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 #include <sys/utsname.h>
  19
  20 #include <linux/kvm.h>
  21 #include <linux/kvm_para.h>
  22
  23 #include "qemu-common.h"
  24 #include "sysemu.h"
  25 #include "kvm.h"
  26 #include "kvm_i386.h"
  27 #include "cpu.h"
  28 #include "gdbstub.h"
  29 #include "host-utils.h"
  30 #include "hw/pc.h"
  31 #include "hw/apic.h"
  32 #include "ioport.h"
  33 #include "hyperv.h"
  34 #include "hw/pci.h"
  35
  36 //#define DEBUG_KVM
  37
  38 #ifdef DEBUG_KVM
  39 #define DPRINTF(fmt, ...) \
  40     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  41 #else
  42 #define DPRINTF(fmt, ...) \
  43     do { } while (0)
  44 #endif
  45
  46 #define MSR_KVM_WALL_CLOCK  0x11
  47 #define MSR_KVM_SYSTEM_TIME 0x12
  48
  49 #ifndef BUS_MCEERR_AR
  50 #define BUS_MCEERR_AR 4
  51 #endif
  52 #ifndef BUS_MCEERR_AO
  53 #define BUS_MCEERR_AO 5
  54 #endif
  55
  56 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  57     KVM_CAP_INFO(SET_TSS_ADDR),
  58     KVM_CAP_INFO(EXT_CPUID),
  59     KVM_CAP_INFO(MP_STATE),
  60     KVM_CAP_LAST_INFO
  61 };
  62
  63 static bool has_msr_star;
  64 static bool has_msr_hsave_pa;
  65 static bool has_msr_tsc_deadline;
  66 static bool has_msr_async_pf_en;
  67 static bool has_msr_pv_eoi_en;
  68 static bool has_msr_misc_enable;
  69 static int lm_capable_kernel;
  70
  71 bool kvm_allows_irq0_override(void)
  72 {
  73     return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
  74 }
  75
  76 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
  77 {
  78     struct kvm_cpuid2 *cpuid;
  79     int r, size;
  80
  81     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
  82     cpuid = (struct kvm_cpuid2 *)g_malloc0(size);
  83     cpuid->nent = max;
  84     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
  85     if (r == 0 && cpuid->nent >= max) {
  86         r = -E2BIG;
  87     }
  88     if (r < 0) {
  89         if (r == -E2BIG) {
  90             g_free(cpuid);
  91             return NULL;
  92         } else {
  93             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
  94                     strerror(-r));
  95             exit(1);
  96         }
  97     }
  98     return cpuid;
  99 }
 100
 101 struct kvm_para_features {
 102     int cap;
 103     int feature;
 104 } para_features[] = {
 105     { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 106     { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 107     { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 108     { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
 109     { -1, -1 }
 110 };
 111
 112 static int get_para_features(KVMState *s)
 113 {
 114     int i, features = 0;
 115
 116     for (i = 0; i < ARRAY_SIZE(para_features) - 1; i++) {
 117         if (kvm_check_extension(s, para_features[i].cap)) {
 118             features |= (1 << para_features[i].feature);
 119         }
 120     }
 121
 122     return features;
 123 }
 124
 125
 126 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
 127                                       uint32_t index, int reg)
 128 {
 129     struct kvm_cpuid2 *cpuid;
 130     int i, max;
 131     uint32_t ret = 0;
 132     uint32_t cpuid_1_edx;
 133     int has_kvm_features = 0;
 134
 135     max = 1;
 136     while ((cpuid = try_get_cpuid(s, max)) == NULL) {
 137         max *= 2;
 138     }
 139
 140     for (i = 0; i < cpuid->nent; ++i) {
 141         if (cpuid->entries[i].function == function &&
 142             cpuid->entries[i].index == index) {
 143             if (cpuid->entries[i].function == KVM_CPUID_FEATURES) {
 144                 has_kvm_features = 1;
 145             }
 146             switch (reg) {
 147             case R_EAX:
 148                 ret = cpuid->entries[i].eax;
 149                 break;
 150             case R_EBX:
 151                 ret = cpuid->entries[i].ebx;
 152                 break;
 153             case R_ECX:
 154                 ret = cpuid->entries[i].ecx;
 155                 break;
 156             case R_EDX:
 157                 ret = cpuid->entries[i].edx;
 158                 switch (function) {
 159                 case 1:
 160                     /* KVM before 2.6.30 misreports the following features */
 161                     ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
 162                     break;
 163                 case 0x80000001:
 164                     /* On Intel, kvm returns cpuid according to the Intel spec,
 165                      * so add missing bits according to the AMD spec:
 166                      */
 167                     cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
 168                     ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
 169                     break;
 170                 }
 171                 break;
 172             }
 173         }
 174     }
 175
 176     g_free(cpuid);
 177
 178     /* fallback for older kernels */
 179     if (!has_kvm_features && (function == KVM_CPUID_FEATURES)) {
 180         ret = get_para_features(s);
 181     }
 182
 183     return ret;
 184 }
 185
 186 typedef struct HWPoisonPage {
 187     ram_addr_t ram_addr;
 188     QLIST_ENTRY(HWPoisonPage) list;
 189 } HWPoisonPage;
 190
 191 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
 192     QLIST_HEAD_INITIALIZER(hwpoison_page_list);
 193
 194 static void kvm_unpoison_all(void *param)
 195 {
 196     HWPoisonPage *page, *next_page;
 197
 198     QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
 199         QLIST_REMOVE(page, list);
 200         qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
 201         g_free(page);
 202     }
 203 }
 204
 205 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
 206 {
 207     HWPoisonPage *page;
 208
 209     QLIST_FOREACH(page, &hwpoison_page_list, list) {
 210         if (page->ram_addr == ram_addr) {
 211             return;
 212         }
 213     }
 214     page = g_malloc(sizeof(HWPoisonPage));
 215     page->ram_addr = ram_addr;
 216     QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
 217 }
 218
 219 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
 220                                      int *max_banks)
 221 {
 222     int r;
 223
 224     r = kvm_check_extension(s, KVM_CAP_MCE);
 225     if (r > 0) {
 226         *max_banks = r;
 227         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
 228     }
 229     return -ENOSYS;
 230 }
 231
 232 static void kvm_mce_inject(CPUX86State *env, target_phys_addr_t paddr, int code)
 233 {
 234     uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
 235                       MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
 236     uint64_t mcg_status = MCG_STATUS_MCIP;
 237
 238     if (code == BUS_MCEERR_AR) {
 239         status |= MCI_STATUS_AR | 0x134;
 240         mcg_status |= MCG_STATUS_EIPV;
 241     } else {
 242         status |= 0xc0;
 243         mcg_status |= MCG_STATUS_RIPV;
 244     }
 245     cpu_x86_inject_mce(NULL, env, 9, status, mcg_status, paddr,
 246                        (MCM_ADDR_PHYS << 6) | 0xc,
 247                        cpu_x86_support_mca_broadcast(env) ?
 248                        MCE_INJECT_BROADCAST : 0);
 249 }
 250
 251 static void hardware_memory_error(void)
 252 {
 253     fprintf(stderr, "Hardware memory error!\n");
 254     exit(1);
 255 }
 256
 257 int kvm_arch_on_sigbus_vcpu(CPUX86State *env, int code, void *addr)
 258 {
 259     ram_addr_t ram_addr;
 260     target_phys_addr_t paddr;
 261
 262     if ((env->mcg_cap & MCG_SER_P) && addr
 263         && (code == BUS_MCEERR_AR || code == BUS_MCEERR_AO)) {
 264         if (qemu_ram_addr_from_host(addr, &ram_addr) ||
 265             !kvm_physical_memory_addr_from_host(env->kvm_state, addr, &paddr)) {
 266             fprintf(stderr, "Hardware memory error for memory used by "
 267                     "QEMU itself instead of guest system!\n");
 268             /* Hope we are lucky for AO MCE */
 269             if (code == BUS_MCEERR_AO) {
 270                 return 0;
 271             } else {
 272                 hardware_memory_error();
 273             }
 274         }
 275         kvm_hwpoison_page_add(ram_addr);
 276         kvm_mce_inject(env, paddr, code);
 277     } else {
 278         if (code == BUS_MCEERR_AO) {
 279             return 0;
 280         } else if (code == BUS_MCEERR_AR) {
 281             hardware_memory_error();
 282         } else {
 283             return 1;
 284         }
 285     }
 286     return 0;
 287 }
 288
 289 int kvm_arch_on_sigbus(int code, void *addr)
 290 {
 291     if ((first_cpu->mcg_cap & MCG_SER_P) && addr && code == BUS_MCEERR_AO) {
 292         ram_addr_t ram_addr;
 293         target_phys_addr_t paddr;
 294
 295         /* Hope we are lucky for AO MCE */
 296         if (qemu_ram_addr_from_host(addr, &ram_addr) ||
 297             !kvm_physical_memory_addr_from_host(first_cpu->kvm_state, addr,
 298                                                 &paddr)) {
 299             fprintf(stderr, "Hardware memory error for memory used by "
 300                     "QEMU itself instead of guest system!: %p\n", addr);
 301             return 0;
 302         }
 303         kvm_hwpoison_page_add(ram_addr);
 304         kvm_mce_inject(first_cpu, paddr, code);
 305     } else {
 306         if (code == BUS_MCEERR_AO) {
 307             return 0;
 308         } else if (code == BUS_MCEERR_AR) {
 309             hardware_memory_error();
 310         } else {
 311             return 1;
 312         }
 313     }
 314     return 0;
 315 }
 316
 317 static int kvm_inject_mce_oldstyle(CPUX86State *env)
 318 {
 319     if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
 320         unsigned int bank, bank_num = env->mcg_cap & 0xff;
 321         struct kvm_x86_mce mce;
 322
 323         env->exception_injected = -1;
 324
 325         /*
 326          * There must be at least one bank in use if an MCE is pending.
 327          * Find it and use its values for the event injection.
 328          */
 329         for (bank = 0; bank < bank_num; bank++) {
 330             if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
 331                 break;
 332             }
 333         }
 334         assert(bank < bank_num);
 335
 336         mce.bank = bank;
 337         mce.status = env->mce_banks[bank * 4 + 1];
 338         mce.mcg_status = env->mcg_status;
 339         mce.addr = env->mce_banks[bank * 4 + 2];
 340         mce.misc = env->mce_banks[bank * 4 + 3];
 341
 342         return kvm_vcpu_ioctl(env, KVM_X86_SET_MCE, &mce);
 343     }
 344     return 0;
 345 }
 346
 347 static void cpu_update_state(void *opaque, int running, RunState state)
 348 {
 349     CPUX86State *env = opaque;
 350
 351     if (running) {
 352         env->tsc_valid = false;
 353     }
 354 }
 355
 356 int kvm_arch_init_vcpu(CPUX86State *env)
 357 {
 358     struct {
 359         struct kvm_cpuid2 cpuid;
 360         struct kvm_cpuid_entry2 entries[100];
 361     } QEMU_PACKED cpuid_data;
 362     KVMState *s = env->kvm_state;
 363     uint32_t limit, i, j, cpuid_i;
 364     uint32_t unused;
 365     struct kvm_cpuid_entry2 *c;
 366     uint32_t signature[3];
 367     int r;
 368
 369     env->cpuid_features &= kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
 370
 371     i = env->cpuid_ext_features & CPUID_EXT_HYPERVISOR;
 372     j = env->cpuid_ext_features & CPUID_EXT_TSC_DEADLINE_TIMER;
 373     env->cpuid_ext_features &= kvm_arch_get_supported_cpuid(s, 1, 0, R_ECX);
 374     env->cpuid_ext_features |= i;
 375     if (j && kvm_irqchip_in_kernel() &&
 376         kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
 377         env->cpuid_ext_features |= CPUID_EXT_TSC_DEADLINE_TIMER;
 378     }
 379
 380     env->cpuid_ext2_features &= kvm_arch_get_supported_cpuid(s, 0x80000001,
 381                                                              0, R_EDX);
 382     env->cpuid_ext3_features &= kvm_arch_get_supported_cpuid(s, 0x80000001,
 383                                                              0, R_ECX);
 384     env->cpuid_svm_features  &= kvm_arch_get_supported_cpuid(s, 0x8000000A,
 385                                                              0, R_EDX);
 386
 387     cpuid_i = 0;
 388
 389     /* Paravirtualization CPUIDs */
 390     c = &cpuid_data.entries[cpuid_i++];
 391     memset(c, 0, sizeof(*c));
 392     c->function = KVM_CPUID_SIGNATURE;
 393     if (!hyperv_enabled()) {
 394         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 395         c->eax = 0;
 396     } else {
 397         memcpy(signature, "Microsoft Hv", 12);
 398         c->eax = HYPERV_CPUID_MIN;
 399     }
 400     c->ebx = signature[0];
 401     c->ecx = signature[1];
 402     c->edx = signature[2];
 403
 404     c = &cpuid_data.entries[cpuid_i++];
 405     memset(c, 0, sizeof(*c));
 406     c->function = KVM_CPUID_FEATURES;
 407     c->eax = env->cpuid_kvm_features &
 408         kvm_arch_get_supported_cpuid(s, KVM_CPUID_FEATURES, 0, R_EAX);
 409
 410     if (hyperv_enabled()) {
 411         memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
 412         c->eax = signature[0];
 413
 414         c = &cpuid_data.entries[cpuid_i++];
 415         memset(c, 0, sizeof(*c));
 416         c->function = HYPERV_CPUID_VERSION;
 417         c->eax = 0x00001bbc;
 418         c->ebx = 0x00060001;
 419
 420         c = &cpuid_data.entries[cpuid_i++];
 421         memset(c, 0, sizeof(*c));
 422         c->function = HYPERV_CPUID_FEATURES;
 423         if (hyperv_relaxed_timing_enabled()) {
 424             c->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
 425         }
 426         if (hyperv_vapic_recommended()) {
 427             c->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
 428             c->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
 429         }
 430
 431         c = &cpuid_data.entries[cpuid_i++];
 432         memset(c, 0, sizeof(*c));
 433         c->function = HYPERV_CPUID_ENLIGHTMENT_INFO;
 434         if (hyperv_relaxed_timing_enabled()) {
 435             c->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
 436         }
 437         if (hyperv_vapic_recommended()) {
 438             c->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
 439         }
 440         c->ebx = hyperv_get_spinlock_retries();
 441
 442         c = &cpuid_data.entries[cpuid_i++];
 443         memset(c, 0, sizeof(*c));
 444         c->function = HYPERV_CPUID_IMPLEMENT_LIMITS;
 445         c->eax = 0x40;
 446         c->ebx = 0x40;
 447
 448         c = &cpuid_data.entries[cpuid_i++];
 449         memset(c, 0, sizeof(*c));
 450         c->function = KVM_CPUID_SIGNATURE_NEXT;
 451         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 452         c->eax = 0;
 453         c->ebx = signature[0];
 454         c->ecx = signature[1];
 455         c->edx = signature[2];
 456     }
 457
 458     has_msr_async_pf_en = c->eax & (1 << KVM_FEATURE_ASYNC_PF);
 459
 460     has_msr_pv_eoi_en = c->eax & (1 << KVM_FEATURE_PV_EOI);
 461
 462     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 463
 464     for (i = 0; i <= limit; i++) {
 465         c = &cpuid_data.entries[cpuid_i++];
 466
 467         switch (i) {
 468         case 2: {
 469             /* Keep reading function 2 till all the input is received */
 470             int times;
 471
 472             c->function = i;
 473             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 474                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 475             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 476             times = c->eax & 0xff;
 477
 478             for (j = 1; j < times; ++j) {
 479                 c = &cpuid_data.entries[cpuid_i++];
 480                 c->function = i;
 481                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 482                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 483             }
 484             break;
 485         }
 486         case 4:
 487         case 0xb:
 488         case 0xd:
 489             for (j = 0; ; j++) {
 490                 if (i == 0xd && j == 64) {
 491                     break;
 492                 }
 493                 c->function = i;
 494                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 495                 c->index = j;
 496                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
 497
 498                 if (i == 4 && c->eax == 0) {
 499                     break;
 500                 }
 501                 if (i == 0xb && !(c->ecx & 0xff00)) {
 502                     break;
 503                 }
 504                 if (i == 0xd && c->eax == 0) {
 505                     continue;
 506                 }
 507                 c = &cpuid_data.entries[cpuid_i++];
 508             }
 509             break;
 510         default:
 511             c->function = i;
 512             c->flags = 0;
 513             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 514             break;
 515         }
 516     }
 517     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
 518
 519     for (i = 0x80000000; i <= limit; i++) {
 520         c = &cpuid_data.entries[cpuid_i++];
 521
 522         c->function = i;
 523         c->flags = 0;
 524         cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 525     }
 526
 527     /* Call Centaur's CPUID instructions they are supported. */
 528     if (env->cpuid_xlevel2 > 0) {
 529         env->cpuid_ext4_features &=
 530             kvm_arch_get_supported_cpuid(s, 0xC0000001, 0, R_EDX);
 531         cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
 532
 533         for (i = 0xC0000000; i <= limit; i++) {
 534             c = &cpuid_data.entries[cpuid_i++];
 535
 536             c->function = i;
 537             c->flags = 0;
 538             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 539         }
 540     }
 541
 542     cpuid_data.cpuid.nent = cpuid_i;
 543
 544     if (((env->cpuid_version >> 8)&0xF) >= 6
 545         && (env->cpuid_features&(CPUID_MCE|CPUID_MCA)) == (CPUID_MCE|CPUID_MCA)
 546         && kvm_check_extension(env->kvm_state, KVM_CAP_MCE) > 0) {
 547         uint64_t mcg_cap;
 548         int banks;
 549         int ret;
 550
 551         ret = kvm_get_mce_cap_supported(env->kvm_state, &mcg_cap, &banks);
 552         if (ret < 0) {
 553             fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
 554             return ret;
 555         }
 556
 557         if (banks > MCE_BANKS_DEF) {
 558             banks = MCE_BANKS_DEF;
 559         }
 560         mcg_cap &= MCE_CAP_DEF;
 561         mcg_cap |= banks;
 562         ret = kvm_vcpu_ioctl(env, KVM_X86_SETUP_MCE, &mcg_cap);
 563         if (ret < 0) {
 564             fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
 565             return ret;
 566         }
 567
 568         env->mcg_cap = mcg_cap;
 569     }
 570
 571     qemu_add_vm_change_state_handler(cpu_update_state, env);
 572
 573     cpuid_data.cpuid.padding = 0;
 574     r = kvm_vcpu_ioctl(env, KVM_SET_CPUID2, &cpuid_data);
 575     if (r) {
 576         return r;
 577     }
 578
 579     r = kvm_check_extension(env->kvm_state, KVM_CAP_TSC_CONTROL);
 580     if (r && env->tsc_khz) {
 581         r = kvm_vcpu_ioctl(env, KVM_SET_TSC_KHZ, env->tsc_khz);
 582         if (r < 0) {
 583             fprintf(stderr, "KVM_SET_TSC_KHZ failed\n");
 584             return r;
 585         }
 586     }
 587
 588     if (kvm_has_xsave()) {
 589         env->kvm_xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
 590     }
 591
 592     return 0;
 593 }
 594
 595 void kvm_arch_reset_vcpu(CPUX86State *env)
 596 {
 597     X86CPU *cpu = x86_env_get_cpu(env);
 598
 599     env->exception_injected = -1;
 600     env->interrupt_injected = -1;
 601     env->xcr0 = 1;
 602     if (kvm_irqchip_in_kernel()) {
 603         env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
 604                                           KVM_MP_STATE_UNINITIALIZED;
 605     } else {
 606         env->mp_state = KVM_MP_STATE_RUNNABLE;
 607     }
 608 }
 609
 610 static int kvm_get_supported_msrs(KVMState *s)
 611 {
 612     static int kvm_supported_msrs;
 613     int ret = 0;
 614
 615     /* first time */
 616     if (kvm_supported_msrs == 0) {
 617         struct kvm_msr_list msr_list, *kvm_msr_list;
 618
 619         kvm_supported_msrs = -1;
 620
 621         /* Obtain MSR list from KVM.  These are the MSRs that we must
 622          * save/restore */
 623         msr_list.nmsrs = 0;
 624         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
 625         if (ret < 0 && ret != -E2BIG) {
 626             return ret;
 627         }
 628         /* Old kernel modules had a bug and could write beyond the provided
 629            memory. Allocate at least a safe amount of 1K. */
 630         kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
 631                                               msr_list.nmsrs *
 632                                               sizeof(msr_list.indices[0])));
 633
 634         kvm_msr_list->nmsrs = msr_list.nmsrs;
 635         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
 636         if (ret >= 0) {
 637             int i;
 638
 639             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
 640                 if (kvm_msr_list->indices[i] == MSR_STAR) {
 641                     has_msr_star = true;
 642                     continue;
 643                 }
 644                 if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA) {
 645                     has_msr_hsave_pa = true;
 646                     continue;
 647                 }
 648                 if (kvm_msr_list->indices[i] == MSR_IA32_TSCDEADLINE) {
 649                     has_msr_tsc_deadline = true;
 650                     continue;
 651                 }
 652                 if (kvm_msr_list->indices[i] == MSR_IA32_MISC_ENABLE) {
 653                     has_msr_misc_enable = true;
 654                     continue;
 655                 }
 656             }
 657         }
 658
 659         g_free(kvm_msr_list);
 660     }
 661
 662     return ret;
 663 }
 664
 665 int kvm_arch_init(KVMState *s)
 666 {
 667     QemuOptsList *list = qemu_find_opts("machine");
 668     uint64_t identity_base = 0xfffbc000;
 669     uint64_t shadow_mem;
 670     int ret;
 671     struct utsname utsname;
 672
 673     ret = kvm_get_supported_msrs(s);
 674     if (ret < 0) {
 675         return ret;
 676     }
 677
 678     uname(&utsname);
 679     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
 680
 681     /*
 682      * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
 683      * In order to use vm86 mode, an EPT identity map and a TSS  are needed.
 684      * Since these must be part of guest physical memory, we need to allocate
 685      * them, both by setting their start addresses in the kernel and by
 686      * creating a corresponding e820 entry. We need 4 pages before the BIOS.
 687      *
 688      * Older KVM versions may not support setting the identity map base. In
 689      * that case we need to stick with the default, i.e. a 256K maximum BIOS
 690      * size.
 691      */
 692     if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
 693         /* Allows up to 16M BIOSes. */
 694         identity_base = 0xfeffc000;
 695
 696         ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
 697         if (ret < 0) {
 698             return ret;
 699         }
 700     }
 701
 702     /* Set TSS base one page after EPT identity map. */
 703     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
 704     if (ret < 0) {
 705         return ret;
 706     }
 707
 708     /* Tell fw_cfg to notify the BIOS to reserve the range. */
 709     ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
 710     if (ret < 0) {
 711         fprintf(stderr, "e820_add_entry() table is full\n");
 712         return ret;
 713     }
 714     qemu_register_reset(kvm_unpoison_all, NULL);
 715
 716     if (!QTAILQ_EMPTY(&list->head)) {
 717         shadow_mem = qemu_opt_get_size(QTAILQ_FIRST(&list->head),
 718                                        "kvm_shadow_mem", -1);
 719         if (shadow_mem != -1) {
 720             shadow_mem /= 4096;
 721             ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
 722             if (ret < 0) {
 723                 return ret;
 724             }
 725         }
 726     }
 727     return 0;
 728 }
 729
 730 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 731 {
 732     lhs->selector = rhs->selector;
 733     lhs->base = rhs->base;
 734     lhs->limit = rhs->limit;
 735     lhs->type = 3;
 736     lhs->present = 1;
 737     lhs->dpl = 3;
 738     lhs->db = 0;
 739     lhs->s = 1;
 740     lhs->l = 0;
 741     lhs->g = 0;
 742     lhs->avl = 0;
 743     lhs->unusable = 0;
 744 }
 745
 746 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 747 {
 748     unsigned flags = rhs->flags;
 749     lhs->selector = rhs->selector;
 750     lhs->base = rhs->base;
 751     lhs->limit = rhs->limit;
 752     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 753     lhs->present = (flags & DESC_P_MASK) != 0;
 754     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
 755     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 756     lhs->s = (flags & DESC_S_MASK) != 0;
 757     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 758     lhs->g = (flags & DESC_G_MASK) != 0;
 759     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 760     lhs->unusable = 0;
 761     lhs->padding = 0;
 762 }
 763
 764 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 765 {
 766     lhs->selector = rhs->selector;
 767     lhs->base = rhs->base;
 768     lhs->limit = rhs->limit;
 769     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
 770                  (rhs->present * DESC_P_MASK) |
 771                  (rhs->dpl << DESC_DPL_SHIFT) |
 772                  (rhs->db << DESC_B_SHIFT) |
 773                  (rhs->s * DESC_S_MASK) |
 774                  (rhs->l << DESC_L_SHIFT) |
 775                  (rhs->g * DESC_G_MASK) |
 776                  (rhs->avl * DESC_AVL_MASK);
 777 }
 778
 779 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
 780 {
 781     if (set) {
 782         *kvm_reg = *qemu_reg;
 783     } else {
 784         *qemu_reg = *kvm_reg;
 785     }
 786 }
 787
 788 static int kvm_getput_regs(CPUX86State *env, int set)
 789 {
 790     struct kvm_regs regs;
 791     int ret = 0;
 792
 793     if (!set) {
 794         ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 795         if (ret < 0) {
 796             return ret;
 797         }
 798     }
 799
 800     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
 801     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
 802     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
 803     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
 804     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
 805     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
 806     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
 807     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
 808 #ifdef TARGET_X86_64
 809     kvm_getput_reg(&regs.r8, &env->regs[8], set);
 810     kvm_getput_reg(&regs.r9, &env->regs[9], set);
 811     kvm_getput_reg(&regs.r10, &env->regs[10], set);
 812     kvm_getput_reg(&regs.r11, &env->regs[11], set);
 813     kvm_getput_reg(&regs.r12, &env->regs[12], set);
 814     kvm_getput_reg(&regs.r13, &env->regs[13], set);
 815     kvm_getput_reg(&regs.r14, &env->regs[14], set);
 816     kvm_getput_reg(&regs.r15, &env->regs[15], set);
 817 #endif
 818
 819     kvm_getput_reg(&regs.rflags, &env->eflags, set);
 820     kvm_getput_reg(&regs.rip, &env->eip, set);
 821
 822     if (set) {
 823         ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 824     }
 825
 826     return ret;
 827 }
 828
 829 static int kvm_put_fpu(CPUX86State *env)
 830 {
 831     struct kvm_fpu fpu;
 832     int i;
 833
 834     memset(&fpu, 0, sizeof fpu);
 835     fpu.fsw = env->fpus & ~(7 << 11);
 836     fpu.fsw |= (env->fpstt & 7) << 11;
 837     fpu.fcw = env->fpuc;
 838     fpu.last_opcode = env->fpop;
 839     fpu.last_ip = env->fpip;
 840     fpu.last_dp = env->fpdp;
 841     for (i = 0; i < 8; ++i) {
 842         fpu.ftwx |= (!env->fptags[i]) << i;
 843     }
 844     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 845     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 846     fpu.mxcsr = env->mxcsr;
 847
 848     return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
 849 }
 850
 851 #define XSAVE_FCW_FSW     0
 852 #define XSAVE_FTW_FOP     1
 853 #define XSAVE_CWD_RIP     2
 854 #define XSAVE_CWD_RDP     4
 855 #define XSAVE_MXCSR       6
 856 #define XSAVE_ST_SPACE    8
 857 #define XSAVE_XMM_SPACE   40
 858 #define XSAVE_XSTATE_BV   128
 859 #define XSAVE_YMMH_SPACE  144
 860
 861 static int kvm_put_xsave(CPUX86State *env)
 862 {
 863     struct kvm_xsave* xsave = env->kvm_xsave_buf;
 864     uint16_t cwd, swd, twd;
 865     int i, r;
 866
 867     if (!kvm_has_xsave()) {
 868         return kvm_put_fpu(env);
 869     }
 870
 871     memset(xsave, 0, sizeof(struct kvm_xsave));
 872     twd = 0;
 873     swd = env->fpus & ~(7 << 11);
 874     swd |= (env->fpstt & 7) << 11;
 875     cwd = env->fpuc;
 876     for (i = 0; i < 8; ++i) {
 877         twd |= (!env->fptags[i]) << i;
 878     }
 879     xsave->region[XSAVE_FCW_FSW] = (uint32_t)(swd << 16) + cwd;
 880     xsave->region[XSAVE_FTW_FOP] = (uint32_t)(env->fpop << 16) + twd;
 881     memcpy(&xsave->region[XSAVE_CWD_RIP], &env->fpip, sizeof(env->fpip));
 882     memcpy(&xsave->region[XSAVE_CWD_RDP], &env->fpdp, sizeof(env->fpdp));
 883     memcpy(&xsave->region[XSAVE_ST_SPACE], env->fpregs,
 884             sizeof env->fpregs);
 885     memcpy(&xsave->region[XSAVE_XMM_SPACE], env->xmm_regs,
 886             sizeof env->xmm_regs);
 887     xsave->region[XSAVE_MXCSR] = env->mxcsr;
 888     *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV] = env->xstate_bv;
 889     memcpy(&xsave->region[XSAVE_YMMH_SPACE], env->ymmh_regs,
 890             sizeof env->ymmh_regs);
 891     r = kvm_vcpu_ioctl(env, KVM_SET_XSAVE, xsave);
 892     return r;
 893 }
 894
 895 static int kvm_put_xcrs(CPUX86State *env)
 896 {
 897     struct kvm_xcrs xcrs;
 898
 899     if (!kvm_has_xcrs()) {
 900         return 0;
 901     }
 902
 903     xcrs.nr_xcrs = 1;
 904     xcrs.flags = 0;
 905     xcrs.xcrs[0].xcr = 0;
 906     xcrs.xcrs[0].value = env->xcr0;
 907     return kvm_vcpu_ioctl(env, KVM_SET_XCRS, &xcrs);
 908 }
 909
 910 static int kvm_put_sregs(CPUX86State *env)
 911 {
 912     struct kvm_sregs sregs;
 913
 914     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
 915     if (env->interrupt_injected >= 0) {
 916         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
 917                 (uint64_t)1 << (env->interrupt_injected % 64);
 918     }
 919
 920     if ((env->eflags & VM_MASK)) {
 921         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 922         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 923         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 924         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 925         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 926         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 927     } else {
 928         set_seg(&sregs.cs, &env->segs[R_CS]);
 929         set_seg(&sregs.ds, &env->segs[R_DS]);
 930         set_seg(&sregs.es, &env->segs[R_ES]);
 931         set_seg(&sregs.fs, &env->segs[R_FS]);
 932         set_seg(&sregs.gs, &env->segs[R_GS]);
 933         set_seg(&sregs.ss, &env->segs[R_SS]);
 934     }
 935
 936     set_seg(&sregs.tr, &env->tr);
 937     set_seg(&sregs.ldt, &env->ldt);
 938
 939     sregs.idt.limit = env->idt.limit;
 940     sregs.idt.base = env->idt.base;
 941     memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
 942     sregs.gdt.limit = env->gdt.limit;
 943     sregs.gdt.base = env->gdt.base;
 944     memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
 945
 946     sregs.cr0 = env->cr[0];
 947     sregs.cr2 = env->cr[2];
 948     sregs.cr3 = env->cr[3];
 949     sregs.cr4 = env->cr[4];
 950
 951     sregs.cr8 = cpu_get_apic_tpr(env->apic_state);
 952     sregs.apic_base = cpu_get_apic_base(env->apic_state);
 953
 954     sregs.efer = env->efer;
 955
 956     return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
 957 }
 958
 959 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
 960                               uint32_t index, uint64_t value)
 961 {
 962     entry->index = index;
 963     entry->data = value;
 964 }
 965
 966 static int kvm_put_msrs(CPUX86State *env, int level)
 967 {
 968     struct {
 969         struct kvm_msrs info;
 970         struct kvm_msr_entry entries[100];
 971     } msr_data;
 972     struct kvm_msr_entry *msrs = msr_data.entries;
 973     int n = 0;
 974
 975     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
 976     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 977     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 978     kvm_msr_entry_set(&msrs[n++], MSR_PAT, env->pat);
 979     if (has_msr_star) {
 980         kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
 981     }
 982     if (has_msr_hsave_pa) {
 983         kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
 984     }
 985     if (has_msr_tsc_deadline) {
 986         kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSCDEADLINE, env->tsc_deadline);
 987     }
 988     if (has_msr_misc_enable) {
 989         kvm_msr_entry_set(&msrs[n++], MSR_IA32_MISC_ENABLE,
 990                           env->msr_ia32_misc_enable);
 991     }
 992 #ifdef TARGET_X86_64
 993     if (lm_capable_kernel) {
 994         kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
 995         kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
 996         kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
 997         kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
 998     }
 999 #endif
1000     if (level == KVM_PUT_FULL_STATE) {
1001         /*
1002          * KVM is yet unable to synchronize TSC values of multiple VCPUs on
1003          * writeback. Until this is fixed, we only write the offset to SMP
1004          * guests after migration, desynchronizing the VCPUs, but avoiding
1005          * huge jump-backs that would occur without any writeback at all.
1006          */
1007         if (smp_cpus == 1 || env->tsc != 0) {
1008             kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
1009         }
1010     }
1011     /*
1012      * The following paravirtual MSRs have side effects on the guest or are
1013      * too heavy for normal writeback. Limit them to reset or full state
1014      * updates.
1015      */
1016     if (level >= KVM_PUT_RESET_STATE) {
1017         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
1018                           env->system_time_msr);
1019         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
1020         if (has_msr_async_pf_en) {
1021             kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN,
1022                               env->async_pf_en_msr);
1023         }
1024         if (has_msr_pv_eoi_en) {
1025             kvm_msr_entry_set(&msrs[n++], MSR_KVM_PV_EOI_EN,
1026                               env->pv_eoi_en_msr);
1027         }
1028         if (hyperv_hypercall_available()) {
1029             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_GUEST_OS_ID, 0);
1030             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_HYPERCALL, 0);
1031         }
1032         if (hyperv_vapic_recommended()) {
1033             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_APIC_ASSIST_PAGE, 0);
1034         }
1035     }
1036     if (env->mcg_cap) {
1037         int i;
1038
1039         kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
1040         kvm_msr_entry_set(&msrs[n++], MSR_MCG_CTL, env->mcg_ctl);
1041         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1042             kvm_msr_entry_set(&msrs[n++], MSR_MC0_CTL + i, env->mce_banks[i]);
1043         }
1044     }
1045
1046     msr_data.info.nmsrs = n;
1047
1048     return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
1049
1050 }
1051
1052
1053 static int kvm_get_fpu(CPUX86State *env)
1054 {
1055     struct kvm_fpu fpu;
1056     int i, ret;
1057
1058     ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
1059     if (ret < 0) {
1060         return ret;
1061     }
1062
1063     env->fpstt = (fpu.fsw >> 11) & 7;
1064     env->fpus = fpu.fsw;
1065     env->fpuc = fpu.fcw;
1066     env->fpop = fpu.last_opcode;
1067     env->fpip = fpu.last_ip;
1068     env->fpdp = fpu.last_dp;
1069     for (i = 0; i < 8; ++i) {
1070         env->fptags[i] = !((fpu.ftwx >> i) & 1);
1071     }
1072     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
1073     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
1074     env->mxcsr = fpu.mxcsr;
1075
1076     return 0;
1077 }
1078
1079 static int kvm_get_xsave(CPUX86State *env)
1080 {
1081     struct kvm_xsave* xsave = env->kvm_xsave_buf;
1082     int ret, i;
1083     uint16_t cwd, swd, twd;
1084
1085     if (!kvm_has_xsave()) {
1086         return kvm_get_fpu(env);
1087     }
1088
1089     ret = kvm_vcpu_ioctl(env, KVM_GET_XSAVE, xsave);
1090     if (ret < 0) {
1091         return ret;
1092     }
1093
1094     cwd = (uint16_t)xsave->region[XSAVE_FCW_FSW];
1095     swd = (uint16_t)(xsave->region[XSAVE_FCW_FSW] >> 16);
1096     twd = (uint16_t)xsave->region[XSAVE_FTW_FOP];
1097     env->fpop = (uint16_t)(xsave->region[XSAVE_FTW_FOP] >> 16);
1098     env->fpstt = (swd >> 11) & 7;
1099     env->fpus = swd;
1100     env->fpuc = cwd;
1101     for (i = 0; i < 8; ++i) {
1102         env->fptags[i] = !((twd >> i) & 1);
1103     }
1104     memcpy(&env->fpip, &xsave->region[XSAVE_CWD_RIP], sizeof(env->fpip));
1105     memcpy(&env->fpdp, &xsave->region[XSAVE_CWD_RDP], sizeof(env->fpdp));
1106     env->mxcsr = xsave->region[XSAVE_MXCSR];
1107     memcpy(env->fpregs, &xsave->region[XSAVE_ST_SPACE],
1108             sizeof env->fpregs);
1109     memcpy(env->xmm_regs, &xsave->region[XSAVE_XMM_SPACE],
1110             sizeof env->xmm_regs);
1111     env->xstate_bv = *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV];
1112     memcpy(env->ymmh_regs, &xsave->region[XSAVE_YMMH_SPACE],
1113             sizeof env->ymmh_regs);
1114     return 0;
1115 }
1116
1117 static int kvm_get_xcrs(CPUX86State *env)
1118 {
1119     int i, ret;
1120     struct kvm_xcrs xcrs;
1121
1122     if (!kvm_has_xcrs()) {
1123         return 0;
1124     }
1125
1126     ret = kvm_vcpu_ioctl(env, KVM_GET_XCRS, &xcrs);
1127     if (ret < 0) {
1128         return ret;
1129     }
1130
1131     for (i = 0; i < xcrs.nr_xcrs; i++) {
1132         /* Only support xcr0 now */
1133         if (xcrs.xcrs[0].xcr == 0) {
1134             env->xcr0 = xcrs.xcrs[0].value;
1135             break;
1136         }
1137     }
1138     return 0;
1139 }
1140
1141 static int kvm_get_sregs(CPUX86State *env)
1142 {
1143     struct kvm_sregs sregs;
1144     uint32_t hflags;
1145     int bit, i, ret;
1146
1147     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
1148     if (ret < 0) {
1149         return ret;
1150     }
1151
1152     /* There can only be one pending IRQ set in the bitmap at a time, so try
1153        to find it and save its number instead (-1 for none). */
1154     env->interrupt_injected = -1;
1155     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
1156         if (sregs.interrupt_bitmap[i]) {
1157             bit = ctz64(sregs.interrupt_bitmap[i]);
1158             env->interrupt_injected = i * 64 + bit;
1159             break;
1160         }
1161     }
1162
1163     get_seg(&env->segs[R_CS], &sregs.cs);
1164     get_seg(&env->segs[R_DS], &sregs.ds);
1165     get_seg(&env->segs[R_ES], &sregs.es);
1166     get_seg(&env->segs[R_FS], &sregs.fs);
1167     get_seg(&env->segs[R_GS], &sregs.gs);
1168     get_seg(&env->segs[R_SS], &sregs.ss);
1169
1170     get_seg(&env->tr, &sregs.tr);
1171     get_seg(&env->ldt, &sregs.ldt);
1172
1173     env->idt.limit = sregs.idt.limit;
1174     env->idt.base = sregs.idt.base;
1175     env->gdt.limit = sregs.gdt.limit;
1176     env->gdt.base = sregs.gdt.base;
1177
1178     env->cr[0] = sregs.cr0;
1179     env->cr[2] = sregs.cr2;
1180     env->cr[3] = sregs.cr3;
1181     env->cr[4] = sregs.cr4;
1182
1183     env->efer = sregs.efer;
1184
1185     /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
1186
1187 #define HFLAG_COPY_MASK \
1188     ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1189        HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1190        HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1191        HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1192
1193     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
1194     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
1195     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
1196                 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
1197     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
1198     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
1199                 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
1200
1201     if (env->efer & MSR_EFER_LMA) {
1202         hflags |= HF_LMA_MASK;
1203     }
1204
1205     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
1206         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
1207     } else {
1208         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
1209                     (DESC_B_SHIFT - HF_CS32_SHIFT);
1210         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
1211                     (DESC_B_SHIFT - HF_SS32_SHIFT);
1212         if (!(env->cr[0] & CR0_PE_MASK) || (env->eflags & VM_MASK) ||
1213             !(hflags & HF_CS32_MASK)) {
1214             hflags |= HF_ADDSEG_MASK;
1215         } else {
1216             hflags |= ((env->segs[R_DS].base | env->segs[R_ES].base |
1217                         env->segs[R_SS].base) != 0) << HF_ADDSEG_SHIFT;
1218         }
1219     }
1220     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
1221
1222     return 0;
1223 }
1224
1225 static int kvm_get_msrs(CPUX86State *env)
1226 {
1227     struct {
1228         struct kvm_msrs info;
1229         struct kvm_msr_entry entries[100];
1230     } msr_data;
1231     struct kvm_msr_entry *msrs = msr_data.entries;
1232     int ret, i, n;
1233
1234     n = 0;
1235     msrs[n++].index = MSR_IA32_SYSENTER_CS;
1236     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
1237     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
1238     msrs[n++].index = MSR_PAT;
1239     if (has_msr_star) {
1240         msrs[n++].index = MSR_STAR;
1241     }
1242     if (has_msr_hsave_pa) {
1243         msrs[n++].index = MSR_VM_HSAVE_PA;
1244     }
1245     if (has_msr_tsc_deadline) {
1246         msrs[n++].index = MSR_IA32_TSCDEADLINE;
1247     }
1248     if (has_msr_misc_enable) {
1249         msrs[n++].index = MSR_IA32_MISC_ENABLE;
1250     }
1251
1252     if (!env->tsc_valid) {
1253         msrs[n++].index = MSR_IA32_TSC;
1254         env->tsc_valid = !runstate_is_running();
1255     }
1256
1257 #ifdef TARGET_X86_64
1258     if (lm_capable_kernel) {
1259         msrs[n++].index = MSR_CSTAR;
1260         msrs[n++].index = MSR_KERNELGSBASE;
1261         msrs[n++].index = MSR_FMASK;
1262         msrs[n++].index = MSR_LSTAR;
1263     }
1264 #endif
1265     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
1266     msrs[n++].index = MSR_KVM_WALL_CLOCK;
1267     if (has_msr_async_pf_en) {
1268         msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
1269     }
1270     if (has_msr_pv_eoi_en) {
1271         msrs[n++].index = MSR_KVM_PV_EOI_EN;
1272     }
1273
1274     if (env->mcg_cap) {
1275         msrs[n++].index = MSR_MCG_STATUS;
1276         msrs[n++].index = MSR_MCG_CTL;
1277         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1278             msrs[n++].index = MSR_MC0_CTL + i;
1279         }
1280     }
1281
1282     msr_data.info.nmsrs = n;
1283     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
1284     if (ret < 0) {
1285         return ret;
1286     }
1287
1288     for (i = 0; i < ret; i++) {
1289         switch (msrs[i].index) {
1290         case MSR_IA32_SYSENTER_CS:
1291             env->sysenter_cs = msrs[i].data;
1292             break;
1293         case MSR_IA32_SYSENTER_ESP:
1294             env->sysenter_esp = msrs[i].data;
1295             break;
1296         case MSR_IA32_SYSENTER_EIP:
1297             env->sysenter_eip = msrs[i].data;
1298             break;
1299         case MSR_PAT:
1300             env->pat = msrs[i].data;
1301             break;
1302         case MSR_STAR:
1303             env->star = msrs[i].data;
1304             break;
1305 #ifdef TARGET_X86_64
1306         case MSR_CSTAR:
1307             env->cstar = msrs[i].data;
1308             break;
1309         case MSR_KERNELGSBASE:
1310             env->kernelgsbase = msrs[i].data;
1311             break;
1312         case MSR_FMASK:
1313             env->fmask = msrs[i].data;
1314             break;
1315         case MSR_LSTAR:
1316             env->lstar = msrs[i].data;
1317             break;
1318 #endif
1319         case MSR_IA32_TSC:
1320             env->tsc = msrs[i].data;
1321             break;
1322         case MSR_IA32_TSCDEADLINE:
1323             env->tsc_deadline = msrs[i].data;
1324             break;
1325         case MSR_VM_HSAVE_PA:
1326             env->vm_hsave = msrs[i].data;
1327             break;
1328         case MSR_KVM_SYSTEM_TIME:
1329             env->system_time_msr = msrs[i].data;
1330             break;
1331         case MSR_KVM_WALL_CLOCK:
1332             env->wall_clock_msr = msrs[i].data;
1333             break;
1334         case MSR_MCG_STATUS:
1335             env->mcg_status = msrs[i].data;
1336             break;
1337         case MSR_MCG_CTL:
1338             env->mcg_ctl = msrs[i].data;
1339             break;
1340         case MSR_IA32_MISC_ENABLE:
1341             env->msr_ia32_misc_enable = msrs[i].data;
1342             break;
1343         default:
1344             if (msrs[i].index >= MSR_MC0_CTL &&
1345                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
1346                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
1347             }
1348             break;
1349         case MSR_KVM_ASYNC_PF_EN:
1350             env->async_pf_en_msr = msrs[i].data;
1351             break;
1352         case MSR_KVM_PV_EOI_EN:
1353             env->pv_eoi_en_msr = msrs[i].data;
1354             break;
1355         }
1356     }
1357
1358     return 0;
1359 }
1360
1361 static int kvm_put_mp_state(CPUX86State *env)
1362 {
1363     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
1364
1365     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
1366 }
1367
1368 static int kvm_get_mp_state(CPUX86State *env)
1369 {
1370     struct kvm_mp_state mp_state;
1371     int ret;
1372
1373     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
1374     if (ret < 0) {
1375         return ret;
1376     }
1377     env->mp_state = mp_state.mp_state;
1378     if (kvm_irqchip_in_kernel()) {
1379         env->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
1380     }
1381     return 0;
1382 }
1383
1384 static int kvm_get_apic(CPUX86State *env)
1385 {
1386     DeviceState *apic = env->apic_state;
1387     struct kvm_lapic_state kapic;
1388     int ret;
1389
1390     if (apic && kvm_irqchip_in_kernel()) {
1391         ret = kvm_vcpu_ioctl(env, KVM_GET_LAPIC, &kapic);
1392         if (ret < 0) {
1393             return ret;
1394         }
1395
1396         kvm_get_apic_state(apic, &kapic);
1397     }
1398     return 0;
1399 }
1400
1401 static int kvm_put_apic(CPUX86State *env)
1402 {
1403     DeviceState *apic = env->apic_state;
1404     struct kvm_lapic_state kapic;
1405
1406     if (apic && kvm_irqchip_in_kernel()) {
1407         kvm_put_apic_state(apic, &kapic);
1408
1409         return kvm_vcpu_ioctl(env, KVM_SET_LAPIC, &kapic);
1410     }
1411     return 0;
1412 }
1413
1414 static int kvm_put_vcpu_events(CPUX86State *env, int level)
1415 {
1416     struct kvm_vcpu_events events;
1417
1418     if (!kvm_has_vcpu_events()) {
1419         return 0;
1420     }
1421
1422     events.exception.injected = (env->exception_injected >= 0);
1423     events.exception.nr = env->exception_injected;
1424     events.exception.has_error_code = env->has_error_code;
1425     events.exception.error_code = env->error_code;
1426     events.exception.pad = 0;
1427
1428     events.interrupt.injected = (env->interrupt_injected >= 0);
1429     events.interrupt.nr = env->interrupt_injected;
1430     events.interrupt.soft = env->soft_interrupt;
1431
1432     events.nmi.injected = env->nmi_injected;
1433     events.nmi.pending = env->nmi_pending;
1434     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
1435     events.nmi.pad = 0;
1436
1437     events.sipi_vector = env->sipi_vector;
1438
1439     events.flags = 0;
1440     if (level >= KVM_PUT_RESET_STATE) {
1441         events.flags |=
1442             KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;
1443     }
1444
1445     return kvm_vcpu_ioctl(env, KVM_SET_VCPU_EVENTS, &events);
1446 }
1447
1448 static int kvm_get_vcpu_events(CPUX86State *env)
1449 {
1450     struct kvm_vcpu_events events;
1451     int ret;
1452
1453     if (!kvm_has_vcpu_events()) {
1454         return 0;
1455     }
1456
1457     ret = kvm_vcpu_ioctl(env, KVM_GET_VCPU_EVENTS, &events);
1458     if (ret < 0) {
1459        return ret;
1460     }
1461     env->exception_injected =
1462        events.exception.injected ? events.exception.nr : -1;
1463     env->has_error_code = events.exception.has_error_code;
1464     env->error_code = events.exception.error_code;
1465
1466     env->interrupt_injected =
1467         events.interrupt.injected ? events.interrupt.nr : -1;
1468     env->soft_interrupt = events.interrupt.soft;
1469
1470     env->nmi_injected = events.nmi.injected;
1471     env->nmi_pending = events.nmi.pending;
1472     if (events.nmi.masked) {
1473         env->hflags2 |= HF2_NMI_MASK;
1474     } else {
1475         env->hflags2 &= ~HF2_NMI_MASK;
1476     }
1477
1478     env->sipi_vector = events.sipi_vector;
1479
1480     return 0;
1481 }
1482
1483 static int kvm_guest_debug_workarounds(CPUX86State *env)
1484 {
1485     int ret = 0;
1486     unsigned long reinject_trap = 0;
1487
1488     if (!kvm_has_vcpu_events()) {
1489         if (env->exception_injected == 1) {
1490             reinject_trap = KVM_GUESTDBG_INJECT_DB;
1491         } else if (env->exception_injected == 3) {
1492             reinject_trap = KVM_GUESTDBG_INJECT_BP;
1493         }
1494         env->exception_injected = -1;
1495     }
1496
1497     /*
1498      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1499      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1500      * by updating the debug state once again if single-stepping is on.
1501      * Another reason to call kvm_update_guest_debug here is a pending debug
1502      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1503      * reinject them via SET_GUEST_DEBUG.
1504      */
1505     if (reinject_trap ||
1506         (!kvm_has_robust_singlestep() && env->singlestep_enabled)) {
1507         ret = kvm_update_guest_debug(env, reinject_trap);
1508     }
1509     return ret;
1510 }
1511
1512 static int kvm_put_debugregs(CPUX86State *env)
1513 {
1514     struct kvm_debugregs dbgregs;
1515     int i;
1516
1517     if (!kvm_has_debugregs()) {
1518         return 0;
1519     }
1520
1521     for (i = 0; i < 4; i++) {
1522         dbgregs.db[i] = env->dr[i];
1523     }
1524     dbgregs.dr6 = env->dr[6];
1525     dbgregs.dr7 = env->dr[7];
1526     dbgregs.flags = 0;
1527
1528     return kvm_vcpu_ioctl(env, KVM_SET_DEBUGREGS, &dbgregs);
1529 }
1530
1531 static int kvm_get_debugregs(CPUX86State *env)
1532 {
1533     struct kvm_debugregs dbgregs;
1534     int i, ret;
1535
1536     if (!kvm_has_debugregs()) {
1537         return 0;
1538     }
1539
1540     ret = kvm_vcpu_ioctl(env, KVM_GET_DEBUGREGS, &dbgregs);
1541     if (ret < 0) {
1542         return ret;
1543     }
1544     for (i = 0; i < 4; i++) {
1545         env->dr[i] = dbgregs.db[i];
1546     }
1547     env->dr[4] = env->dr[6] = dbgregs.dr6;
1548     env->dr[5] = env->dr[7] = dbgregs.dr7;
1549
1550     return 0;
1551 }
1552
1553 int kvm_arch_put_registers(CPUX86State *env, int level)
1554 {
1555     int ret;
1556
1557     assert(cpu_is_stopped(env) || qemu_cpu_is_self(env));
1558
1559     ret = kvm_getput_regs(env, 1);
1560     if (ret < 0) {
1561         return ret;
1562     }
1563     ret = kvm_put_xsave(env);
1564     if (ret < 0) {
1565         return ret;
1566     }
1567     ret = kvm_put_xcrs(env);
1568     if (ret < 0) {
1569         return ret;
1570     }
1571     ret = kvm_put_sregs(env);
1572     if (ret < 0) {
1573         return ret;
1574     }
1575     /* must be before kvm_put_msrs */
1576     ret = kvm_inject_mce_oldstyle(env);
1577     if (ret < 0) {
1578         return ret;
1579     }
1580     ret = kvm_put_msrs(env, level);
1581     if (ret < 0) {
1582         return ret;
1583     }
1584     if (level >= KVM_PUT_RESET_STATE) {
1585         ret = kvm_put_mp_state(env);
1586         if (ret < 0) {
1587             return ret;
1588         }
1589         ret = kvm_put_apic(env);
1590         if (ret < 0) {
1591             return ret;
1592         }
1593     }
1594     ret = kvm_put_vcpu_events(env, level);
1595     if (ret < 0) {
1596         return ret;
1597     }
1598     ret = kvm_put_debugregs(env);
1599     if (ret < 0) {
1600         return ret;
1601     }
1602     /* must be last */
1603     ret = kvm_guest_debug_workarounds(env);
1604     if (ret < 0) {
1605         return ret;
1606     }
1607     return 0;
1608 }
1609
1610 int kvm_arch_get_registers(CPUX86State *env)
1611 {
1612     int ret;
1613
1614     assert(cpu_is_stopped(env) || qemu_cpu_is_self(env));
1615
1616     ret = kvm_getput_regs(env, 0);
1617     if (ret < 0) {
1618         return ret;
1619     }
1620     ret = kvm_get_xsave(env);
1621     if (ret < 0) {
1622         return ret;
1623     }
1624     ret = kvm_get_xcrs(env);
1625     if (ret < 0) {
1626         return ret;
1627     }
1628     ret = kvm_get_sregs(env);
1629     if (ret < 0) {
1630         return ret;
1631     }
1632     ret = kvm_get_msrs(env);
1633     if (ret < 0) {
1634         return ret;
1635     }
1636     ret = kvm_get_mp_state(env);
1637     if (ret < 0) {
1638         return ret;
1639     }
1640     ret = kvm_get_apic(env);
1641     if (ret < 0) {
1642         return ret;
1643     }
1644     ret = kvm_get_vcpu_events(env);
1645     if (ret < 0) {
1646         return ret;
1647     }
1648     ret = kvm_get_debugregs(env);
1649     if (ret < 0) {
1650         return ret;
1651     }
1652     return 0;
1653 }
1654
1655 void kvm_arch_pre_run(CPUX86State *env, struct kvm_run *run)
1656 {
1657     int ret;
1658
1659     /* Inject NMI */
1660     if (env->interrupt_request & CPU_INTERRUPT_NMI) {
1661         env->interrupt_request &= ~CPU_INTERRUPT_NMI;
1662         DPRINTF("injected NMI\n");
1663         ret = kvm_vcpu_ioctl(env, KVM_NMI);
1664         if (ret < 0) {
1665             fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
1666                     strerror(-ret));
1667         }
1668     }
1669
1670     if (!kvm_irqchip_in_kernel()) {
1671         /* Force the VCPU out of its inner loop to process any INIT requests
1672          * or pending TPR access reports. */
1673         if (env->interrupt_request &
1674             (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
1675             env->exit_request = 1;
1676         }
1677
1678         /* Try to inject an interrupt if the guest can accept it */
1679         if (run->ready_for_interrupt_injection &&
1680             (env->interrupt_request & CPU_INTERRUPT_HARD) &&
1681             (env->eflags & IF_MASK)) {
1682             int irq;
1683
1684             env->interrupt_request &= ~CPU_INTERRUPT_HARD;
1685             irq = cpu_get_pic_interrupt(env);
1686             if (irq >= 0) {
1687                 struct kvm_interrupt intr;
1688
1689                 intr.irq = irq;
1690                 DPRINTF("injected interrupt %d\n", irq);
1691                 ret = kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
1692                 if (ret < 0) {
1693                     fprintf(stderr,
1694                             "KVM: injection failed, interrupt lost (%s)\n",
1695                             strerror(-ret));
1696                 }
1697             }
1698         }
1699
1700         /* If we have an interrupt but the guest is not ready to receive an
1701          * interrupt, request an interrupt window exit.  This will
1702          * cause a return to userspace as soon as the guest is ready to
1703          * receive interrupts. */
1704         if ((env->interrupt_request & CPU_INTERRUPT_HARD)) {
1705             run->request_interrupt_window = 1;
1706         } else {
1707             run->request_interrupt_window = 0;
1708         }
1709
1710         DPRINTF("setting tpr\n");
1711         run->cr8 = cpu_get_apic_tpr(env->apic_state);
1712     }
1713 }
1714
1715 void kvm_arch_post_run(CPUX86State *env, struct kvm_run *run)
1716 {
1717     if (run->if_flag) {
1718         env->eflags |= IF_MASK;
1719     } else {
1720         env->eflags &= ~IF_MASK;
1721     }
1722     cpu_set_apic_tpr(env->apic_state, run->cr8);
1723     cpu_set_apic_base(env->apic_state, run->apic_base);
1724 }
1725
1726 int kvm_arch_process_async_events(CPUX86State *env)
1727 {
1728     X86CPU *cpu = x86_env_get_cpu(env);
1729
1730     if (env->interrupt_request & CPU_INTERRUPT_MCE) {
1731         /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
1732         assert(env->mcg_cap);
1733
1734         env->interrupt_request &= ~CPU_INTERRUPT_MCE;
1735
1736         kvm_cpu_synchronize_state(env);
1737
1738         if (env->exception_injected == EXCP08_DBLE) {
1739             /* this means triple fault */
1740             qemu_system_reset_request();
1741             env->exit_request = 1;
1742             return 0;
1743         }
1744         env->exception_injected = EXCP12_MCHK;
1745         env->has_error_code = 0;
1746
1747         env->halted = 0;
1748         if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
1749             env->mp_state = KVM_MP_STATE_RUNNABLE;
1750         }
1751     }
1752
1753     if (kvm_irqchip_in_kernel()) {
1754         return 0;
1755     }
1756
1757     if (env->interrupt_request & CPU_INTERRUPT_POLL) {
1758         env->interrupt_request &= ~CPU_INTERRUPT_POLL;
1759         apic_poll_irq(env->apic_state);
1760     }
1761     if (((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1762          (env->eflags & IF_MASK)) ||
1763         (env->interrupt_request & CPU_INTERRUPT_NMI)) {
1764         env->halted = 0;
1765     }
1766     if (env->interrupt_request & CPU_INTERRUPT_INIT) {
1767         kvm_cpu_synchronize_state(env);
1768         do_cpu_init(cpu);
1769     }
1770     if (env->interrupt_request & CPU_INTERRUPT_SIPI) {
1771         kvm_cpu_synchronize_state(env);
1772         do_cpu_sipi(cpu);
1773     }
1774     if (env->interrupt_request & CPU_INTERRUPT_TPR) {
1775         env->interrupt_request &= ~CPU_INTERRUPT_TPR;
1776         kvm_cpu_synchronize_state(env);
1777         apic_handle_tpr_access_report(env->apic_state, env->eip,
1778                                       env->tpr_access_type);
1779     }
1780
1781     return env->halted;
1782 }
1783
1784 static int kvm_handle_halt(CPUX86State *env)
1785 {
1786     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1787           (env->eflags & IF_MASK)) &&
1788         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
1789         env->halted = 1;
1790         return EXCP_HLT;
1791     }
1792
1793     return 0;
1794 }
1795
1796 static int kvm_handle_tpr_access(CPUX86State *env)
1797 {
1798     struct kvm_run *run = env->kvm_run;
1799
1800     apic_handle_tpr_access_report(env->apic_state, run->tpr_access.rip,
1801                                   run->tpr_access.is_write ? TPR_ACCESS_WRITE
1802                                                            : TPR_ACCESS_READ);
1803     return 1;
1804 }
1805
1806 int kvm_arch_insert_sw_breakpoint(CPUX86State *env, struct kvm_sw_breakpoint *bp)
1807 {
1808     static const uint8_t int3 = 0xcc;
1809
1810     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
1811         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1)) {
1812         return -EINVAL;
1813     }
1814     return 0;
1815 }
1816
1817 int kvm_arch_remove_sw_breakpoint(CPUX86State *env, struct kvm_sw_breakpoint *bp)
1818 {
1819     uint8_t int3;
1820
1821     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
1822         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
1823         return -EINVAL;
1824     }
1825     return 0;
1826 }
1827
1828 static struct {
1829     target_ulong addr;
1830     int len;
1831     int type;
1832 } hw_breakpoint[4];
1833
1834 static int nb_hw_breakpoint;
1835
1836 static int find_hw_breakpoint(target_ulong addr, int len, int type)
1837 {
1838     int n;
1839
1840     for (n = 0; n < nb_hw_breakpoint; n++) {
1841         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
1842             (hw_breakpoint[n].len == len || len == -1)) {
1843             return n;
1844         }
1845     }
1846     return -1;
1847 }
1848
1849 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1850                                   target_ulong len, int type)
1851 {
1852     switch (type) {
1853     case GDB_BREAKPOINT_HW:
1854         len = 1;
1855         break;
1856     case GDB_WATCHPOINT_WRITE:
1857     case GDB_WATCHPOINT_ACCESS:
1858         switch (len) {
1859         case 1:
1860             break;
1861         case 2:
1862         case 4:
1863         case 8:
1864             if (addr & (len - 1)) {
1865                 return -EINVAL;
1866             }
1867             break;
1868         default:
1869             return -EINVAL;
1870         }
1871         break;
1872     default:
1873         return -ENOSYS;
1874     }
1875
1876     if (nb_hw_breakpoint == 4) {
1877         return -ENOBUFS;
1878     }
1879     if (find_hw_breakpoint(addr, len, type) >= 0) {
1880         return -EEXIST;
1881     }
1882     hw_breakpoint[nb_hw_breakpoint].addr = addr;
1883     hw_breakpoint[nb_hw_breakpoint].len = len;
1884     hw_breakpoint[nb_hw_breakpoint].type = type;
1885     nb_hw_breakpoint++;
1886
1887     return 0;
1888 }
1889
1890 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1891                                   target_ulong len, int type)
1892 {
1893     int n;
1894
1895     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
1896     if (n < 0) {
1897         return -ENOENT;
1898     }
1899     nb_hw_breakpoint--;
1900     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
1901
1902     return 0;
1903 }
1904
1905 void kvm_arch_remove_all_hw_breakpoints(void)
1906 {
1907     nb_hw_breakpoint = 0;
1908 }
1909
1910 static CPUWatchpoint hw_watchpoint;
1911
1912 static int kvm_handle_debug(struct kvm_debug_exit_arch *arch_info)
1913 {
1914     int ret = 0;
1915     int n;
1916
1917     if (arch_info->exception == 1) {
1918         if (arch_info->dr6 & (1 << 14)) {
1919             if (cpu_single_env->singlestep_enabled) {
1920                 ret = EXCP_DEBUG;
1921             }
1922         } else {
1923             for (n = 0; n < 4; n++) {
1924                 if (arch_info->dr6 & (1 << n)) {
1925                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
1926                     case 0x0:
1927                         ret = EXCP_DEBUG;
1928                         break;
1929                     case 0x1:
1930                         ret = EXCP_DEBUG;
1931                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1932                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1933                         hw_watchpoint.flags = BP_MEM_WRITE;
1934                         break;
1935                     case 0x3:
1936                         ret = EXCP_DEBUG;
1937                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1938                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1939                         hw_watchpoint.flags = BP_MEM_ACCESS;
1940                         break;
1941                     }
1942                 }
1943             }
1944         }
1945     } else if (kvm_find_sw_breakpoint(cpu_single_env, arch_info->pc)) {
1946         ret = EXCP_DEBUG;
1947     }
1948     if (ret == 0) {
1949         cpu_synchronize_state(cpu_single_env);
1950         assert(cpu_single_env->exception_injected == -1);
1951
1952         /* pass to guest */
1953         cpu_single_env->exception_injected = arch_info->exception;
1954         cpu_single_env->has_error_code = 0;
1955     }
1956
1957     return ret;
1958 }
1959
1960 void kvm_arch_update_guest_debug(CPUX86State *env, struct kvm_guest_debug *dbg)
1961 {
1962     const uint8_t type_code[] = {
1963         [GDB_BREAKPOINT_HW] = 0x0,
1964         [GDB_WATCHPOINT_WRITE] = 0x1,
1965         [GDB_WATCHPOINT_ACCESS] = 0x3
1966     };
1967     const uint8_t len_code[] = {
1968         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1969     };
1970     int n;
1971
1972     if (kvm_sw_breakpoints_active(env)) {
1973         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1974     }
1975     if (nb_hw_breakpoint > 0) {
1976         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1977         dbg->arch.debugreg[7] = 0x0600;
1978         for (n = 0; n < nb_hw_breakpoint; n++) {
1979             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
1980             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
1981                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
1982                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
1983         }
1984     }
1985 }
1986
1987 static bool host_supports_vmx(void)
1988 {
1989     uint32_t ecx, unused;
1990
1991     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
1992     return ecx & CPUID_EXT_VMX;
1993 }
1994
1995 #define VMX_INVALID_GUEST_STATE 0x80000021
1996
1997 int kvm_arch_handle_exit(CPUX86State *env, struct kvm_run *run)
1998 {
1999     uint64_t code;
2000     int ret;
2001
2002     switch (run->exit_reason) {
2003     case KVM_EXIT_HLT:
2004         DPRINTF("handle_hlt\n");
2005         ret = kvm_handle_halt(env);
2006         break;
2007     case KVM_EXIT_SET_TPR:
2008         ret = 0;
2009         break;
2010     case KVM_EXIT_TPR_ACCESS:
2011         ret = kvm_handle_tpr_access(env);
2012         break;
2013     case KVM_EXIT_FAIL_ENTRY:
2014         code = run->fail_entry.hardware_entry_failure_reason;
2015         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
2016                 code);
2017         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
2018             fprintf(stderr,
2019                     "\nIf you're running a guest on an Intel machine without "
2020                         "unrestricted mode\n"
2021                     "support, the failure can be most likely due to the guest "
2022                         "entering an invalid\n"
2023                     "state for Intel VT. For example, the guest maybe running "
2024                         "in big real mode\n"
2025                     "which is not supported on less recent Intel processors."
2026                         "\n\n");
2027         }
2028         ret = -1;
2029         break;
2030     case KVM_EXIT_EXCEPTION:
2031         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
2032                 run->ex.exception, run->ex.error_code);
2033         ret = -1;
2034         break;
2035     case KVM_EXIT_DEBUG:
2036         DPRINTF("kvm_exit_debug\n");
2037         ret = kvm_handle_debug(&run->debug.arch);
2038         break;
2039     default:
2040         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
2041         ret = -1;
2042         break;
2043     }
2044
2045     return ret;
2046 }
2047
2048 bool kvm_arch_stop_on_emulation_error(CPUX86State *env)
2049 {
2050     kvm_cpu_synchronize_state(env);
2051     return !(env->cr[0] & CR0_PE_MASK) ||
2052            ((env->segs[R_CS].selector  & 3) != 3);
2053 }
2054
2055 void kvm_arch_init_irq_routing(KVMState *s)
2056 {
2057     if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2058         /* If kernel can't do irq routing, interrupt source
2059          * override 0->2 cannot be set up as required by HPET.
2060          * So we have to disable it.
2061          */
2062         no_hpet = 1;
2063     }
2064     /* We know at this point that we're using the in-kernel
2065      * irqchip, so we can use irqfds, and on x86 we know
2066      * we can use msi via irqfd and GSI routing.
2067      */
2068     kvm_irqfds_allowed = true;
2069     kvm_msi_via_irqfd_allowed = true;
2070     kvm_gsi_routing_allowed = true;
2071 }
2072
2073 /* Classic KVM device assignment interface. Will remain x86 only. */
2074 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
2075                           uint32_t flags, uint32_t *dev_id)
2076 {
2077     struct kvm_assigned_pci_dev dev_data = {
2078         .segnr = dev_addr->domain,
2079         .busnr = dev_addr->bus,
2080         .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
2081         .flags = flags,
2082     };
2083     int ret;
2084
2085     dev_data.assigned_dev_id =
2086         (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
2087
2088     ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
2089     if (ret < 0) {
2090         return ret;
2091     }
2092
2093     *dev_id = dev_data.assigned_dev_id;
2094
2095     return 0;
2096 }
2097
2098 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
2099 {
2100     struct kvm_assigned_pci_dev dev_data = {
2101         .assigned_dev_id = dev_id,
2102     };
2103
2104     return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
2105 }
2106
2107 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
2108                                    uint32_t irq_type, uint32_t guest_irq)
2109 {
2110     struct kvm_assigned_irq assigned_irq = {
2111         .assigned_dev_id = dev_id,
2112         .guest_irq = guest_irq,
2113         .flags = irq_type,
2114     };
2115
2116     if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
2117         return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
2118     } else {
2119         return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
2120     }
2121 }
2122
2123 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
2124                            uint32_t guest_irq)
2125 {
2126     uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
2127         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
2128
2129     return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
2130 }
2131
2132 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
2133 {
2134     struct kvm_assigned_pci_dev dev_data = {
2135         .assigned_dev_id = dev_id,
2136         .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
2137     };
2138
2139     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
2140 }
2141
2142 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
2143                                      uint32_t type)
2144 {
2145     struct kvm_assigned_irq assigned_irq = {
2146         .assigned_dev_id = dev_id,
2147         .flags = type,
2148     };
2149
2150     return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
2151 }
2152
2153 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
2154 {
2155     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
2156         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
2157 }
2158
2159 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
2160 {
2161     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
2162                                               KVM_DEV_IRQ_GUEST_MSI, virq);
2163 }
2164
2165 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
2166 {
2167     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
2168                                                 KVM_DEV_IRQ_HOST_MSI);
2169 }
2170
2171 bool kvm_device_msix_supported(KVMState *s)
2172 {
2173     /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
2174      * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
2175     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
2176 }
2177
2178 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
2179                                  uint32_t nr_vectors)
2180 {
2181     struct kvm_assigned_msix_nr msix_nr = {
2182         .assigned_dev_id = dev_id,
2183         .entry_nr = nr_vectors,
2184     };
2185
2186     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
2187 }
2188
2189 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
2190                                int virq)
2191 {
2192     struct kvm_assigned_msix_entry msix_entry = {
2193         .assigned_dev_id = dev_id,
2194         .gsi = virq,
2195         .entry = vector,
2196     };
2197
2198     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
2199 }
2200
2201 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
2202 {
2203     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
2204                                               KVM_DEV_IRQ_GUEST_MSIX, 0);
2205 }
2206
2207 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
2208 {
2209     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
2210                                                 KVM_DEV_IRQ_HOST_MSIX);
2211 }