Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 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> |
Jan Kiszka | bb5a798 | 2011-12-14 17:58:18 +0100 | [diff] [blame] | 17 | #include <linux/vmalloc.h> |
| 18 | #include <linux/uaccess.h> |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 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 | |
| 26 | void kvm_update_cpuid(struct kvm_vcpu *vcpu) |
| 27 | { |
| 28 | struct kvm_cpuid_entry2 *best; |
| 29 | struct kvm_lapic *apic = vcpu->arch.apic; |
| 30 | |
| 31 | best = kvm_find_cpuid_entry(vcpu, 1, 0); |
| 32 | if (!best) |
| 33 | return; |
| 34 | |
| 35 | /* Update OSXSAVE bit */ |
| 36 | if (cpu_has_xsave && best->function == 0x1) { |
| 37 | best->ecx &= ~(bit(X86_FEATURE_OSXSAVE)); |
| 38 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) |
| 39 | best->ecx |= bit(X86_FEATURE_OSXSAVE); |
| 40 | } |
| 41 | |
| 42 | if (apic) { |
| 43 | if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER)) |
| 44 | apic->lapic_timer.timer_mode_mask = 3 << 17; |
| 45 | else |
| 46 | apic->lapic_timer.timer_mode_mask = 1 << 17; |
| 47 | } |
Gleb Natapov | f5132b0 | 2011-11-10 14:57:22 +0200 | [diff] [blame] | 48 | |
| 49 | kvm_pmu_cpuid_update(vcpu); |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 50 | } |
| 51 | |
| 52 | static int is_efer_nx(void) |
| 53 | { |
| 54 | unsigned long long efer = 0; |
| 55 | |
| 56 | rdmsrl_safe(MSR_EFER, &efer); |
| 57 | return efer & EFER_NX; |
| 58 | } |
| 59 | |
| 60 | static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) |
| 61 | { |
| 62 | int i; |
| 63 | struct kvm_cpuid_entry2 *e, *entry; |
| 64 | |
| 65 | entry = NULL; |
| 66 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { |
| 67 | e = &vcpu->arch.cpuid_entries[i]; |
| 68 | if (e->function == 0x80000001) { |
| 69 | entry = e; |
| 70 | break; |
| 71 | } |
| 72 | } |
| 73 | if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) { |
| 74 | entry->edx &= ~(1 << 20); |
| 75 | printk(KERN_INFO "kvm: guest NX capability removed\n"); |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | /* when an old userspace process fills a new kernel module */ |
| 80 | int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, |
| 81 | struct kvm_cpuid *cpuid, |
| 82 | struct kvm_cpuid_entry __user *entries) |
| 83 | { |
| 84 | int r, i; |
| 85 | struct kvm_cpuid_entry *cpuid_entries; |
| 86 | |
| 87 | r = -E2BIG; |
| 88 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) |
| 89 | goto out; |
| 90 | r = -ENOMEM; |
| 91 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent); |
| 92 | if (!cpuid_entries) |
| 93 | goto out; |
| 94 | r = -EFAULT; |
| 95 | if (copy_from_user(cpuid_entries, entries, |
| 96 | cpuid->nent * sizeof(struct kvm_cpuid_entry))) |
| 97 | goto out_free; |
| 98 | for (i = 0; i < cpuid->nent; i++) { |
| 99 | vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; |
| 100 | vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; |
| 101 | vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; |
| 102 | vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; |
| 103 | vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; |
| 104 | vcpu->arch.cpuid_entries[i].index = 0; |
| 105 | vcpu->arch.cpuid_entries[i].flags = 0; |
| 106 | vcpu->arch.cpuid_entries[i].padding[0] = 0; |
| 107 | vcpu->arch.cpuid_entries[i].padding[1] = 0; |
| 108 | vcpu->arch.cpuid_entries[i].padding[2] = 0; |
| 109 | } |
| 110 | vcpu->arch.cpuid_nent = cpuid->nent; |
| 111 | cpuid_fix_nx_cap(vcpu); |
| 112 | r = 0; |
| 113 | kvm_apic_set_version(vcpu); |
| 114 | kvm_x86_ops->cpuid_update(vcpu); |
| 115 | kvm_update_cpuid(vcpu); |
| 116 | |
| 117 | out_free: |
| 118 | vfree(cpuid_entries); |
| 119 | out: |
| 120 | return r; |
| 121 | } |
| 122 | |
| 123 | int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, |
| 124 | struct kvm_cpuid2 *cpuid, |
| 125 | struct kvm_cpuid_entry2 __user *entries) |
| 126 | { |
| 127 | int r; |
| 128 | |
| 129 | r = -E2BIG; |
| 130 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) |
| 131 | goto out; |
| 132 | r = -EFAULT; |
| 133 | if (copy_from_user(&vcpu->arch.cpuid_entries, entries, |
| 134 | cpuid->nent * sizeof(struct kvm_cpuid_entry2))) |
| 135 | goto out; |
| 136 | vcpu->arch.cpuid_nent = cpuid->nent; |
| 137 | kvm_apic_set_version(vcpu); |
| 138 | kvm_x86_ops->cpuid_update(vcpu); |
| 139 | kvm_update_cpuid(vcpu); |
| 140 | return 0; |
| 141 | |
| 142 | out: |
| 143 | return r; |
| 144 | } |
| 145 | |
| 146 | int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, |
| 147 | struct kvm_cpuid2 *cpuid, |
| 148 | struct kvm_cpuid_entry2 __user *entries) |
| 149 | { |
| 150 | int r; |
| 151 | |
| 152 | r = -E2BIG; |
| 153 | if (cpuid->nent < vcpu->arch.cpuid_nent) |
| 154 | goto out; |
| 155 | r = -EFAULT; |
| 156 | if (copy_to_user(entries, &vcpu->arch.cpuid_entries, |
| 157 | vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) |
| 158 | goto out; |
| 159 | return 0; |
| 160 | |
| 161 | out: |
| 162 | cpuid->nent = vcpu->arch.cpuid_nent; |
| 163 | return r; |
| 164 | } |
| 165 | |
| 166 | static void cpuid_mask(u32 *word, int wordnum) |
| 167 | { |
| 168 | *word &= boot_cpu_data.x86_capability[wordnum]; |
| 169 | } |
| 170 | |
| 171 | static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, |
| 172 | u32 index) |
| 173 | { |
| 174 | entry->function = function; |
| 175 | entry->index = index; |
| 176 | cpuid_count(entry->function, entry->index, |
| 177 | &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); |
| 178 | entry->flags = 0; |
| 179 | } |
| 180 | |
| 181 | static bool supported_xcr0_bit(unsigned bit) |
| 182 | { |
| 183 | u64 mask = ((u64)1 << bit); |
| 184 | |
| 185 | return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0; |
| 186 | } |
| 187 | |
| 188 | #define F(x) bit(X86_FEATURE_##x) |
| 189 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 190 | static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 191 | u32 index, int *nent, int maxnent) |
| 192 | { |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 193 | int r; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 194 | unsigned f_nx = is_efer_nx() ? F(NX) : 0; |
| 195 | #ifdef CONFIG_X86_64 |
| 196 | unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) |
| 197 | ? F(GBPAGES) : 0; |
| 198 | unsigned f_lm = F(LM); |
| 199 | #else |
| 200 | unsigned f_gbpages = 0; |
| 201 | unsigned f_lm = 0; |
| 202 | #endif |
| 203 | unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; |
Mao, Junjie | ad756a1 | 2012-07-02 01:18:48 +0000 | [diff] [blame] | 204 | unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 205 | |
| 206 | /* cpuid 1.edx */ |
| 207 | const u32 kvm_supported_word0_x86_features = |
| 208 | F(FPU) | F(VME) | F(DE) | F(PSE) | |
| 209 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | |
| 210 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | |
| 211 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | |
| 212 | F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) | |
| 213 | 0 /* Reserved, DS, ACPI */ | F(MMX) | |
| 214 | F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | |
| 215 | 0 /* HTT, TM, Reserved, PBE */; |
| 216 | /* cpuid 0x80000001.edx */ |
| 217 | const u32 kvm_supported_word1_x86_features = |
| 218 | F(FPU) | F(VME) | F(DE) | F(PSE) | |
| 219 | F(TSC) | F(MSR) | F(PAE) | F(MCE) | |
| 220 | F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | |
| 221 | F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | |
| 222 | F(PAT) | F(PSE36) | 0 /* Reserved */ | |
| 223 | f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | |
| 224 | F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | |
| 225 | 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); |
| 226 | /* cpuid 1.ecx */ |
| 227 | const u32 kvm_supported_word4_x86_features = |
| 228 | F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | |
| 229 | 0 /* DS-CPL, VMX, SMX, EST */ | |
| 230 | 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | |
Liu, Jinsong | fb21536 | 2011-11-28 03:55:19 -0800 | [diff] [blame] | 231 | F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | |
Mao, Junjie | ad756a1 | 2012-07-02 01:18:48 +0000 | [diff] [blame] | 232 | F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 233 | F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | |
| 234 | 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | |
| 235 | F(F16C) | F(RDRAND); |
| 236 | /* cpuid 0x80000001.ecx */ |
| 237 | const u32 kvm_supported_word6_x86_features = |
| 238 | F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | |
| 239 | F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | |
Boris Ostrovsky | 2b036c6 | 2012-01-09 14:00:35 -0500 | [diff] [blame] | 240 | F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 241 | 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); |
| 242 | |
| 243 | /* cpuid 0xC0000001.edx */ |
| 244 | const u32 kvm_supported_word5_x86_features = |
| 245 | F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | |
| 246 | F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | |
| 247 | F(PMM) | F(PMM_EN); |
| 248 | |
| 249 | /* cpuid 7.0.ebx */ |
| 250 | const u32 kvm_supported_word9_x86_features = |
Liu, Jinsong | 83c5291 | 2012-02-28 05:15:46 +0000 | [diff] [blame] | 251 | F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | |
Mao, Junjie | ad756a1 | 2012-07-02 01:18:48 +0000 | [diff] [blame] | 252 | F(BMI2) | F(ERMS) | f_invpcid | F(RTM); |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 253 | |
| 254 | /* all calls to cpuid_count() should be made on the same cpu */ |
| 255 | get_cpu(); |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 256 | |
| 257 | r = -E2BIG; |
| 258 | |
| 259 | if (*nent >= maxnent) |
| 260 | goto out; |
| 261 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 262 | do_cpuid_1_ent(entry, function, index); |
| 263 | ++*nent; |
| 264 | |
| 265 | switch (function) { |
| 266 | case 0: |
| 267 | entry->eax = min(entry->eax, (u32)0xd); |
| 268 | break; |
| 269 | case 1: |
| 270 | entry->edx &= kvm_supported_word0_x86_features; |
| 271 | cpuid_mask(&entry->edx, 0); |
| 272 | entry->ecx &= kvm_supported_word4_x86_features; |
| 273 | cpuid_mask(&entry->ecx, 4); |
| 274 | /* we support x2apic emulation even if host does not support |
| 275 | * it since we emulate x2apic in software */ |
| 276 | entry->ecx |= F(X2APIC); |
| 277 | break; |
| 278 | /* function 2 entries are STATEFUL. That is, repeated cpuid commands |
| 279 | * may return different values. This forces us to get_cpu() before |
| 280 | * issuing the first command, and also to emulate this annoying behavior |
| 281 | * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ |
| 282 | case 2: { |
| 283 | int t, times = entry->eax & 0xff; |
| 284 | |
| 285 | entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; |
| 286 | entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 287 | for (t = 1; t < times; ++t) { |
| 288 | if (*nent >= maxnent) |
| 289 | goto out; |
| 290 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 291 | do_cpuid_1_ent(&entry[t], function, 0); |
| 292 | entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; |
| 293 | ++*nent; |
| 294 | } |
| 295 | break; |
| 296 | } |
| 297 | /* function 4 has additional index. */ |
| 298 | case 4: { |
| 299 | int i, cache_type; |
| 300 | |
| 301 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
| 302 | /* read more entries until cache_type is zero */ |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 303 | for (i = 1; ; ++i) { |
| 304 | if (*nent >= maxnent) |
| 305 | goto out; |
| 306 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 307 | cache_type = entry[i - 1].eax & 0x1f; |
| 308 | if (!cache_type) |
| 309 | break; |
| 310 | do_cpuid_1_ent(&entry[i], function, i); |
| 311 | entry[i].flags |= |
| 312 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
| 313 | ++*nent; |
| 314 | } |
| 315 | break; |
| 316 | } |
| 317 | case 7: { |
| 318 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
Guo Chao | bbbda79 | 2012-06-28 15:20:58 +0800 | [diff] [blame] | 319 | /* Mask ebx against host capability word 9 */ |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 320 | if (index == 0) { |
| 321 | entry->ebx &= kvm_supported_word9_x86_features; |
| 322 | cpuid_mask(&entry->ebx, 9); |
| 323 | } else |
| 324 | entry->ebx = 0; |
| 325 | entry->eax = 0; |
| 326 | entry->ecx = 0; |
| 327 | entry->edx = 0; |
| 328 | break; |
| 329 | } |
| 330 | case 9: |
| 331 | break; |
Gleb Natapov | a6c06ed | 2011-11-10 14:57:28 +0200 | [diff] [blame] | 332 | case 0xa: { /* Architectural Performance Monitoring */ |
| 333 | struct x86_pmu_capability cap; |
| 334 | union cpuid10_eax eax; |
| 335 | union cpuid10_edx edx; |
| 336 | |
| 337 | perf_get_x86_pmu_capability(&cap); |
| 338 | |
| 339 | /* |
| 340 | * Only support guest architectural pmu on a host |
| 341 | * with architectural pmu. |
| 342 | */ |
| 343 | if (!cap.version) |
| 344 | memset(&cap, 0, sizeof(cap)); |
| 345 | |
| 346 | eax.split.version_id = min(cap.version, 2); |
| 347 | eax.split.num_counters = cap.num_counters_gp; |
| 348 | eax.split.bit_width = cap.bit_width_gp; |
| 349 | eax.split.mask_length = cap.events_mask_len; |
| 350 | |
| 351 | edx.split.num_counters_fixed = cap.num_counters_fixed; |
| 352 | edx.split.bit_width_fixed = cap.bit_width_fixed; |
| 353 | edx.split.reserved = 0; |
| 354 | |
| 355 | entry->eax = eax.full; |
| 356 | entry->ebx = cap.events_mask; |
| 357 | entry->ecx = 0; |
| 358 | entry->edx = edx.full; |
| 359 | break; |
| 360 | } |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 361 | /* function 0xb has additional index. */ |
| 362 | case 0xb: { |
| 363 | int i, level_type; |
| 364 | |
| 365 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
| 366 | /* read more entries until level_type is zero */ |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 367 | for (i = 1; ; ++i) { |
| 368 | if (*nent >= maxnent) |
| 369 | goto out; |
| 370 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 371 | level_type = entry[i - 1].ecx & 0xff00; |
| 372 | if (!level_type) |
| 373 | break; |
| 374 | do_cpuid_1_ent(&entry[i], function, i); |
| 375 | entry[i].flags |= |
| 376 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
| 377 | ++*nent; |
| 378 | } |
| 379 | break; |
| 380 | } |
| 381 | case 0xd: { |
| 382 | int idx, i; |
| 383 | |
| 384 | entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 385 | for (idx = 1, i = 1; idx < 64; ++idx) { |
| 386 | if (*nent >= maxnent) |
| 387 | goto out; |
| 388 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 389 | do_cpuid_1_ent(&entry[i], function, idx); |
| 390 | if (entry[i].eax == 0 || !supported_xcr0_bit(idx)) |
| 391 | continue; |
| 392 | entry[i].flags |= |
| 393 | KVM_CPUID_FLAG_SIGNIFCANT_INDEX; |
| 394 | ++*nent; |
| 395 | ++i; |
| 396 | } |
| 397 | break; |
| 398 | } |
| 399 | case KVM_CPUID_SIGNATURE: { |
Mathias Krause | 326d07c | 2012-08-30 01:30:13 +0200 | [diff] [blame] | 400 | static const char signature[12] = "KVMKVMKVM\0\0"; |
| 401 | const u32 *sigptr = (const u32 *)signature; |
Michael S. Tsirkin | 57c22e5 | 2012-05-02 17:55:56 +0300 | [diff] [blame] | 402 | entry->eax = KVM_CPUID_FEATURES; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 403 | entry->ebx = sigptr[0]; |
| 404 | entry->ecx = sigptr[1]; |
| 405 | entry->edx = sigptr[2]; |
| 406 | break; |
| 407 | } |
| 408 | case KVM_CPUID_FEATURES: |
| 409 | entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | |
| 410 | (1 << KVM_FEATURE_NOP_IO_DELAY) | |
| 411 | (1 << KVM_FEATURE_CLOCKSOURCE2) | |
| 412 | (1 << KVM_FEATURE_ASYNC_PF) | |
Michael S. Tsirkin | ae7a2a3 | 2012-06-24 19:25:07 +0300 | [diff] [blame] | 413 | (1 << KVM_FEATURE_PV_EOI) | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 414 | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT); |
| 415 | |
| 416 | if (sched_info_on()) |
| 417 | entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); |
| 418 | |
| 419 | entry->ebx = 0; |
| 420 | entry->ecx = 0; |
| 421 | entry->edx = 0; |
| 422 | break; |
| 423 | case 0x80000000: |
| 424 | entry->eax = min(entry->eax, 0x8000001a); |
| 425 | break; |
| 426 | case 0x80000001: |
| 427 | entry->edx &= kvm_supported_word1_x86_features; |
| 428 | cpuid_mask(&entry->edx, 1); |
| 429 | entry->ecx &= kvm_supported_word6_x86_features; |
| 430 | cpuid_mask(&entry->ecx, 6); |
| 431 | break; |
| 432 | case 0x80000008: { |
| 433 | unsigned g_phys_as = (entry->eax >> 16) & 0xff; |
| 434 | unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); |
| 435 | unsigned phys_as = entry->eax & 0xff; |
| 436 | |
| 437 | if (!g_phys_as) |
| 438 | g_phys_as = phys_as; |
| 439 | entry->eax = g_phys_as | (virt_as << 8); |
| 440 | entry->ebx = entry->edx = 0; |
| 441 | break; |
| 442 | } |
| 443 | case 0x80000019: |
| 444 | entry->ecx = entry->edx = 0; |
| 445 | break; |
| 446 | case 0x8000001a: |
| 447 | break; |
| 448 | case 0x8000001d: |
| 449 | break; |
| 450 | /*Add support for Centaur's CPUID instruction*/ |
| 451 | case 0xC0000000: |
| 452 | /*Just support up to 0xC0000004 now*/ |
| 453 | entry->eax = min(entry->eax, 0xC0000004); |
| 454 | break; |
| 455 | case 0xC0000001: |
| 456 | entry->edx &= kvm_supported_word5_x86_features; |
| 457 | cpuid_mask(&entry->edx, 5); |
| 458 | break; |
| 459 | case 3: /* Processor serial number */ |
| 460 | case 5: /* MONITOR/MWAIT */ |
| 461 | case 6: /* Thermal management */ |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 462 | case 0x80000007: /* Advanced power management */ |
| 463 | case 0xC0000002: |
| 464 | case 0xC0000003: |
| 465 | case 0xC0000004: |
| 466 | default: |
| 467 | entry->eax = entry->ebx = entry->ecx = entry->edx = 0; |
| 468 | break; |
| 469 | } |
| 470 | |
| 471 | kvm_x86_ops->set_supported_cpuid(function, entry); |
| 472 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 473 | r = 0; |
| 474 | |
| 475 | out: |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 476 | put_cpu(); |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 477 | |
| 478 | return r; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 479 | } |
| 480 | |
| 481 | #undef F |
| 482 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 483 | struct kvm_cpuid_param { |
| 484 | u32 func; |
| 485 | u32 idx; |
| 486 | bool has_leaf_count; |
Mathias Krause | 326d07c | 2012-08-30 01:30:13 +0200 | [diff] [blame] | 487 | bool (*qualifier)(const struct kvm_cpuid_param *param); |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 488 | }; |
| 489 | |
Mathias Krause | 326d07c | 2012-08-30 01:30:13 +0200 | [diff] [blame] | 490 | static bool is_centaur_cpu(const struct kvm_cpuid_param *param) |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 491 | { |
| 492 | return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; |
| 493 | } |
| 494 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 495 | int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid, |
| 496 | struct kvm_cpuid_entry2 __user *entries) |
| 497 | { |
| 498 | struct kvm_cpuid_entry2 *cpuid_entries; |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 499 | int limit, nent = 0, r = -E2BIG, i; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 500 | u32 func; |
Mathias Krause | 326d07c | 2012-08-30 01:30:13 +0200 | [diff] [blame] | 501 | static const struct kvm_cpuid_param param[] = { |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 502 | { .func = 0, .has_leaf_count = true }, |
| 503 | { .func = 0x80000000, .has_leaf_count = true }, |
| 504 | { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, |
| 505 | { .func = KVM_CPUID_SIGNATURE }, |
| 506 | { .func = KVM_CPUID_FEATURES }, |
| 507 | }; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 508 | |
| 509 | if (cpuid->nent < 1) |
| 510 | goto out; |
| 511 | if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) |
| 512 | cpuid->nent = KVM_MAX_CPUID_ENTRIES; |
| 513 | r = -ENOMEM; |
| 514 | cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); |
| 515 | if (!cpuid_entries) |
| 516 | goto out; |
| 517 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 518 | r = 0; |
| 519 | for (i = 0; i < ARRAY_SIZE(param); i++) { |
Mathias Krause | 326d07c | 2012-08-30 01:30:13 +0200 | [diff] [blame] | 520 | const struct kvm_cpuid_param *ent = ¶m[i]; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 521 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 522 | if (ent->qualifier && !ent->qualifier(ent)) |
| 523 | continue; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 524 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 525 | r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 526 | &nent, cpuid->nent); |
| 527 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 528 | if (r) |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 529 | goto out_free; |
| 530 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 531 | if (!ent->has_leaf_count) |
| 532 | continue; |
| 533 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 534 | limit = cpuid_entries[nent - 1].eax; |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 535 | for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) |
| 536 | r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, |
| 537 | &nent, cpuid->nent); |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 538 | |
Sasha Levin | 831bf66 | 2011-11-28 11:20:29 +0200 | [diff] [blame] | 539 | if (r) |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 540 | goto out_free; |
| 541 | } |
| 542 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 543 | r = -EFAULT; |
| 544 | if (copy_to_user(entries, cpuid_entries, |
| 545 | nent * sizeof(struct kvm_cpuid_entry2))) |
| 546 | goto out_free; |
| 547 | cpuid->nent = nent; |
| 548 | r = 0; |
| 549 | |
| 550 | out_free: |
| 551 | vfree(cpuid_entries); |
| 552 | out: |
| 553 | return r; |
| 554 | } |
| 555 | |
| 556 | static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) |
| 557 | { |
| 558 | struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; |
| 559 | int j, nent = vcpu->arch.cpuid_nent; |
| 560 | |
| 561 | e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; |
| 562 | /* when no next entry is found, the current entry[i] is reselected */ |
| 563 | for (j = i + 1; ; j = (j + 1) % nent) { |
| 564 | struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; |
| 565 | if (ej->function == e->function) { |
| 566 | ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; |
| 567 | return j; |
| 568 | } |
| 569 | } |
| 570 | return 0; /* silence gcc, even though control never reaches here */ |
| 571 | } |
| 572 | |
| 573 | /* find an entry with matching function, matching index (if needed), and that |
| 574 | * should be read next (if it's stateful) */ |
| 575 | static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, |
| 576 | u32 function, u32 index) |
| 577 | { |
| 578 | if (e->function != function) |
| 579 | return 0; |
| 580 | if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) |
| 581 | return 0; |
| 582 | if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && |
| 583 | !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) |
| 584 | return 0; |
| 585 | return 1; |
| 586 | } |
| 587 | |
| 588 | struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, |
| 589 | u32 function, u32 index) |
| 590 | { |
| 591 | int i; |
| 592 | struct kvm_cpuid_entry2 *best = NULL; |
| 593 | |
| 594 | for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { |
| 595 | struct kvm_cpuid_entry2 *e; |
| 596 | |
| 597 | e = &vcpu->arch.cpuid_entries[i]; |
| 598 | if (is_matching_cpuid_entry(e, function, index)) { |
| 599 | if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) |
| 600 | move_to_next_stateful_cpuid_entry(vcpu, i); |
| 601 | best = e; |
| 602 | break; |
| 603 | } |
| 604 | } |
| 605 | return best; |
| 606 | } |
| 607 | EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); |
| 608 | |
| 609 | int cpuid_maxphyaddr(struct kvm_vcpu *vcpu) |
| 610 | { |
| 611 | struct kvm_cpuid_entry2 *best; |
| 612 | |
| 613 | best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); |
| 614 | if (!best || best->eax < 0x80000008) |
| 615 | goto not_found; |
| 616 | best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); |
| 617 | if (best) |
| 618 | return best->eax & 0xff; |
| 619 | not_found: |
| 620 | return 36; |
| 621 | } |
| 622 | |
| 623 | /* |
| 624 | * If no match is found, check whether we exceed the vCPU's limit |
| 625 | * and return the content of the highest valid _standard_ leaf instead. |
| 626 | * This is to satisfy the CPUID specification. |
| 627 | */ |
| 628 | static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, |
| 629 | u32 function, u32 index) |
| 630 | { |
| 631 | struct kvm_cpuid_entry2 *maxlevel; |
| 632 | |
| 633 | maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); |
| 634 | if (!maxlevel || maxlevel->eax >= function) |
| 635 | return NULL; |
| 636 | if (function & 0x80000000) { |
| 637 | maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); |
| 638 | if (!maxlevel) |
| 639 | return NULL; |
| 640 | } |
| 641 | return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); |
| 642 | } |
| 643 | |
Avi Kivity | 62046e5 | 2012-06-07 14:07:48 +0300 | [diff] [blame] | 644 | void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 645 | { |
Avi Kivity | 62046e5 | 2012-06-07 14:07:48 +0300 | [diff] [blame] | 646 | u32 function = *eax, index = *ecx; |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 647 | struct kvm_cpuid_entry2 *best; |
| 648 | |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 649 | best = kvm_find_cpuid_entry(vcpu, function, index); |
| 650 | |
| 651 | if (!best) |
| 652 | best = check_cpuid_limit(vcpu, function, index); |
| 653 | |
| 654 | if (best) { |
Avi Kivity | 62046e5 | 2012-06-07 14:07:48 +0300 | [diff] [blame] | 655 | *eax = best->eax; |
| 656 | *ebx = best->ebx; |
| 657 | *ecx = best->ecx; |
| 658 | *edx = best->edx; |
| 659 | } else |
| 660 | *eax = *ebx = *ecx = *edx = 0; |
| 661 | } |
| 662 | |
| 663 | void kvm_emulate_cpuid(struct kvm_vcpu *vcpu) |
| 664 | { |
| 665 | u32 function, eax, ebx, ecx, edx; |
| 666 | |
| 667 | function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
| 668 | ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); |
| 669 | kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx); |
| 670 | kvm_register_write(vcpu, VCPU_REGS_RAX, eax); |
| 671 | kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); |
| 672 | kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); |
| 673 | kvm_register_write(vcpu, VCPU_REGS_RDX, edx); |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 674 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
Avi Kivity | 62046e5 | 2012-06-07 14:07:48 +0300 | [diff] [blame] | 675 | trace_kvm_cpuid(function, eax, ebx, ecx, edx); |
Avi Kivity | 00b27a3 | 2011-11-23 16:30:32 +0200 | [diff] [blame] | 676 | } |
| 677 | EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); |