target-i386/kvm.c

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright (C) 2006-2008 Qumranet Technologies
   5  * Copyright IBM, Corp. 2008
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  11  * See the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include <sys/types.h>
  16 #include <sys/ioctl.h>
  17 #include <sys/mman.h>
  18
  19 #include <linux/kvm.h>
  20
  21 #include "qemu-common.h"
  22 #include "sysemu.h"
  23 #include "kvm.h"
  24 #include "cpu.h"
  25 #include "gdbstub.h"
  26
  27 //#define DEBUG_KVM
  28
  29 #ifdef DEBUG_KVM
  30 #define dprintf(fmt, ...) \
  31     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  32 #else
  33 #define dprintf(fmt, ...) \
  34     do { } while (0)
  35 #endif
  36
  37 #ifdef KVM_CAP_EXT_CPUID
  38
  39 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
  40 {
  41     struct kvm_cpuid2 *cpuid;
  42     int r, size;
  43
  44     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
  45     cpuid = (struct kvm_cpuid2 *)qemu_mallocz(size);
  46     cpuid->nent = max;
  47     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
  48     if (r == 0 && cpuid->nent >= max) {
  49         r = -E2BIG;
  50     }
  51     if (r < 0) {
  52         if (r == -E2BIG) {
  53             qemu_free(cpuid);
  54             return NULL;
  55         } else {
  56             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
  57                     strerror(-r));
  58             exit(1);
  59         }
  60     }
  61     return cpuid;
  62 }
  63
  64 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function, int reg)
  65 {
  66     struct kvm_cpuid2 *cpuid;
  67     int i, max;
  68     uint32_t ret = 0;
  69     uint32_t cpuid_1_edx;
  70
  71     if (!kvm_check_extension(env->kvm_state, KVM_CAP_EXT_CPUID)) {
  72         return -1U;
  73     }
  74
  75     max = 1;
  76     while ((cpuid = try_get_cpuid(env->kvm_state, max)) == NULL) {
  77         max *= 2;
  78     }
  79
  80     for (i = 0; i < cpuid->nent; ++i) {
  81         if (cpuid->entries[i].function == function) {
  82             switch (reg) {
  83             case R_EAX:
  84                 ret = cpuid->entries[i].eax;
  85                 break;
  86             case R_EBX:
  87                 ret = cpuid->entries[i].ebx;
  88                 break;
  89             case R_ECX:
  90                 ret = cpuid->entries[i].ecx;
  91                 break;
  92             case R_EDX:
  93                 ret = cpuid->entries[i].edx;
  94                 if (function == 0x80000001) {
  95                     /* On Intel, kvm returns cpuid according to the Intel spec,
  96                      * so add missing bits according to the AMD spec:
  97                      */
  98                     cpuid_1_edx = kvm_arch_get_supported_cpuid(env, 1, R_EDX);
  99                     ret |= cpuid_1_edx & 0xdfeff7ff;
 100                 }
 101                 break;
 102             }
 103         }
 104     }
 105
 106     qemu_free(cpuid);
 107
 108     return ret;
 109 }
 110
 111 #else
 112
 113 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function, int reg)
 114 {
 115     return -1U;
 116 }
 117
 118 #endif
 119
 120 static void kvm_trim_features(uint32_t *features, uint32_t supported)
 121 {
 122     int i;
 123     uint32_t mask;
 124
 125     for (i = 0; i < 32; ++i) {
 126         mask = 1U << i;
 127         if ((*features & mask) && !(supported & mask)) {
 128             *features &= ~mask;
 129         }
 130     }
 131 }
 132
 133 int kvm_arch_init_vcpu(CPUState *env)
 134 {
 135     struct {
 136         struct kvm_cpuid2 cpuid;
 137         struct kvm_cpuid_entry2 entries[100];
 138     } __attribute__((packed)) cpuid_data;
 139     uint32_t limit, i, j, cpuid_i;
 140     uint32_t unused;
 141
 142     env->mp_state = KVM_MP_STATE_RUNNABLE;
 143
 144     kvm_trim_features(&env->cpuid_features,
 145         kvm_arch_get_supported_cpuid(env, 1, R_EDX));
 146
 147     i = env->cpuid_ext_features & CPUID_EXT_HYPERVISOR;
 148     kvm_trim_features(&env->cpuid_ext_features,
 149         kvm_arch_get_supported_cpuid(env, 1, R_ECX));
 150     env->cpuid_ext_features |= i;
 151
 152     kvm_trim_features(&env->cpuid_ext2_features,
 153         kvm_arch_get_supported_cpuid(env, 0x80000001, R_EDX));
 154     kvm_trim_features(&env->cpuid_ext3_features,
 155         kvm_arch_get_supported_cpuid(env, 0x80000001, R_ECX));
 156
 157     cpuid_i = 0;
 158
 159     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 160
 161     for (i = 0; i <= limit; i++) {
 162         struct kvm_cpuid_entry2 *c = &cpuid_data.entries[cpuid_i++];
 163
 164         switch (i) {
 165         case 2: {
 166             /* Keep reading function 2 till all the input is received */
 167             int times;
 168
 169             c->function = i;
 170             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 171                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 172             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 173             times = c->eax & 0xff;
 174
 175             for (j = 1; j < times; ++j) {
 176                 c = &cpuid_data.entries[cpuid_i++];
 177                 c->function = i;
 178                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 179                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 180             }
 181             break;
 182         }
 183         case 4:
 184         case 0xb:
 185         case 0xd:
 186             for (j = 0; ; j++) {
 187                 c->function = i;
 188                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 189                 c->index = j;
 190                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
 191
 192                 if (i == 4 && c->eax == 0)
 193                     break;
 194                 if (i == 0xb && !(c->ecx & 0xff00))
 195                     break;
 196                 if (i == 0xd && c->eax == 0)
 197                     break;
 198
 199                 c = &cpuid_data.entries[cpuid_i++];
 200             }
 201             break;
 202         default:
 203             c->function = i;
 204             c->flags = 0;
 205             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 206             break;
 207         }
 208     }
 209     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
 210
 211     for (i = 0x80000000; i <= limit; i++) {
 212         struct kvm_cpuid_entry2 *c = &cpuid_data.entries[cpuid_i++];
 213
 214         c->function = i;
 215         c->flags = 0;
 216         cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 217     }
 218
 219     cpuid_data.cpuid.nent = cpuid_i;
 220
 221     return kvm_vcpu_ioctl(env, KVM_SET_CPUID2, &cpuid_data);
 222 }
 223
 224 static int kvm_has_msr_star(CPUState *env)
 225 {
 226     static int has_msr_star;
 227     int ret;
 228
 229     /* first time */
 230     if (has_msr_star == 0) {
 231         struct kvm_msr_list msr_list, *kvm_msr_list;
 232
 233         has_msr_star = -1;
 234
 235         /* Obtain MSR list from KVM.  These are the MSRs that we must
 236          * save/restore */
 237         msr_list.nmsrs = 0;
 238         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
 239         if (ret < 0)
 240             return 0;
 241
 242         /* Old kernel modules had a bug and could write beyond the provided
 243            memory. Allocate at least a safe amount of 1K. */
 244         kvm_msr_list = qemu_mallocz(MAX(1024, sizeof(msr_list) +
 245                                               msr_list.nmsrs *
 246                                               sizeof(msr_list.indices[0])));
 247
 248         kvm_msr_list->nmsrs = msr_list.nmsrs;
 249         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
 250         if (ret >= 0) {
 251             int i;
 252
 253             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
 254                 if (kvm_msr_list->indices[i] == MSR_STAR) {
 255                     has_msr_star = 1;
 256                     break;
 257                 }
 258             }
 259         }
 260
 261         free(kvm_msr_list);
 262     }
 263
 264     if (has_msr_star == 1)
 265         return 1;
 266     return 0;
 267 }
 268
 269 int kvm_arch_init(KVMState *s, int smp_cpus)
 270 {
 271     int ret;
 272
 273     /* create vm86 tss.  KVM uses vm86 mode to emulate 16-bit code
 274      * directly.  In order to use vm86 mode, a TSS is needed.  Since this
 275      * must be part of guest physical memory, we need to allocate it.  Older
 276      * versions of KVM just assumed that it would be at the end of physical
 277      * memory but that doesn't work with more than 4GB of memory.  We simply
 278      * refuse to work with those older versions of KVM. */
 279     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_SET_TSS_ADDR);
 280     if (ret <= 0) {
 281         fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
 282         return ret;
 283     }
 284
 285     /* this address is 3 pages before the bios, and the bios should present
 286      * as unavaible memory.  FIXME, need to ensure the e820 map deals with
 287      * this?
 288      */
 289     return kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, 0xfffbd000);
 290 }
 291
 292 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 293 {
 294     lhs->selector = rhs->selector;
 295     lhs->base = rhs->base;
 296     lhs->limit = rhs->limit;
 297     lhs->type = 3;
 298     lhs->present = 1;
 299     lhs->dpl = 3;
 300     lhs->db = 0;
 301     lhs->s = 1;
 302     lhs->l = 0;
 303     lhs->g = 0;
 304     lhs->avl = 0;
 305     lhs->unusable = 0;
 306 }
 307
 308 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 309 {
 310     unsigned flags = rhs->flags;
 311     lhs->selector = rhs->selector;
 312     lhs->base = rhs->base;
 313     lhs->limit = rhs->limit;
 314     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 315     lhs->present = (flags & DESC_P_MASK) != 0;
 316     lhs->dpl = rhs->selector & 3;
 317     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 318     lhs->s = (flags & DESC_S_MASK) != 0;
 319     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 320     lhs->g = (flags & DESC_G_MASK) != 0;
 321     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 322     lhs->unusable = 0;
 323 }
 324
 325 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 326 {
 327     lhs->selector = rhs->selector;
 328     lhs->base = rhs->base;
 329     lhs->limit = rhs->limit;
 330     lhs->flags =
 331         (rhs->type << DESC_TYPE_SHIFT)
 332         | (rhs->present * DESC_P_MASK)
 333         | (rhs->dpl << DESC_DPL_SHIFT)
 334         | (rhs->db << DESC_B_SHIFT)
 335         | (rhs->s * DESC_S_MASK)
 336         | (rhs->l << DESC_L_SHIFT)
 337         | (rhs->g * DESC_G_MASK)
 338         | (rhs->avl * DESC_AVL_MASK);
 339 }
 340
 341 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
 342 {
 343     if (set)
 344         *kvm_reg = *qemu_reg;
 345     else
 346         *qemu_reg = *kvm_reg;
 347 }
 348
 349 static int kvm_getput_regs(CPUState *env, int set)
 350 {
 351     struct kvm_regs regs;
 352     int ret = 0;
 353
 354     if (!set) {
 355         ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 356         if (ret < 0)
 357             return ret;
 358     }
 359
 360     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
 361     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
 362     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
 363     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
 364     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
 365     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
 366     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
 367     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
 368 #ifdef TARGET_X86_64
 369     kvm_getput_reg(&regs.r8, &env->regs[8], set);
 370     kvm_getput_reg(&regs.r9, &env->regs[9], set);
 371     kvm_getput_reg(&regs.r10, &env->regs[10], set);
 372     kvm_getput_reg(&regs.r11, &env->regs[11], set);
 373     kvm_getput_reg(&regs.r12, &env->regs[12], set);
 374     kvm_getput_reg(&regs.r13, &env->regs[13], set);
 375     kvm_getput_reg(&regs.r14, &env->regs[14], set);
 376     kvm_getput_reg(&regs.r15, &env->regs[15], set);
 377 #endif
 378
 379     kvm_getput_reg(&regs.rflags, &env->eflags, set);
 380     kvm_getput_reg(&regs.rip, &env->eip, set);
 381
 382     if (set)
 383         ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 384
 385     return ret;
 386 }
 387
 388 static int kvm_put_fpu(CPUState *env)
 389 {
 390     struct kvm_fpu fpu;
 391     int i;
 392
 393     memset(&fpu, 0, sizeof fpu);
 394     fpu.fsw = env->fpus & ~(7 << 11);
 395     fpu.fsw |= (env->fpstt & 7) << 11;
 396     fpu.fcw = env->fpuc;
 397     for (i = 0; i < 8; ++i)
 398         fpu.ftwx |= (!env->fptags[i]) << i;
 399     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 400     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 401     fpu.mxcsr = env->mxcsr;
 402
 403     return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
 404 }
 405
 406 static int kvm_put_sregs(CPUState *env)
 407 {
 408     struct kvm_sregs sregs;
 409
 410     memcpy(sregs.interrupt_bitmap,
 411            env->interrupt_bitmap,
 412            sizeof(sregs.interrupt_bitmap));
 413
 414     if ((env->eflags & VM_MASK)) {
 415             set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 416             set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 417             set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 418             set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 419             set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 420             set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 421     } else {
 422             set_seg(&sregs.cs, &env->segs[R_CS]);
 423             set_seg(&sregs.ds, &env->segs[R_DS]);
 424             set_seg(&sregs.es, &env->segs[R_ES]);
 425             set_seg(&sregs.fs, &env->segs[R_FS]);
 426             set_seg(&sregs.gs, &env->segs[R_GS]);
 427             set_seg(&sregs.ss, &env->segs[R_SS]);
 428
 429             if (env->cr[0] & CR0_PE_MASK) {
 430                 /* force ss cpl to cs cpl */
 431                 sregs.ss.selector = (sregs.ss.selector & ~3) |
 432                         (sregs.cs.selector & 3);
 433                 sregs.ss.dpl = sregs.ss.selector & 3;
 434             }
 435     }
 436
 437     set_seg(&sregs.tr, &env->tr);
 438     set_seg(&sregs.ldt, &env->ldt);
 439
 440     sregs.idt.limit = env->idt.limit;
 441     sregs.idt.base = env->idt.base;
 442     sregs.gdt.limit = env->gdt.limit;
 443     sregs.gdt.base = env->gdt.base;
 444
 445     sregs.cr0 = env->cr[0];
 446     sregs.cr2 = env->cr[2];
 447     sregs.cr3 = env->cr[3];
 448     sregs.cr4 = env->cr[4];
 449
 450     sregs.cr8 = cpu_get_apic_tpr(env);
 451     sregs.apic_base = cpu_get_apic_base(env);
 452
 453     sregs.efer = env->efer;
 454
 455     return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
 456 }
 457
 458 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
 459                               uint32_t index, uint64_t value)
 460 {
 461     entry->index = index;
 462     entry->data = value;
 463 }
 464
 465 static int kvm_put_msrs(CPUState *env)
 466 {
 467     struct {
 468         struct kvm_msrs info;
 469         struct kvm_msr_entry entries[100];
 470     } msr_data;
 471     struct kvm_msr_entry *msrs = msr_data.entries;
 472     int n = 0;
 473
 474     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
 475     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 476     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 477     if (kvm_has_msr_star(env))
 478         kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
 479     kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
 480     kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
 481 #ifdef TARGET_X86_64
 482     /* FIXME if lm capable */
 483     kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
 484     kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
 485     kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
 486     kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
 487 #endif
 488     msr_data.info.nmsrs = n;
 489
 490     return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
 491
 492 }
 493
 494
 495 static int kvm_get_fpu(CPUState *env)
 496 {
 497     struct kvm_fpu fpu;
 498     int i, ret;
 499
 500     ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
 501     if (ret < 0)
 502         return ret;
 503
 504     env->fpstt = (fpu.fsw >> 11) & 7;
 505     env->fpus = fpu.fsw;
 506     env->fpuc = fpu.fcw;
 507     for (i = 0; i < 8; ++i)
 508         env->fptags[i] = !((fpu.ftwx >> i) & 1);
 509     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
 510     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
 511     env->mxcsr = fpu.mxcsr;
 512
 513     return 0;
 514 }
 515
 516 static int kvm_get_sregs(CPUState *env)
 517 {
 518     struct kvm_sregs sregs;
 519     uint32_t hflags;
 520     int ret;
 521
 522     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 523     if (ret < 0)
 524         return ret;
 525
 526     memcpy(env->interrupt_bitmap,
 527            sregs.interrupt_bitmap,
 528            sizeof(sregs.interrupt_bitmap));
 529
 530     get_seg(&env->segs[R_CS], &sregs.cs);
 531     get_seg(&env->segs[R_DS], &sregs.ds);
 532     get_seg(&env->segs[R_ES], &sregs.es);
 533     get_seg(&env->segs[R_FS], &sregs.fs);
 534     get_seg(&env->segs[R_GS], &sregs.gs);
 535     get_seg(&env->segs[R_SS], &sregs.ss);
 536
 537     get_seg(&env->tr, &sregs.tr);
 538     get_seg(&env->ldt, &sregs.ldt);
 539
 540     env->idt.limit = sregs.idt.limit;
 541     env->idt.base = sregs.idt.base;
 542     env->gdt.limit = sregs.gdt.limit;
 543     env->gdt.base = sregs.gdt.base;
 544
 545     env->cr[0] = sregs.cr0;
 546     env->cr[2] = sregs.cr2;
 547     env->cr[3] = sregs.cr3;
 548     env->cr[4] = sregs.cr4;
 549
 550     cpu_set_apic_base(env, sregs.apic_base);
 551
 552     env->efer = sregs.efer;
 553     //cpu_set_apic_tpr(env, sregs.cr8);
 554
 555 #define HFLAG_COPY_MASK ~( \
 556                         HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
 557                         HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
 558                         HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
 559                         HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
 560
 561
 562
 563     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 564     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
 565     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
 566             (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
 567     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
 568     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
 569             (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
 570
 571     if (env->efer & MSR_EFER_LMA) {
 572         hflags |= HF_LMA_MASK;
 573     }
 574
 575     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
 576         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
 577     } else {
 578         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
 579                 (DESC_B_SHIFT - HF_CS32_SHIFT);
 580         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
 581                 (DESC_B_SHIFT - HF_SS32_SHIFT);
 582         if (!(env->cr[0] & CR0_PE_MASK) ||
 583                    (env->eflags & VM_MASK) ||
 584                    !(hflags & HF_CS32_MASK)) {
 585                 hflags |= HF_ADDSEG_MASK;
 586             } else {
 587                 hflags |= ((env->segs[R_DS].base |
 588                                 env->segs[R_ES].base |
 589                                 env->segs[R_SS].base) != 0) <<
 590                     HF_ADDSEG_SHIFT;
 591             }
 592     }
 593     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
 594
 595     return 0;
 596 }
 597
 598 static int kvm_get_msrs(CPUState *env)
 599 {
 600     struct {
 601         struct kvm_msrs info;
 602         struct kvm_msr_entry entries[100];
 603     } msr_data;
 604     struct kvm_msr_entry *msrs = msr_data.entries;
 605     int ret, i, n;
 606
 607     n = 0;
 608     msrs[n++].index = MSR_IA32_SYSENTER_CS;
 609     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
 610     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
 611     if (kvm_has_msr_star(env))
 612         msrs[n++].index = MSR_STAR;
 613     msrs[n++].index = MSR_IA32_TSC;
 614     msrs[n++].index = MSR_VM_HSAVE_PA;
 615 #ifdef TARGET_X86_64
 616     /* FIXME lm_capable_kernel */
 617     msrs[n++].index = MSR_CSTAR;
 618     msrs[n++].index = MSR_KERNELGSBASE;
 619     msrs[n++].index = MSR_FMASK;
 620     msrs[n++].index = MSR_LSTAR;
 621 #endif
 622     msr_data.info.nmsrs = n;
 623     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
 624     if (ret < 0)
 625         return ret;
 626
 627     for (i = 0; i < ret; i++) {
 628         switch (msrs[i].index) {
 629         case MSR_IA32_SYSENTER_CS:
 630             env->sysenter_cs = msrs[i].data;
 631             break;
 632         case MSR_IA32_SYSENTER_ESP:
 633             env->sysenter_esp = msrs[i].data;
 634             break;
 635         case MSR_IA32_SYSENTER_EIP:
 636             env->sysenter_eip = msrs[i].data;
 637             break;
 638         case MSR_STAR:
 639             env->star = msrs[i].data;
 640             break;
 641 #ifdef TARGET_X86_64
 642         case MSR_CSTAR:
 643             env->cstar = msrs[i].data;
 644             break;
 645         case MSR_KERNELGSBASE:
 646             env->kernelgsbase = msrs[i].data;
 647             break;
 648         case MSR_FMASK:
 649             env->fmask = msrs[i].data;
 650             break;
 651         case MSR_LSTAR:
 652             env->lstar = msrs[i].data;
 653             break;
 654 #endif
 655         case MSR_IA32_TSC:
 656             env->tsc = msrs[i].data;
 657             break;
 658         case MSR_VM_HSAVE_PA:
 659             env->vm_hsave = msrs[i].data;
 660             break;
 661         }
 662     }
 663
 664     return 0;
 665 }
 666
 667 int kvm_arch_put_registers(CPUState *env)
 668 {
 669     int ret;
 670
 671     ret = kvm_getput_regs(env, 1);
 672     if (ret < 0)
 673         return ret;
 674
 675     ret = kvm_put_fpu(env);
 676     if (ret < 0)
 677         return ret;
 678
 679     ret = kvm_put_sregs(env);
 680     if (ret < 0)
 681         return ret;
 682
 683     ret = kvm_put_msrs(env);
 684     if (ret < 0)
 685         return ret;
 686
 687     ret = kvm_put_mp_state(env);
 688     if (ret < 0)
 689         return ret;
 690
 691     ret = kvm_get_mp_state(env);
 692     if (ret < 0)
 693         return ret;
 694
 695     return 0;
 696 }
 697
 698 int kvm_arch_get_registers(CPUState *env)
 699 {
 700     int ret;
 701
 702     ret = kvm_getput_regs(env, 0);
 703     if (ret < 0)
 704         return ret;
 705
 706     ret = kvm_get_fpu(env);
 707     if (ret < 0)
 708         return ret;
 709
 710     ret = kvm_get_sregs(env);
 711     if (ret < 0)
 712         return ret;
 713
 714     ret = kvm_get_msrs(env);
 715     if (ret < 0)
 716         return ret;
 717
 718     return 0;
 719 }
 720
 721 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
 722 {
 723     /* Try to inject an interrupt if the guest can accept it */
 724     if (run->ready_for_interrupt_injection &&
 725         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
 726         (env->eflags & IF_MASK)) {
 727         int irq;
 728
 729         env->interrupt_request &= ~CPU_INTERRUPT_HARD;
 730         irq = cpu_get_pic_interrupt(env);
 731         if (irq >= 0) {
 732             struct kvm_interrupt intr;
 733             intr.irq = irq;
 734             /* FIXME: errors */
 735             dprintf("injected interrupt %d\n", irq);
 736             kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
 737         }
 738     }
 739
 740     /* If we have an interrupt but the guest is not ready to receive an
 741      * interrupt, request an interrupt window exit.  This will
 742      * cause a return to userspace as soon as the guest is ready to
 743      * receive interrupts. */
 744     if ((env->interrupt_request & CPU_INTERRUPT_HARD))
 745         run->request_interrupt_window = 1;
 746     else
 747         run->request_interrupt_window = 0;
 748
 749     dprintf("setting tpr\n");
 750     run->cr8 = cpu_get_apic_tpr(env);
 751
 752     return 0;
 753 }
 754
 755 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
 756 {
 757     if (run->if_flag)
 758         env->eflags |= IF_MASK;
 759     else
 760         env->eflags &= ~IF_MASK;
 761
 762     cpu_set_apic_tpr(env, run->cr8);
 763     cpu_set_apic_base(env, run->apic_base);
 764
 765     return 0;
 766 }
 767
 768 static int kvm_handle_halt(CPUState *env)
 769 {
 770     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
 771           (env->eflags & IF_MASK)) &&
 772         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
 773         env->halted = 1;
 774         env->exception_index = EXCP_HLT;
 775         return 0;
 776     }
 777
 778     return 1;
 779 }
 780
 781 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
 782 {
 783     int ret = 0;
 784
 785     switch (run->exit_reason) {
 786     case KVM_EXIT_HLT:
 787         dprintf("handle_hlt\n");
 788         ret = kvm_handle_halt(env);
 789         break;
 790     }
 791
 792     return ret;
 793 }
 794
 795 #ifdef KVM_CAP_SET_GUEST_DEBUG
 796 int kvm_arch_insert_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
 797 {
 798     const static uint8_t int3 = 0xcc;
 799
 800     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
 801         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1))
 802         return -EINVAL;
 803     return 0;
 804 }
 805
 806 int kvm_arch_remove_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
 807 {
 808     uint8_t int3;
 809
 810     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
 811         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1))
 812         return -EINVAL;
 813     return 0;
 814 }
 815
 816 static struct {
 817     target_ulong addr;
 818     int len;
 819     int type;
 820 } hw_breakpoint[4];
 821
 822 static int nb_hw_breakpoint;
 823
 824 static int find_hw_breakpoint(target_ulong addr, int len, int type)
 825 {
 826     int n;
 827
 828     for (n = 0; n < nb_hw_breakpoint; n++)
 829         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
 830             (hw_breakpoint[n].len == len || len == -1))
 831             return n;
 832     return -1;
 833 }
 834
 835 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 836                                   target_ulong len, int type)
 837 {
 838     switch (type) {
 839     case GDB_BREAKPOINT_HW:
 840         len = 1;
 841         break;
 842     case GDB_WATCHPOINT_WRITE:
 843     case GDB_WATCHPOINT_ACCESS:
 844         switch (len) {
 845         case 1:
 846             break;
 847         case 2:
 848         case 4:
 849         case 8:
 850             if (addr & (len - 1))
 851                 return -EINVAL;
 852             break;
 853         default:
 854             return -EINVAL;
 855         }
 856         break;
 857     default:
 858         return -ENOSYS;
 859     }
 860
 861     if (nb_hw_breakpoint == 4)
 862         return -ENOBUFS;
 863
 864     if (find_hw_breakpoint(addr, len, type) >= 0)
 865         return -EEXIST;
 866
 867     hw_breakpoint[nb_hw_breakpoint].addr = addr;
 868     hw_breakpoint[nb_hw_breakpoint].len = len;
 869     hw_breakpoint[nb_hw_breakpoint].type = type;
 870     nb_hw_breakpoint++;
 871
 872     return 0;
 873 }
 874
 875 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 876                                   target_ulong len, int type)
 877 {
 878     int n;
 879
 880     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
 881     if (n < 0)
 882         return -ENOENT;
 883
 884     nb_hw_breakpoint--;
 885     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
 886
 887     return 0;
 888 }
 889
 890 void kvm_arch_remove_all_hw_breakpoints(void)
 891 {
 892     nb_hw_breakpoint = 0;
 893 }
 894
 895 static CPUWatchpoint hw_watchpoint;
 896
 897 int kvm_arch_debug(struct kvm_debug_exit_arch *arch_info)
 898 {
 899     int handle = 0;
 900     int n;
 901
 902     if (arch_info->exception == 1) {
 903         if (arch_info->dr6 & (1 << 14)) {
 904             if (cpu_single_env->singlestep_enabled)
 905                 handle = 1;
 906         } else {
 907             for (n = 0; n < 4; n++)
 908                 if (arch_info->dr6 & (1 << n))
 909                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
 910                     case 0x0:
 911                         handle = 1;
 912                         break;
 913                     case 0x1:
 914                         handle = 1;
 915                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
 916                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
 917                         hw_watchpoint.flags = BP_MEM_WRITE;
 918                         break;
 919                     case 0x3:
 920                         handle = 1;
 921                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
 922                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
 923                         hw_watchpoint.flags = BP_MEM_ACCESS;
 924                         break;
 925                     }
 926         }
 927     } else if (kvm_find_sw_breakpoint(cpu_single_env, arch_info->pc))
 928         handle = 1;
 929
 930     if (!handle)
 931         kvm_update_guest_debug(cpu_single_env,
 932                         (arch_info->exception == 1) ?
 933                         KVM_GUESTDBG_INJECT_DB : KVM_GUESTDBG_INJECT_BP);
 934
 935     return handle;
 936 }
 937
 938 void kvm_arch_update_guest_debug(CPUState *env, struct kvm_guest_debug *dbg)
 939 {
 940     const uint8_t type_code[] = {
 941         [GDB_BREAKPOINT_HW] = 0x0,
 942         [GDB_WATCHPOINT_WRITE] = 0x1,
 943         [GDB_WATCHPOINT_ACCESS] = 0x3
 944     };
 945     const uint8_t len_code[] = {
 946         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
 947     };
 948     int n;
 949
 950     if (kvm_sw_breakpoints_active(env))
 951         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
 952
 953     if (nb_hw_breakpoint > 0) {
 954         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 955         dbg->arch.debugreg[7] = 0x0600;
 956         for (n = 0; n < nb_hw_breakpoint; n++) {
 957             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
 958             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
 959                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
 960                 (len_code[hw_breakpoint[n].len] << (18 + n*4));
 961         }
 962     }
 963 }
 964 #endif /* KVM_CAP_SET_GUEST_DEBUG */