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
  15 #include "qemu-kvm.h"
  16 #include <libkvm.h>
  17 #include <pthread.h>
  18 #include <sys/utsname.h>
  19 #include <linux/kvm_para.h>
  20
  21 #define MSR_IA32_TSC            0x10
  22
  23 static struct kvm_msr_list *kvm_msr_list;
  24 extern unsigned int kvm_shadow_memory;
  25 extern kvm_context_t kvm_context;
  26 static int kvm_has_msr_star;
  27
  28 static int lm_capable_kernel;
  29
  30 int kvm_arch_qemu_create_context(void)
  31 {
  32     int i;
  33     struct utsname utsname;
  34
  35     uname(&utsname);
  36     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
  37
  38     if (kvm_shadow_memory)
  39         kvm_set_shadow_pages(kvm_context, kvm_shadow_memory);
  40
  41     kvm_msr_list = kvm_get_msr_list(kvm_context);
  42     if (!kvm_msr_list)
  43                 return -1;
  44     for (i = 0; i < kvm_msr_list->nmsrs; ++i)
  45         if (kvm_msr_list->indices[i] == MSR_STAR)
  46             kvm_has_msr_star = 1;
  47         return 0;
  48 }
  49
  50 static void set_msr_entry(struct kvm_msr_entry *entry, uint32_t index,
  51                           uint64_t data)
  52 {
  53     entry->index = index;
  54     entry->data  = data;
  55 }
  56
  57 /* returns 0 on success, non-0 on failure */
  58 static int get_msr_entry(struct kvm_msr_entry *entry, CPUState *env)
  59 {
  60         switch (entry->index) {
  61         case MSR_IA32_SYSENTER_CS:
  62             env->sysenter_cs  = entry->data;
  63             break;
  64         case MSR_IA32_SYSENTER_ESP:
  65             env->sysenter_esp = entry->data;
  66             break;
  67         case MSR_IA32_SYSENTER_EIP:
  68             env->sysenter_eip = entry->data;
  69             break;
  70         case MSR_STAR:
  71             env->star         = entry->data;
  72             break;
  73 #ifdef TARGET_X86_64
  74         case MSR_CSTAR:
  75             env->cstar        = entry->data;
  76             break;
  77         case MSR_KERNELGSBASE:
  78             env->kernelgsbase = entry->data;
  79             break;
  80         case MSR_FMASK:
  81             env->fmask        = entry->data;
  82             break;
  83         case MSR_LSTAR:
  84             env->lstar        = entry->data;
  85             break;
  86 #endif
  87         case MSR_IA32_TSC:
  88             env->tsc          = entry->data;
  89             break;
  90         default:
  91             printf("Warning unknown msr index 0x%x\n", entry->index);
  92             return 1;
  93         }
  94         return 0;
  95 }
  96
  97 #ifdef TARGET_X86_64
  98 #define MSR_COUNT 9
  99 #else
 100 #define MSR_COUNT 5
 101 #endif
 102
 103 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 104 {
 105     lhs->selector = rhs->selector;
 106     lhs->base = rhs->base;
 107     lhs->limit = rhs->limit;
 108     lhs->type = 3;
 109     lhs->present = 1;
 110     lhs->dpl = 3;
 111     lhs->db = 0;
 112     lhs->s = 1;
 113     lhs->l = 0;
 114     lhs->g = 0;
 115     lhs->avl = 0;
 116     lhs->unusable = 0;
 117 }
 118
 119 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 120 {
 121     unsigned flags = rhs->flags;
 122     lhs->selector = rhs->selector;
 123     lhs->base = rhs->base;
 124     lhs->limit = rhs->limit;
 125     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 126     lhs->present = (flags & DESC_P_MASK) != 0;
 127     lhs->dpl = rhs->selector & 3;
 128     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 129     lhs->s = (flags & DESC_S_MASK) != 0;
 130     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 131     lhs->g = (flags & DESC_G_MASK) != 0;
 132     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 133     lhs->unusable = 0;
 134 }
 135
 136 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 137 {
 138     lhs->selector = rhs->selector;
 139     lhs->base = rhs->base;
 140     lhs->limit = rhs->limit;
 141     lhs->flags =
 142         (rhs->type << DESC_TYPE_SHIFT)
 143         | (rhs->present * DESC_P_MASK)
 144         | (rhs->dpl << DESC_DPL_SHIFT)
 145         | (rhs->db << DESC_B_SHIFT)
 146         | (rhs->s * DESC_S_MASK)
 147         | (rhs->l << DESC_L_SHIFT)
 148         | (rhs->g * DESC_G_MASK)
 149         | (rhs->avl * DESC_AVL_MASK);
 150 }
 151
 152 /* the reset values of qemu are not compatible to SVM
 153  * this function is used to fix the segment descriptor values */
 154 static void fix_realmode_dataseg(struct kvm_segment *seg)
 155 {
 156         seg->type = 0x02;
 157         seg->present = 1;
 158         seg->s = 1;
 159 }
 160
 161 void kvm_arch_load_regs(CPUState *env)
 162 {
 163     struct kvm_regs regs;
 164     struct kvm_fpu fpu;
 165     struct kvm_sregs sregs;
 166     struct kvm_msr_entry msrs[MSR_COUNT];
 167     int rc, n, i;
 168
 169     regs.rax = env->regs[R_EAX];
 170     regs.rbx = env->regs[R_EBX];
 171     regs.rcx = env->regs[R_ECX];
 172     regs.rdx = env->regs[R_EDX];
 173     regs.rsi = env->regs[R_ESI];
 174     regs.rdi = env->regs[R_EDI];
 175     regs.rsp = env->regs[R_ESP];
 176     regs.rbp = env->regs[R_EBP];
 177 #ifdef TARGET_X86_64
 178     regs.r8 = env->regs[8];
 179     regs.r9 = env->regs[9];
 180     regs.r10 = env->regs[10];
 181     regs.r11 = env->regs[11];
 182     regs.r12 = env->regs[12];
 183     regs.r13 = env->regs[13];
 184     regs.r14 = env->regs[14];
 185     regs.r15 = env->regs[15];
 186 #endif
 187
 188     regs.rflags = env->eflags;
 189     regs.rip = env->eip;
 190
 191     kvm_set_regs(kvm_context, env->cpu_index, &regs);
 192
 193     memset(&fpu, 0, sizeof fpu);
 194     fpu.fsw = env->fpus & ~(7 << 11);
 195     fpu.fsw |= (env->fpstt & 7) << 11;
 196     fpu.fcw = env->fpuc;
 197     for (i = 0; i < 8; ++i)
 198         fpu.ftwx |= (!env->fptags[i]) << i;
 199     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 200     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 201     fpu.mxcsr = env->mxcsr;
 202     kvm_set_fpu(kvm_context, env->cpu_index, &fpu);
 203
 204     memcpy(sregs.interrupt_bitmap, env->kvm_interrupt_bitmap, sizeof(sregs.interrupt_bitmap));
 205
 206     if ((env->eflags & VM_MASK)) {
 207             set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 208             set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 209             set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 210             set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 211             set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 212             set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 213     } else {
 214             set_seg(&sregs.cs, &env->segs[R_CS]);
 215             set_seg(&sregs.ds, &env->segs[R_DS]);
 216             set_seg(&sregs.es, &env->segs[R_ES]);
 217             set_seg(&sregs.fs, &env->segs[R_FS]);
 218             set_seg(&sregs.gs, &env->segs[R_GS]);
 219             set_seg(&sregs.ss, &env->segs[R_SS]);
 220
 221             if (env->cr[0] & CR0_PE_MASK) {
 222                 /* force ss cpl to cs cpl */
 223                 sregs.ss.selector = (sregs.ss.selector & ~3) |
 224                         (sregs.cs.selector & 3);
 225                 sregs.ss.dpl = sregs.ss.selector & 3;
 226             }
 227
 228             if (!(env->cr[0] & CR0_PG_MASK)) {
 229                     fix_realmode_dataseg(&sregs.cs);
 230                     fix_realmode_dataseg(&sregs.ds);
 231                     fix_realmode_dataseg(&sregs.es);
 232                     fix_realmode_dataseg(&sregs.fs);
 233                     fix_realmode_dataseg(&sregs.gs);
 234                     fix_realmode_dataseg(&sregs.ss);
 235             }
 236     }
 237
 238     set_seg(&sregs.tr, &env->tr);
 239     set_seg(&sregs.ldt, &env->ldt);
 240
 241     sregs.idt.limit = env->idt.limit;
 242     sregs.idt.base = env->idt.base;
 243     sregs.gdt.limit = env->gdt.limit;
 244     sregs.gdt.base = env->gdt.base;
 245
 246     sregs.cr0 = env->cr[0];
 247     sregs.cr2 = env->cr[2];
 248     sregs.cr3 = env->cr[3];
 249     sregs.cr4 = env->cr[4];
 250
 251     sregs.apic_base = cpu_get_apic_base(env);
 252     sregs.efer = env->efer;
 253     sregs.cr8 = cpu_get_apic_tpr(env);
 254
 255     kvm_set_sregs(kvm_context, env->cpu_index, &sregs);
 256
 257     /* msrs */
 258     n = 0;
 259     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_CS,  env->sysenter_cs);
 260     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 261     set_msr_entry(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 262     if (kvm_has_msr_star)
 263         set_msr_entry(&msrs[n++], MSR_STAR,              env->star);
 264     set_msr_entry(&msrs[n++], MSR_IA32_TSC, env->tsc);
 265 #ifdef TARGET_X86_64
 266     if (lm_capable_kernel) {
 267         set_msr_entry(&msrs[n++], MSR_CSTAR,             env->cstar);
 268         set_msr_entry(&msrs[n++], MSR_KERNELGSBASE,      env->kernelgsbase);
 269         set_msr_entry(&msrs[n++], MSR_FMASK,             env->fmask);
 270         set_msr_entry(&msrs[n++], MSR_LSTAR  ,           env->lstar);
 271     }
 272 #endif
 273
 274     rc = kvm_set_msrs(kvm_context, env->cpu_index, msrs, n);
 275     if (rc == -1)
 276         perror("kvm_set_msrs FAILED");
 277 }
 278
 279
 280 void kvm_arch_save_regs(CPUState *env)
 281 {
 282     struct kvm_regs regs;
 283     struct kvm_fpu fpu;
 284     struct kvm_sregs sregs;
 285     struct kvm_msr_entry msrs[MSR_COUNT];
 286     uint32_t hflags;
 287     uint32_t i, n, rc;
 288
 289     kvm_get_regs(kvm_context, env->cpu_index, &regs);
 290
 291     env->regs[R_EAX] = regs.rax;
 292     env->regs[R_EBX] = regs.rbx;
 293     env->regs[R_ECX] = regs.rcx;
 294     env->regs[R_EDX] = regs.rdx;
 295     env->regs[R_ESI] = regs.rsi;
 296     env->regs[R_EDI] = regs.rdi;
 297     env->regs[R_ESP] = regs.rsp;
 298     env->regs[R_EBP] = regs.rbp;
 299 #ifdef TARGET_X86_64
 300     env->regs[8] = regs.r8;
 301     env->regs[9] = regs.r9;
 302     env->regs[10] = regs.r10;
 303     env->regs[11] = regs.r11;
 304     env->regs[12] = regs.r12;
 305     env->regs[13] = regs.r13;
 306     env->regs[14] = regs.r14;
 307     env->regs[15] = regs.r15;
 308 #endif
 309
 310     env->eflags = regs.rflags;
 311     env->eip = regs.rip;
 312
 313     kvm_get_fpu(kvm_context, env->cpu_index, &fpu);
 314     env->fpstt = (fpu.fsw >> 11) & 7;
 315     env->fpus = fpu.fsw;
 316     env->fpuc = fpu.fcw;
 317     for (i = 0; i < 8; ++i)
 318         env->fptags[i] = !((fpu.ftwx >> i) & 1);
 319     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
 320     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
 321     env->mxcsr = fpu.mxcsr;
 322
 323     kvm_get_sregs(kvm_context, env->cpu_index, &sregs);
 324
 325     memcpy(env->kvm_interrupt_bitmap, sregs.interrupt_bitmap, sizeof(env->kvm_interrupt_bitmap));
 326
 327     get_seg(&env->segs[R_CS], &sregs.cs);
 328     get_seg(&env->segs[R_DS], &sregs.ds);
 329     get_seg(&env->segs[R_ES], &sregs.es);
 330     get_seg(&env->segs[R_FS], &sregs.fs);
 331     get_seg(&env->segs[R_GS], &sregs.gs);
 332     get_seg(&env->segs[R_SS], &sregs.ss);
 333
 334     get_seg(&env->tr, &sregs.tr);
 335     get_seg(&env->ldt, &sregs.ldt);
 336
 337     env->idt.limit = sregs.idt.limit;
 338     env->idt.base = sregs.idt.base;
 339     env->gdt.limit = sregs.gdt.limit;
 340     env->gdt.base = sregs.gdt.base;
 341
 342     env->cr[0] = sregs.cr0;
 343     env->cr[2] = sregs.cr2;
 344     env->cr[3] = sregs.cr3;
 345     env->cr[4] = sregs.cr4;
 346
 347     cpu_set_apic_base(env, sregs.apic_base);
 348
 349     env->efer = sregs.efer;
 350     //cpu_set_apic_tpr(env, sregs.cr8);
 351
 352 #define HFLAG_COPY_MASK ~( \
 353                         HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
 354                         HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
 355                         HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
 356                         HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
 357
 358
 359
 360     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 361     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
 362     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
 363             (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
 364     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
 365     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
 366             (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
 367
 368     if (env->efer & MSR_EFER_LMA) {
 369         hflags |= HF_LMA_MASK;
 370     }
 371
 372     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
 373         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
 374     } else {
 375         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
 376                 (DESC_B_SHIFT - HF_CS32_SHIFT);
 377         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
 378                 (DESC_B_SHIFT - HF_SS32_SHIFT);
 379         if (!(env->cr[0] & CR0_PE_MASK) ||
 380                    (env->eflags & VM_MASK) ||
 381                    !(hflags & HF_CS32_MASK)) {
 382                 hflags |= HF_ADDSEG_MASK;
 383             } else {
 384                 hflags |= ((env->segs[R_DS].base |
 385                                 env->segs[R_ES].base |
 386                                 env->segs[R_SS].base) != 0) <<
 387                     HF_ADDSEG_SHIFT;
 388             }
 389     }
 390     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
 391     env->cc_src = env->eflags & (CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C);
 392     env->df = 1 - (2 * ((env->eflags >> 10) & 1));
 393     env->cc_op = CC_OP_EFLAGS;
 394     env->eflags &= ~(DF_MASK | CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C);
 395
 396     /* msrs */
 397     n = 0;
 398     msrs[n++].index = MSR_IA32_SYSENTER_CS;
 399     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
 400     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
 401     if (kvm_has_msr_star)
 402         msrs[n++].index = MSR_STAR;
 403     msrs[n++].index = MSR_IA32_TSC;
 404 #ifdef TARGET_X86_64
 405     if (lm_capable_kernel) {
 406         msrs[n++].index = MSR_CSTAR;
 407         msrs[n++].index = MSR_KERNELGSBASE;
 408         msrs[n++].index = MSR_FMASK;
 409         msrs[n++].index = MSR_LSTAR;
 410     }
 411 #endif
 412     rc = kvm_get_msrs(kvm_context, env->cpu_index, msrs, n);
 413     if (rc == -1) {
 414         perror("kvm_get_msrs FAILED");
 415     }
 416     else {
 417         n = rc; /* actual number of MSRs */
 418         for (i=0 ; i<n; i++) {
 419             if (get_msr_entry(&msrs[i], env))
 420                 return;
 421         }
 422     }
 423 }
 424
 425 static void host_cpuid(uint32_t function, uint32_t *eax, uint32_t *ebx,
 426                        uint32_t *ecx, uint32_t *edx)
 427 {
 428     uint32_t vec[4];
 429
 430 #ifdef __x86_64__
 431     asm volatile("cpuid"
 432                  : "=a"(vec[0]), "=b"(vec[1]),
 433                    "=c"(vec[2]), "=d"(vec[3])
 434                  : "0"(function) : "cc");
 435 #else
 436     asm volatile("pusha \n\t"
 437                  "cpuid \n\t"
 438                  "mov %%eax, 0(%1) \n\t"
 439                  "mov %%ebx, 4(%1) \n\t"
 440                  "mov %%ecx, 8(%1) \n\t"
 441                  "mov %%edx, 12(%1) \n\t"
 442                  "popa"
 443                  : : "a"(function), "S"(vec)
 444                  : "memory", "cc");
 445 #endif
 446
 447     if (eax)
 448         *eax = vec[0];
 449     if (ebx)
 450         *ebx = vec[1];
 451     if (ecx)
 452         *ecx = vec[2];
 453     if (edx)
 454         *edx = vec[3];
 455 }
 456
 457
 458 static void do_cpuid_ent(struct kvm_cpuid_entry *e, uint32_t function,
 459                          CPUState *env)
 460 {
 461     env->regs[R_EAX] = function;
 462     qemu_kvm_cpuid_on_env(env);
 463     e->function = function;
 464     e->eax = env->regs[R_EAX];
 465     e->ebx = env->regs[R_EBX];
 466     e->ecx = env->regs[R_ECX];
 467     e->edx = env->regs[R_EDX];
 468     if (function == 0x80000001) {
 469         uint32_t h_eax, h_edx;
 470
 471         host_cpuid(function, &h_eax, NULL, NULL, &h_edx);
 472
 473         // long mode
 474         if ((h_edx & 0x20000000) == 0 || !lm_capable_kernel)
 475             e->edx &= ~0x20000000u;
 476         // syscall
 477         if ((h_edx & 0x00000800) == 0)
 478             e->edx &= ~0x00000800u;
 479         // nx
 480         if ((h_edx & 0x00100000) == 0)
 481             e->edx &= ~0x00100000u;
 482         // svm
 483         if (e->ecx & 4)
 484             e->ecx &= ~4u;
 485     }
 486     // sysenter isn't supported on compatibility mode on AMD.  and syscall
 487     // isn't supported in compatibility mode on Intel.  so advertise the
 488     // actuall cpu, and say goodbye to migration between different vendors
 489     // is you use compatibility mode.
 490     if (function == 0) {
 491         uint32_t bcd[3];
 492
 493         host_cpuid(0, NULL, &bcd[0], &bcd[1], &bcd[2]);
 494         e->ebx = bcd[0];
 495         e->ecx = bcd[1];
 496         e->edx = bcd[2];
 497     }
 498 }
 499
 500 int kvm_arch_qemu_init_env(CPUState *cenv)
 501 {
 502     struct kvm_cpuid_entry cpuid_ent[100];
 503 #ifdef KVM_CPUID_SIGNATURE
 504     struct kvm_cpuid_entry *pv_ent;
 505     uint32_t signature[3];
 506 #endif
 507     int cpuid_nent = 0;
 508     CPUState copy;
 509     uint32_t i, limit;
 510     int has_clocksource = 0;
 511 #ifdef KVM_CAP_CLOCKSOURCE
 512     has_clocksource = kvm_check_extension(kvm_context, KVM_CAP_CLOCKSOURCE);
 513 #endif
 514
 515     copy = *cenv;
 516
 517 #ifdef KVM_CPUID_SIGNATURE
 518     /* Paravirtualization CPUIDs */
 519     memcpy(signature, "KVMKVMKVM", 12);
 520     pv_ent = &cpuid_ent[cpuid_nent++];
 521     memset(pv_ent, 0, sizeof(*pv_ent));
 522     pv_ent->function = KVM_CPUID_SIGNATURE;
 523     pv_ent->eax = 0;
 524     pv_ent->ebx = signature[0];
 525     pv_ent->ecx = signature[1];
 526     pv_ent->edx = signature[2];
 527
 528     pv_ent = &cpuid_ent[cpuid_nent++];
 529     memset(pv_ent, 0, sizeof(*pv_ent));
 530     pv_ent->function = KVM_CPUID_FEATURES;
 531 #ifdef KVM_CAP_CLOCKSOURCE
 532     pv_ent->eax = (has_clocksource << KVM_FEATURE_CLOCKSOURCE);
 533 #endif
 534 #endif
 535
 536     copy.regs[R_EAX] = 0;
 537     qemu_kvm_cpuid_on_env(&copy);
 538     limit = copy.regs[R_EAX];
 539
 540     for (i = 0; i <= limit; ++i)
 541         do_cpuid_ent(&cpuid_ent[cpuid_nent++], i, &copy);
 542
 543     copy.regs[R_EAX] = 0x80000000;
 544     qemu_kvm_cpuid_on_env(&copy);
 545     limit = copy.regs[R_EAX];
 546
 547     for (i = 0x80000000; i <= limit; ++i)
 548         do_cpuid_ent(&cpuid_ent[cpuid_nent++], i, &copy);
 549
 550     kvm_setup_cpuid(kvm_context, cenv->cpu_index, cpuid_nent, cpuid_ent);
 551     return 0;
 552 }
 553
 554 int kvm_arch_halt(void *opaque, int vcpu)
 555 {
 556     CPUState *env = cpu_single_env;
 557
 558     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
 559           (env->eflags & IF_MASK))) {
 560             env->hflags |= HF_HALTED_MASK;
 561             env->exception_index = EXCP_HLT;
 562     }
 563     return 1;
 564 }
 565
 566 void kvm_arch_pre_kvm_run(void *opaque, int vcpu)
 567 {
 568     CPUState *env = cpu_single_env;
 569
 570     if (!kvm_irqchip_in_kernel(kvm_context))
 571         kvm_set_cr8(kvm_context, vcpu, cpu_get_apic_tpr(env));
 572 }
 573
 574 void kvm_arch_post_kvm_run(void *opaque, int vcpu)
 575 {
 576     CPUState *env = qemu_kvm_cpu_env(vcpu);
 577     cpu_single_env = env;
 578
 579     env->eflags = kvm_get_interrupt_flag(kvm_context, vcpu)
 580         ? env->eflags | IF_MASK : env->eflags & ~IF_MASK;
 581     env->ready_for_interrupt_injection
 582         = kvm_is_ready_for_interrupt_injection(kvm_context, vcpu);
 583
 584     cpu_set_apic_tpr(env, kvm_get_cr8(kvm_context, vcpu));
 585     cpu_set_apic_base(env, kvm_get_apic_base(kvm_context, vcpu));
 586 }
 587
 588 int kvm_arch_has_work(CPUState *env)
 589 {
 590     if ((env->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_EXIT)) &&
 591         (env->eflags & IF_MASK))
 592         return 1;
 593     return 0;
 594 }
 595
 596 int kvm_arch_try_push_interrupts(void *opaque)
 597 {
 598     CPUState *env = cpu_single_env;
 599     int r, irq;
 600
 601     if (env->ready_for_interrupt_injection &&
 602         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
 603         (env->eflags & IF_MASK)) {
 604             env->interrupt_request &= ~CPU_INTERRUPT_HARD;
 605             irq = cpu_get_pic_interrupt(env);
 606             if (irq >= 0) {
 607                 r = kvm_inject_irq(kvm_context, env->cpu_index, irq);
 608                 if (r < 0)
 609                     printf("cpu %d fail inject %x\n", env->cpu_index, irq);
 610             }
 611     }
 612
 613     return (env->interrupt_request & CPU_INTERRUPT_HARD) != 0;
 614 }
 615
 616 void kvm_arch_update_regs_for_sipi(CPUState *env)
 617 {
 618     SegmentCache cs = env->segs[R_CS];
 619
 620     kvm_arch_save_regs(env);
 621     env->segs[R_CS] = cs;
 622     env->eip = 0;
 623     kvm_arch_load_regs(env);
 624 }
 625
 626 int handle_tpr_access(void *opaque, int vcpu,
 627                              uint64_t rip, int is_write)
 628 {
 629     kvm_tpr_access_report(cpu_single_env, rip, is_write);
 630     return 0;
 631 }