qemu-kvm-x86.c

   1 /*
   2  * qemu/kvm integration, x86 specific code
   3  *
   4  * Copyright (C) 2006-2008 Qumranet Technologies
   5  *
   6  * Licensed under the terms of the GNU GPL version 2 or higher.
   7  */
   8
   9 #include "config.h"
  10 #include "config-host.h"
  11
  12 #include <string.h>
  13 #include "hw/hw.h"
  14 #include "gdbstub.h"
  15 #include <sys/io.h>
  16
  17 #include "qemu-kvm.h"
  18 #include "libkvm-all.h"
  19 #include <pthread.h>
  20 #include <sys/utsname.h>
  21 #include <linux/kvm_para.h>
  22
  23 #include "kvm.h"
  24
  25 #define MSR_IA32_TSC            0x10
  26
  27 static struct kvm_msr_list *kvm_msr_list;
  28 extern unsigned int kvm_shadow_memory;
  29 static int kvm_has_msr_star;
  30 static int kvm_has_vm_hsave_pa;
  31
  32 static int lm_capable_kernel;
  33
  34 int kvm_qemu_create_memory_alias(uint64_t phys_start,
  35                                  uint64_t len,
  36                                  uint64_t target_phys)
  37 {
  38     return kvm_create_memory_alias(kvm_context, phys_start, len, target_phys);
  39 }
  40
  41 int kvm_qemu_destroy_memory_alias(uint64_t phys_start)
  42 {
  43         return kvm_destroy_memory_alias(kvm_context, phys_start);
  44 }
  45
  46 int kvm_arch_qemu_create_context(void)
  47 {
  48     int i;
  49     struct utsname utsname;
  50
  51     uname(&utsname);
  52     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
  53
  54     if (kvm_shadow_memory)
  55         kvm_set_shadow_pages(kvm_context, kvm_shadow_memory);
  56
  57     kvm_msr_list = kvm_get_msr_list(kvm_context);
  58     if (!kvm_msr_list)
  59                 return -1;
  60     for (i = 0; i < kvm_msr_list->nmsrs; ++i) {
  61         if (kvm_msr_list->indices[i] == MSR_STAR)
  62             kvm_has_msr_star = 1;
  63         if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA)
  64             kvm_has_vm_hsave_pa = 1;
  65     }
  66
  67     return 0;
  68 }
  69
  70 static void set_msr_entry(struct kvm_msr_entry *entry, uint32_t index,
  71                           uint64_t data)
  72 {
  73     entry->index = index;
  74     entry->data  = data;
  75 }
  76
  77 /* returns 0 on success, non-0 on failure */
  78 static int get_msr_entry(struct kvm_msr_entry *entry, CPUState *env)
  79 {
  80         switch (entry->index) {
  81         case MSR_IA32_SYSENTER_CS:
  82             env->sysenter_cs  = entry->data;
  83             break;
  84         case MSR_IA32_SYSENTER_ESP:
  85             env->sysenter_esp = entry->data;
  86             break;
  87         case MSR_IA32_SYSENTER_EIP:
  88             env->sysenter_eip = entry->data;
  89             break;
  90         case MSR_STAR:
  91             env->star         = entry->data;
  92             break;
  93 #ifdef TARGET_X86_64
  94         case MSR_CSTAR:
  95             env->cstar        = entry->data;
  96             break;
  97         case MSR_KERNELGSBASE:
  98             env->kernelgsbase = entry->data;
  99             break;
 100         case MSR_FMASK:
 101             env->fmask        = entry->data;
 102             break;
 103         case MSR_LSTAR:
 104             env->lstar        = entry->data;
 105             break;
 106 #endif
 107         case MSR_IA32_TSC:
 108             env->tsc          = entry->data;
 109             break;
 110         case MSR_VM_HSAVE_PA:
 111             env->vm_hsave     = entry->data;
 112             break;
 113         default:
 114             printf("Warning unknown msr index 0x%x\n", entry->index);
 115             return 1;
 116         }
 117         return 0;
 118 }
 119
 120 #ifdef TARGET_X86_64
 121 #define MSR_COUNT 9
 122 #else
 123 #define MSR_COUNT 5
 124 #endif
 125
 126 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 127 {
 128     lhs->selector = rhs->selector;
 129     lhs->base = rhs->base;
 130     lhs->limit = rhs->limit;
 131     lhs->type = 3;
 132     lhs->present = 1;
 133     lhs->dpl = 3;
 134     lhs->db = 0;
 135     lhs->s = 1;
 136     lhs->l = 0;
 137     lhs->g = 0;
 138     lhs->avl = 0;
 139     lhs->unusable = 0;
 140 }
 141
 142 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 143 {
 144     unsigned flags = rhs->flags;
 145     lhs->selector = rhs->selector;
 146     lhs->base = rhs->base;
 147     lhs->limit = rhs->limit;
 148     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 149     lhs->present = (flags & DESC_P_MASK) != 0;
 150     lhs->dpl = rhs->selector & 3;
 151     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 152     lhs->s = (flags & DESC_S_MASK) != 0;
 153     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 154     lhs->g = (flags & DESC_G_MASK) != 0;
 155     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 156     lhs->unusable = 0;
 157 }
 158
 159 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 160 {
 161     lhs->selector = rhs->selector;
 162     lhs->base = rhs->base;
 163     lhs->limit = rhs->limit;
 164     lhs->flags =
 165         (rhs->type << DESC_TYPE_SHIFT)
 166         | (rhs->present * DESC_P_MASK)
 167         | (rhs->dpl << DESC_DPL_SHIFT)
 168         | (rhs->db << DESC_B_SHIFT)
 169         | (rhs->s * DESC_S_MASK)
 170         | (rhs->l << DESC_L_SHIFT)
 171         | (rhs->g * DESC_G_MASK)
 172         | (rhs->avl * DESC_AVL_MASK);
 173 }
 174
 175 void kvm_arch_load_regs(CPUState *env)
 176 {
 177     struct kvm_regs regs;
 178     struct kvm_fpu fpu;
 179     struct kvm_sregs sregs;
 180     struct kvm_msr_entry msrs[MSR_COUNT];
 181     int rc, n, i;
 182
 183     regs.rax = env->regs[R_EAX];
 184     regs.rbx = env->regs[R_EBX];
 185     regs.rcx = env->regs[R_ECX];
 186     regs.rdx = env->regs[R_EDX];
 187     regs.rsi = env->regs[R_ESI];
 188     regs.rdi = env->regs[R_EDI];
 189     regs.rsp = env->regs[R_ESP];
 190     regs.rbp = env->regs[R_EBP];
 191 #ifdef TARGET_X86_64
 192     regs.r8 = env->regs[8];
 193     regs.r9 = env->regs[9];
 194     regs.r10 = env->regs[10];
 195     regs.r11 = env->regs[11];
 196     regs.r12 = env->regs[12];
 197     regs.r13 = env->regs[13];
 198     regs.r14 = env->regs[14];
 199     regs.r15 = env->regs[15];
 200 #endif
 201
 202     regs.rflags = env->eflags;
 203     regs.rip = env->eip;
 204
 205     kvm_set_regs(env->kvm_cpu_state.vcpu_ctx, &regs);
 206
 207     memset(&fpu, 0, sizeof fpu);
 208     fpu.fsw = env->fpus & ~(7 << 11);
 209     fpu.fsw |= (env->fpstt & 7) << 11;
 210     fpu.fcw = env->fpuc;
 211     for (i = 0; i < 8; ++i)
 212         fpu.ftwx |= (!env->fptags[i]) << i;
 213     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 214     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 215     fpu.mxcsr = env->mxcsr;
 216     kvm_set_fpu(env->kvm_cpu_state.vcpu_ctx, &fpu);
 217
 218     memcpy(sregs.interrupt_bitmap, env->interrupt_bitmap, sizeof(sregs.interrupt_bitmap));
 219
 220     if ((env->eflags & VM_MASK)) {
 221             set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 222             set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 223             set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 224             set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 225             set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 226             set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 227     } else {
 228             set_seg(&sregs.cs, &env->segs[R_CS]);
 229             set_seg(&sregs.ds, &env->segs[R_DS]);
 230             set_seg(&sregs.es, &env->segs[R_ES]);
 231             set_seg(&sregs.fs, &env->segs[R_FS]);
 232             set_seg(&sregs.gs, &env->segs[R_GS]);
 233             set_seg(&sregs.ss, &env->segs[R_SS]);
 234
 235             if (env->cr[0] & CR0_PE_MASK) {
 236                 /* force ss cpl to cs cpl */
 237                 sregs.ss.selector = (sregs.ss.selector & ~3) |
 238                         (sregs.cs.selector & 3);
 239                 sregs.ss.dpl = sregs.ss.selector & 3;
 240             }
 241     }
 242
 243     set_seg(&sregs.tr, &env->tr);
 244     set_seg(&sregs.ldt, &env->ldt);
 245
 246     sregs.idt.limit = env->idt.limit;
 247     sregs.idt.base = env->idt.base;
 248     sregs.gdt.limit = env->gdt.limit;
 249     sregs.gdt.base = env->gdt.base;
 250
 251     sregs.cr0 = env->cr[0];
 252     sregs.cr2 = env->cr[2];
 253     sregs.cr3 = env->cr[3];
 254     sregs.cr4 = env->cr[4];
 255
 256     sregs.cr8 = cpu_get_apic_tpr(env);
 257     sregs.apic_base = cpu_get_apic_base(env);
 258
 259     sregs.efer = env->efer;
 260
 261     kvm_set_sregs(env->kvm_cpu_state.vcpu_ctx, &sregs);
 262
 263     /* msrs */
 264     n = 0;
 265     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_CS,  env->sysenter_cs);
 266     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 267     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 268     if (kvm_has_msr_star)
 269         set_msr_entry(&msrs[n++], MSR_STAR,              env->star);
 270     if (kvm_has_vm_hsave_pa)
 271         set_msr_entry(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
 272 #ifdef TARGET_X86_64
 273     if (lm_capable_kernel) {
 274         set_msr_entry(&msrs[n++], MSR_CSTAR,             env->cstar);
 275         set_msr_entry(&msrs[n++], MSR_KERNELGSBASE,      env->kernelgsbase);
 276         set_msr_entry(&msrs[n++], MSR_FMASK,             env->fmask);
 277         set_msr_entry(&msrs[n++], MSR_LSTAR  ,           env->lstar);
 278     }
 279 #endif
 280
 281     rc = kvm_set_msrs(env->kvm_cpu_state.vcpu_ctx, msrs, n);
 282     if (rc == -1)
 283         perror("kvm_set_msrs FAILED");
 284 }
 285
 286 void kvm_load_tsc(CPUState *env)
 287 {
 288     int rc;
 289     struct kvm_msr_entry msr;
 290
 291     set_msr_entry(&msr, MSR_IA32_TSC, env->tsc);
 292
 293     rc = kvm_set_msrs(env->kvm_cpu_state.vcpu_ctx, &msr, 1);
 294     if (rc == -1)
 295         perror("kvm_set_tsc FAILED.\n");
 296 }
 297
 298 void kvm_arch_save_mpstate(CPUState *env)
 299 {
 300 #ifdef KVM_CAP_MP_STATE
 301     int r;
 302     struct kvm_mp_state mp_state;
 303
 304     r = kvm_get_mpstate(env->kvm_cpu_state.vcpu_ctx, &mp_state);
 305     if (r < 0)
 306         env->mp_state = -1;
 307     else
 308         env->mp_state = mp_state.mp_state;
 309 #endif
 310 }
 311
 312 void kvm_arch_load_mpstate(CPUState *env)
 313 {
 314 #ifdef KVM_CAP_MP_STATE
 315     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
 316
 317     /*
 318      * -1 indicates that the host did not support GET_MP_STATE ioctl,
 319      *  so don't touch it.
 320      */
 321     if (env->mp_state != -1)
 322         kvm_set_mpstate(env->kvm_cpu_state.vcpu_ctx, &mp_state);
 323 #endif
 324 }
 325
 326 void kvm_arch_save_regs(CPUState *env)
 327 {
 328     struct kvm_regs regs;
 329     struct kvm_fpu fpu;
 330     struct kvm_sregs sregs;
 331     struct kvm_msr_entry msrs[MSR_COUNT];
 332     uint32_t hflags;
 333     uint32_t i, n, rc;
 334
 335     kvm_get_regs(env->kvm_cpu_state.vcpu_ctx, &regs);
 336
 337     env->regs[R_EAX] = regs.rax;
 338     env->regs[R_EBX] = regs.rbx;
 339     env->regs[R_ECX] = regs.rcx;
 340     env->regs[R_EDX] = regs.rdx;
 341     env->regs[R_ESI] = regs.rsi;
 342     env->regs[R_EDI] = regs.rdi;
 343     env->regs[R_ESP] = regs.rsp;
 344     env->regs[R_EBP] = regs.rbp;
 345 #ifdef TARGET_X86_64
 346     env->regs[8] = regs.r8;
 347     env->regs[9] = regs.r9;
 348     env->regs[10] = regs.r10;
 349     env->regs[11] = regs.r11;
 350     env->regs[12] = regs.r12;
 351     env->regs[13] = regs.r13;
 352     env->regs[14] = regs.r14;
 353     env->regs[15] = regs.r15;
 354 #endif
 355
 356     env->eflags = regs.rflags;
 357     env->eip = regs.rip;
 358
 359     kvm_get_fpu(env->kvm_cpu_state.vcpu_ctx, &fpu);
 360     env->fpstt = (fpu.fsw >> 11) & 7;
 361     env->fpus = fpu.fsw;
 362     env->fpuc = fpu.fcw;
 363     for (i = 0; i < 8; ++i)
 364         env->fptags[i] = !((fpu.ftwx >> i) & 1);
 365     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
 366     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
 367     env->mxcsr = fpu.mxcsr;
 368
 369     kvm_get_sregs(env->kvm_cpu_state.vcpu_ctx, &sregs);
 370
 371     memcpy(env->interrupt_bitmap, sregs.interrupt_bitmap, sizeof(env->interrupt_bitmap));
 372
 373     get_seg(&env->segs[R_CS], &sregs.cs);
 374     get_seg(&env->segs[R_DS], &sregs.ds);
 375     get_seg(&env->segs[R_ES], &sregs.es);
 376     get_seg(&env->segs[R_FS], &sregs.fs);
 377     get_seg(&env->segs[R_GS], &sregs.gs);
 378     get_seg(&env->segs[R_SS], &sregs.ss);
 379
 380     get_seg(&env->tr, &sregs.tr);
 381     get_seg(&env->ldt, &sregs.ldt);
 382
 383     env->idt.limit = sregs.idt.limit;
 384     env->idt.base = sregs.idt.base;
 385     env->gdt.limit = sregs.gdt.limit;
 386     env->gdt.base = sregs.gdt.base;
 387
 388     env->cr[0] = sregs.cr0;
 389     env->cr[2] = sregs.cr2;
 390     env->cr[3] = sregs.cr3;
 391     env->cr[4] = sregs.cr4;
 392
 393     cpu_set_apic_base(env, sregs.apic_base);
 394
 395     env->efer = sregs.efer;
 396     //cpu_set_apic_tpr(env, sregs.cr8);
 397
 398 #define HFLAG_COPY_MASK ~( \
 399                         HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
 400                         HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
 401                         HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
 402                         HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
 403
 404
 405
 406     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 407     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
 408     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
 409             (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
 410     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
 411     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
 412             (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
 413
 414     if (env->efer & MSR_EFER_LMA) {
 415         hflags |= HF_LMA_MASK;
 416     }
 417
 418     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
 419         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
 420     } else {
 421         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
 422                 (DESC_B_SHIFT - HF_CS32_SHIFT);
 423         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
 424                 (DESC_B_SHIFT - HF_SS32_SHIFT);
 425         if (!(env->cr[0] & CR0_PE_MASK) ||
 426                    (env->eflags & VM_MASK) ||
 427                    !(hflags & HF_CS32_MASK)) {
 428                 hflags |= HF_ADDSEG_MASK;
 429             } else {
 430                 hflags |= ((env->segs[R_DS].base |
 431                                 env->segs[R_ES].base |
 432                                 env->segs[R_SS].base) != 0) <<
 433                     HF_ADDSEG_SHIFT;
 434             }
 435     }
 436     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
 437
 438     /* msrs */
 439     n = 0;
 440     msrs[n++].index = MSR_IA32_SYSENTER_CS;
 441     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
 442     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
 443     if (kvm_has_msr_star)
 444         msrs[n++].index = MSR_STAR;
 445     msrs[n++].index = MSR_IA32_TSC;
 446     if (kvm_has_vm_hsave_pa)
 447         msrs[n++].index = MSR_VM_HSAVE_PA;
 448 #ifdef TARGET_X86_64
 449     if (lm_capable_kernel) {
 450         msrs[n++].index = MSR_CSTAR;
 451         msrs[n++].index = MSR_KERNELGSBASE;
 452         msrs[n++].index = MSR_FMASK;
 453         msrs[n++].index = MSR_LSTAR;
 454     }
 455 #endif
 456     rc = kvm_get_msrs(env->kvm_cpu_state.vcpu_ctx, msrs, n);
 457     if (rc == -1) {
 458         perror("kvm_get_msrs FAILED");
 459     }
 460     else {
 461         n = rc; /* actual number of MSRs */
 462         for (i=0 ; i<n; i++) {
 463             if (get_msr_entry(&msrs[i], env))
 464                 return;
 465         }
 466     }
 467 }
 468
 469 static void do_cpuid_ent(struct kvm_cpuid_entry2 *e, uint32_t function,
 470                          uint32_t count, CPUState *env)
 471 {
 472     env->regs[R_EAX] = function;
 473     env->regs[R_ECX] = count;
 474     qemu_kvm_cpuid_on_env(env);
 475     e->function = function;
 476     e->flags = 0;
 477     e->index = 0;
 478     e->eax = env->regs[R_EAX];
 479     e->ebx = env->regs[R_EBX];
 480     e->ecx = env->regs[R_ECX];
 481     e->edx = env->regs[R_EDX];
 482 }
 483
 484 struct kvm_para_features {
 485         int cap;
 486         int feature;
 487 } para_features[] = {
 488 #ifdef KVM_CAP_CLOCKSOURCE
 489         { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 490 #endif
 491 #ifdef KVM_CAP_NOP_IO_DELAY
 492         { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 493 #endif
 494 #ifdef KVM_CAP_PV_MMU
 495         { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 496 #endif
 497 #ifdef KVM_CAP_CR3_CACHE
 498         { KVM_CAP_CR3_CACHE, KVM_FEATURE_CR3_CACHE },
 499 #endif
 500         { -1, -1 }
 501 };
 502
 503 static int get_para_features(kvm_context_t kvm_context)
 504 {
 505         int i, features = 0;
 506
 507         for (i = 0; i < ARRAY_SIZE(para_features)-1; i++) {
 508                 if (kvm_check_extension(kvm_context, para_features[i].cap))
 509                         features |= (1 << para_features[i].feature);
 510         }
 511
 512         return features;
 513 }
 514
 515 static void kvm_trim_features(uint32_t *features, uint32_t supported)
 516 {
 517     int i;
 518     uint32_t mask;
 519
 520     for (i = 0; i < 32; ++i) {
 521         mask = 1U << i;
 522         if ((*features & mask) && !(supported & mask)) {
 523             *features &= ~mask;
 524         }
 525     }
 526 }
 527
 528 int kvm_arch_qemu_init_env(CPUState *cenv)
 529 {
 530     struct kvm_cpuid_entry2 cpuid_ent[100];
 531 #ifdef KVM_CPUID_SIGNATURE
 532     struct kvm_cpuid_entry2 *pv_ent;
 533     uint32_t signature[3];
 534 #endif
 535     int cpuid_nent = 0;
 536     CPUState copy;
 537     uint32_t i, j, limit;
 538
 539     qemu_kvm_load_lapic(cenv);
 540
 541     copy = *cenv;
 542
 543 #ifdef KVM_CPUID_SIGNATURE
 544     /* Paravirtualization CPUIDs */
 545     memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 546     pv_ent = &cpuid_ent[cpuid_nent++];
 547     memset(pv_ent, 0, sizeof(*pv_ent));
 548     pv_ent->function = KVM_CPUID_SIGNATURE;
 549     pv_ent->eax = 0;
 550     pv_ent->ebx = signature[0];
 551     pv_ent->ecx = signature[1];
 552     pv_ent->edx = signature[2];
 553
 554     pv_ent = &cpuid_ent[cpuid_nent++];
 555     memset(pv_ent, 0, sizeof(*pv_ent));
 556     pv_ent->function = KVM_CPUID_FEATURES;
 557     pv_ent->eax = get_para_features(kvm_context);
 558 #endif
 559
 560     copy.regs[R_EAX] = 0;
 561     qemu_kvm_cpuid_on_env(&copy);
 562     limit = copy.regs[R_EAX];
 563
 564     for (i = 0; i <= limit; ++i) {
 565         if (i == 4 || i == 0xb || i == 0xd) {
 566             for (j = 0; ; ++j) {
 567                 do_cpuid_ent(&cpuid_ent[cpuid_nent], i, j, &copy);
 568
 569                 cpuid_ent[cpuid_nent].flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 570                 cpuid_ent[cpuid_nent].index = j;
 571
 572                 cpuid_nent++;
 573
 574                 if (i == 4 && copy.regs[R_EAX] == 0)
 575                     break;
 576                 if (i == 0xb && !(copy.regs[R_ECX] & 0xff00))
 577                     break;
 578                 if (i == 0xd && copy.regs[R_EAX] == 0)
 579                     break;
 580             }
 581         } else
 582             do_cpuid_ent(&cpuid_ent[cpuid_nent++], i, 0, &copy);
 583     }
 584
 585     copy.regs[R_EAX] = 0x80000000;
 586     qemu_kvm_cpuid_on_env(&copy);
 587     limit = copy.regs[R_EAX];
 588
 589     for (i = 0x80000000; i <= limit; ++i)
 590         do_cpuid_ent(&cpuid_ent[cpuid_nent++], i, 0, &copy);
 591
 592     kvm_setup_cpuid2(cenv->kvm_cpu_state.vcpu_ctx, cpuid_nent, cpuid_ent);
 593
 594     kvm_trim_features(&cenv->cpuid_features,
 595                       kvm_arch_get_supported_cpuid(cenv, 1, R_EDX));
 596     kvm_trim_features(&cenv->cpuid_ext_features,
 597                       kvm_arch_get_supported_cpuid(cenv, 1, R_ECX));
 598     kvm_trim_features(&cenv->cpuid_ext2_features,
 599                       kvm_arch_get_supported_cpuid(cenv, 0x80000001, R_EDX));
 600     kvm_trim_features(&cenv->cpuid_ext3_features,
 601                       kvm_arch_get_supported_cpuid(cenv, 0x80000001, R_ECX));
 602
 603     return 0;
 604 }
 605
 606 int kvm_arch_halt(void *opaque, kvm_vcpu_context_t vcpu)
 607 {
 608     CPUState *env = cpu_single_env;
 609
 610     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
 611           (env->eflags & IF_MASK)) &&
 612         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
 613             env->halted = 1;
 614     }
 615     return 1;
 616 }
 617
 618 void kvm_arch_pre_kvm_run(void *opaque, CPUState *env)
 619 {
 620     if (!kvm_irqchip_in_kernel(kvm_context))
 621         kvm_set_cr8(env->kvm_cpu_state.vcpu_ctx, cpu_get_apic_tpr(env));
 622 }
 623
 624 void kvm_arch_post_kvm_run(void *opaque, CPUState *env)
 625 {
 626     cpu_single_env = env;
 627
 628     env->eflags = kvm_get_interrupt_flag(env->kvm_cpu_state.vcpu_ctx)
 629         ? env->eflags | IF_MASK : env->eflags & ~IF_MASK;
 630
 631     cpu_set_apic_tpr(env, kvm_get_cr8(env->kvm_cpu_state.vcpu_ctx));
 632     cpu_set_apic_base(env, kvm_get_apic_base(env->kvm_cpu_state.vcpu_ctx));
 633 }
 634
 635 int kvm_arch_has_work(CPUState *env)
 636 {
 637     if (env->exit_request ||
 638         ((env->interrupt_request & CPU_INTERRUPT_HARD) &&
 639          (env->eflags & IF_MASK)) ||
 640         (env->interrupt_request & CPU_INTERRUPT_NMI))
 641         return 1;
 642     return 0;
 643 }
 644
 645 int kvm_arch_try_push_interrupts(void *opaque)
 646 {
 647     CPUState *env = cpu_single_env;
 648     int r, irq;
 649
 650     if (kvm_is_ready_for_interrupt_injection(env->kvm_cpu_state.vcpu_ctx) &&
 651         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
 652         (env->eflags & IF_MASK)) {
 653             env->interrupt_request &= ~CPU_INTERRUPT_HARD;
 654             irq = cpu_get_pic_interrupt(env);
 655             if (irq >= 0) {
 656                 r = kvm_inject_irq(env->kvm_cpu_state.vcpu_ctx, irq);
 657                 if (r < 0)
 658                     printf("cpu %d fail inject %x\n", env->cpu_index, irq);
 659             }
 660     }
 661
 662     return (env->interrupt_request & CPU_INTERRUPT_HARD) != 0;
 663 }
 664
 665 #ifdef KVM_CAP_USER_NMI
 666 void kvm_arch_push_nmi(void *opaque)
 667 {
 668     CPUState *env = cpu_single_env;
 669     int r;
 670
 671     if (likely(!(env->interrupt_request & CPU_INTERRUPT_NMI)))
 672         return;
 673
 674     env->interrupt_request &= ~CPU_INTERRUPT_NMI;
 675     r = kvm_inject_nmi(env->kvm_cpu_state.vcpu_ctx);
 676     if (r < 0)
 677         printf("cpu %d fail inject NMI\n", env->cpu_index);
 678 }
 679 #endif /* KVM_CAP_USER_NMI */
 680
 681 void kvm_arch_update_regs_for_sipi(CPUState *env)
 682 {
 683     SegmentCache cs = env->segs[R_CS];
 684
 685     kvm_arch_save_regs(env);
 686     env->segs[R_CS] = cs;
 687     env->eip = 0;
 688     kvm_arch_load_regs(env);
 689 }
 690
 691 int handle_tpr_access(void *opaque, kvm_vcpu_context_t vcpu,
 692                              uint64_t rip, int is_write)
 693 {
 694     kvm_tpr_access_report(cpu_single_env, rip, is_write);
 695     return 0;
 696 }
 697
 698 void kvm_arch_cpu_reset(CPUState *env)
 699 {
 700     kvm_arch_load_regs(env);
 701     if (env->cpu_index != 0) {
 702         if (kvm_irqchip_in_kernel(kvm_context)) {
 703 #ifdef KVM_CAP_MP_STATE
 704             kvm_reset_mpstate(env->kvm_cpu_state.vcpu_ctx);
 705 #endif
 706         } else {
 707             env->interrupt_request &= ~CPU_INTERRUPT_HARD;
 708             env->halted = 1;
 709         }
 710     }
 711 }
 712
 713 int kvm_arch_insert_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
 714 {
 715     uint8_t int3 = 0xcc;
 716
 717     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
 718         cpu_memory_rw_debug(env, bp->pc, &int3, 1, 1))
 719         return -EINVAL;
 720     return 0;
 721 }
 722
 723 int kvm_arch_remove_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
 724 {
 725     uint8_t int3;
 726
 727     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
 728         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1))
 729         return -EINVAL;
 730     return 0;
 731 }
 732
 733 #ifdef KVM_CAP_SET_GUEST_DEBUG
 734 static struct {
 735     target_ulong addr;
 736     int len;
 737     int type;
 738 } hw_breakpoint[4];
 739
 740 static int nb_hw_breakpoint;
 741
 742 static int find_hw_breakpoint(target_ulong addr, int len, int type)
 743 {
 744     int n;
 745
 746     for (n = 0; n < nb_hw_breakpoint; n++)
 747         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
 748             (hw_breakpoint[n].len == len || len == -1))
 749             return n;
 750     return -1;
 751 }
 752
 753 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 754                                   target_ulong len, int type)
 755 {
 756     switch (type) {
 757     case GDB_BREAKPOINT_HW:
 758         len = 1;
 759         break;
 760     case GDB_WATCHPOINT_WRITE:
 761     case GDB_WATCHPOINT_ACCESS:
 762         switch (len) {
 763         case 1:
 764             break;
 765         case 2:
 766         case 4:
 767         case 8:
 768             if (addr & (len - 1))
 769                 return -EINVAL;
 770             break;
 771         default:
 772             return -EINVAL;
 773         }
 774         break;
 775     default:
 776         return -ENOSYS;
 777     }
 778
 779     if (nb_hw_breakpoint == 4)
 780         return -ENOBUFS;
 781
 782     if (find_hw_breakpoint(addr, len, type) >= 0)
 783         return -EEXIST;
 784
 785     hw_breakpoint[nb_hw_breakpoint].addr = addr;
 786     hw_breakpoint[nb_hw_breakpoint].len = len;
 787     hw_breakpoint[nb_hw_breakpoint].type = type;
 788     nb_hw_breakpoint++;
 789
 790     return 0;
 791 }
 792
 793 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 794                                   target_ulong len, int type)
 795 {
 796     int n;
 797
 798     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
 799     if (n < 0)
 800         return -ENOENT;
 801
 802     nb_hw_breakpoint--;
 803     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
 804
 805     return 0;
 806 }
 807
 808 void kvm_arch_remove_all_hw_breakpoints(void)
 809 {
 810     nb_hw_breakpoint = 0;
 811 }
 812
 813 static CPUWatchpoint hw_watchpoint;
 814
 815 int kvm_arch_debug(struct kvm_debug_exit_arch *arch_info)
 816 {
 817     int handle = 0;
 818     int n;
 819
 820     if (arch_info->exception == 1) {
 821         if (arch_info->dr6 & (1 << 14)) {
 822             if (cpu_single_env->singlestep_enabled)
 823                 handle = 1;
 824         } else {
 825             for (n = 0; n < 4; n++)
 826                 if (arch_info->dr6 & (1 << n))
 827                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
 828                     case 0x0:
 829                         handle = 1;
 830                         break;
 831                     case 0x1:
 832                         handle = 1;
 833                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
 834                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
 835                         hw_watchpoint.flags = BP_MEM_WRITE;
 836                         break;
 837                     case 0x3:
 838                         handle = 1;
 839                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
 840                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
 841                         hw_watchpoint.flags = BP_MEM_ACCESS;
 842                         break;
 843                     }
 844         }
 845     } else if (kvm_find_sw_breakpoint(arch_info->pc))
 846         handle = 1;
 847
 848     if (!handle)
 849         kvm_update_guest_debug(cpu_single_env,
 850                         (arch_info->exception == 1) ?
 851                         KVM_GUESTDBG_INJECT_DB : KVM_GUESTDBG_INJECT_BP);
 852
 853     return handle;
 854 }
 855
 856 void kvm_arch_update_guest_debug(CPUState *env, struct kvm_guest_debug *dbg)
 857 {
 858     const uint8_t type_code[] = {
 859         [GDB_BREAKPOINT_HW] = 0x0,
 860         [GDB_WATCHPOINT_WRITE] = 0x1,
 861         [GDB_WATCHPOINT_ACCESS] = 0x3
 862     };
 863     const uint8_t len_code[] = {
 864         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
 865     };
 866     int n;
 867
 868     if (!TAILQ_EMPTY(&kvm_sw_breakpoints))
 869         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
 870
 871     if (nb_hw_breakpoint > 0) {
 872         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 873         dbg->arch.debugreg[7] = 0x0600;
 874         for (n = 0; n < nb_hw_breakpoint; n++) {
 875             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
 876             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
 877                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
 878                 (len_code[hw_breakpoint[n].len] << (18 + n*4));
 879         }
 880     }
 881 }
 882 #endif
 883
 884 void kvm_arch_do_ioperm(void *_data)
 885 {
 886     struct ioperm_data *data = _data;
 887     ioperm(data->start_port, data->num, data->turn_on);
 888 }
 889
 890 /*
 891  * Setup x86 specific IRQ routing
 892  */
 893 int kvm_arch_init_irq_routing(void)
 894 {
 895     int i, r;
 896
 897     if (kvm_irqchip && kvm_has_gsi_routing(kvm_context)) {
 898         kvm_clear_gsi_routes(kvm_context);
 899         for (i = 0; i < 8; ++i) {
 900             if (i == 2)
 901                 continue;
 902             r = kvm_add_irq_route(kvm_context, i, KVM_IRQCHIP_PIC_MASTER, i);
 903             if (r < 0)
 904                 return r;
 905         }
 906         for (i = 8; i < 16; ++i) {
 907             r = kvm_add_irq_route(kvm_context, i, KVM_IRQCHIP_PIC_SLAVE, i - 8);
 908             if (r < 0)
 909                 return r;
 910         }
 911         for (i = 0; i < 24; ++i) {
 912             r = kvm_add_irq_route(kvm_context, i, KVM_IRQCHIP_IOAPIC, i);
 913             if (r < 0)
 914                 return r;
 915         }
 916         kvm_commit_irq_routes(kvm_context);
 917     }
 918     return 0;
 919 }
 920
 921 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function,
 922                                       int reg)
 923 {
 924     return kvm_get_supported_cpuid(kvm_context, function, reg);
 925 }