best) { |
62046e5a AK |
805 | *eax = best->eax; |
806 | *ebx = best->ebx; | |
807 | *ecx = best->ecx; | |
808 | *edx = best->edx; | |
809 | } else | |
810 | *eax = *ebx = *ecx = *edx = 0; | |
a9d4e439 | 811 | trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx); |
62046e5a | 812 | } |
66f7b72e | 813 | EXPORT_SYMBOL_GPL(kvm_cpuid); |
62046e5a AK |
814 | |
815 | void kvm_emulate_cpuid(struct kvm_vcpu *vcpu) | |
816 | { | |
817 | u32 function, eax, ebx, ecx, edx; | |
818 | ||
819 | function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX); | |
820 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
821 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx); | |
822 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); | |
823 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); | |
824 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); | |
825 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); | |
00b27a3e | 826 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
00b27a3e AK |
827 | } |
828 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |