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
  22 #include "qemu-common.h"
  23 #include "sysemu.h"
  24 #include "kvm.h"
  25 #include "cpu.h"
  26 #include "gdbstub.h"
  27 #include "host-utils.h"
  28 #include "hw/pc.h"
  29 #include "hw/apic.h"
  30 #include "ioport.h"
  31 #include "kvm_x86.h"
  32
  33 #ifdef CONFIG_KVM_PARA
  34 #include <linux/kvm_para.h>
  35 #endif
  36 //
  37 //#define DEBUG_KVM
  38
  39 #ifdef DEBUG_KVM
  40 #define DPRINTF(fmt, ...) \
  41     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  42 #else
  43 #define DPRINTF(fmt, ...) \
  44     do { } while (0)
  45 #endif
  46
  47 #define MSR_KVM_WALL_CLOCK  0x11
  48 #define MSR_KVM_SYSTEM_TIME 0x12
  49
  50 #ifndef BUS_MCEERR_AR
  51 #define BUS_MCEERR_AR 4
  52 #endif
  53 #ifndef BUS_MCEERR_AO
  54 #define BUS_MCEERR_AO 5
  55 #endif
  56
  57 static int lm_capable_kernel;
  58
  59 #ifdef KVM_CAP_EXT_CPUID
  60
  61 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
  62 {
  63     struct kvm_cpuid2 *cpuid;
  64     int r, size;
  65
  66     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
  67     cpuid = (struct kvm_cpuid2 *)qemu_mallocz(size);
  68     cpuid->nent = max;
  69     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
  70     if (r == 0 && cpuid->nent >= max) {
  71         r = -E2BIG;
  72     }
  73     if (r < 0) {
  74         if (r == -E2BIG) {
  75             qemu_free(cpuid);
  76             return NULL;
  77         } else {
  78             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
  79                     strerror(-r));
  80             exit(1);
  81         }
  82     }
  83     return cpuid;
  84 }
  85
  86 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function,
  87                                       uint32_t index, int reg)
  88 {
  89     struct kvm_cpuid2 *cpuid;
  90     int i, max;
  91     uint32_t ret = 0;
  92     uint32_t cpuid_1_edx;
  93
  94     if (!kvm_check_extension(env->kvm_state, KVM_CAP_EXT_CPUID)) {
  95         return -1U;
  96     }
  97
  98     max = 1;
  99     while ((cpuid = try_get_cpuid(env->kvm_state, max)) == NULL) {
 100         max *= 2;
 101     }
 102
 103     for (i = 0; i < cpuid->nent; ++i) {
 104         if (cpuid->entries[i].function == function &&
 105             cpuid->entries[i].index == index) {
 106             switch (reg) {
 107             case R_EAX:
 108                 ret = cpuid->entries[i].eax;
 109                 break;
 110             case R_EBX:
 111                 ret = cpuid->entries[i].ebx;
 112                 break;
 113             case R_ECX:
 114                 ret = cpuid->entries[i].ecx;
 115                 break;
 116             case R_EDX:
 117                 ret = cpuid->entries[i].edx;
 118                 switch (function) {
 119                 case 1:
 120                     /* KVM before 2.6.30 misreports the following features */
 121                     ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
 122                     break;
 123                 case 0x80000001:
 124                     /* On Intel, kvm returns cpuid according to the Intel spec,
 125                      * so add missing bits according to the AMD spec:
 126                      */
 127                     cpuid_1_edx = kvm_arch_get_supported_cpuid(env, 1, 0, R_EDX);
 128                     ret |= cpuid_1_edx & 0x183f7ff;
 129                     break;
 130                 }
 131                 break;
 132             }
 133         }
 134     }
 135
 136     qemu_free(cpuid);
 137
 138     return ret;
 139 }
 140
 141 #else
 142
 143 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function,
 144                                       uint32_t index, int reg)
 145 {
 146     return -1U;
 147 }
 148
 149 #endif
 150
 151 #ifdef CONFIG_KVM_PARA
 152 struct kvm_para_features {
 153         int cap;
 154         int feature;
 155 } para_features[] = {
 156 #ifdef KVM_CAP_CLOCKSOURCE
 157         { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 158 #endif
 159 #ifdef KVM_CAP_NOP_IO_DELAY
 160         { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 161 #endif
 162 #ifdef KVM_CAP_PV_MMU
 163         { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 164 #endif
 165 #ifdef KVM_CAP_ASYNC_PF
 166         { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
 167 #endif
 168         { -1, -1 }
 169 };
 170
 171 static int get_para_features(CPUState *env)
 172 {
 173         int i, features = 0;
 174
 175         for (i = 0; i < ARRAY_SIZE(para_features) - 1; i++) {
 176                 if (kvm_check_extension(env->kvm_state, para_features[i].cap))
 177                         features |= (1 << para_features[i].feature);
 178         }
 179
 180         return features;
 181 }
 182 #endif
 183
 184 #ifdef KVM_CAP_MCE
 185 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
 186                                      int *max_banks)
 187 {
 188     int r;
 189
 190     r = kvm_check_extension(s, KVM_CAP_MCE);
 191     if (r > 0) {
 192         *max_banks = r;
 193         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
 194     }
 195     return -ENOSYS;
 196 }
 197
 198 static int kvm_setup_mce(CPUState *env, uint64_t *mcg_cap)
 199 {
 200     return kvm_vcpu_ioctl(env, KVM_X86_SETUP_MCE, mcg_cap);
 201 }
 202
 203 static int kvm_set_mce(CPUState *env, struct kvm_x86_mce *m)
 204 {
 205     return kvm_vcpu_ioctl(env, KVM_X86_SET_MCE, m);
 206 }
 207
 208 static int kvm_get_msr(CPUState *env, struct kvm_msr_entry *msrs, int n)
 209 {
 210     struct kvm_msrs *kmsrs = qemu_malloc(sizeof *kmsrs + n * sizeof *msrs);
 211     int r;
 212
 213     kmsrs->nmsrs = n;
 214     memcpy(kmsrs->entries, msrs, n * sizeof *msrs);
 215     r = kvm_vcpu_ioctl(env, KVM_GET_MSRS, kmsrs);
 216     memcpy(msrs, kmsrs->entries, n * sizeof *msrs);
 217     free(kmsrs);
 218     return r;
 219 }
 220
 221 /* FIXME: kill this and kvm_get_msr, use env->mcg_status instead */
 222 static int kvm_mce_in_exception(CPUState *env)
 223 {
 224     struct kvm_msr_entry msr_mcg_status = {
 225         .index = MSR_MCG_STATUS,
 226     };
 227     int r;
 228
 229     r = kvm_get_msr(env, &msr_mcg_status, 1);
 230     if (r == -1 || r == 0) {
 231         return -1;
 232     }
 233     return !!(msr_mcg_status.data & MCG_STATUS_MCIP);
 234 }
 235
 236 struct kvm_x86_mce_data
 237 {
 238     CPUState *env;
 239     struct kvm_x86_mce *mce;
 240     int abort_on_error;
 241 };
 242
 243 static void kvm_do_inject_x86_mce(void *_data)
 244 {
 245     struct kvm_x86_mce_data *data = _data;
 246     int r;
 247
 248     /* If there is an MCE exception being processed, ignore this SRAO MCE */
 249     if ((data->env->mcg_cap & MCG_SER_P) &&
 250         !(data->mce->status & MCI_STATUS_AR)) {
 251         r = kvm_mce_in_exception(data->env);
 252         if (r == -1) {
 253             fprintf(stderr, "Failed to get MCE status\n");
 254         } else if (r) {
 255             return;
 256         }
 257     }
 258
 259     r = kvm_set_mce(data->env, data->mce);
 260     if (r < 0) {
 261         perror("kvm_set_mce FAILED");
 262         if (data->abort_on_error) {
 263             abort();
 264         }
 265     }
 266 }
 267 #endif
 268
 269 void kvm_inject_x86_mce(CPUState *cenv, int bank, uint64_t status,
 270                         uint64_t mcg_status, uint64_t addr, uint64_t misc,
 271                         int abort_on_error)
 272 {
 273 #ifdef KVM_CAP_MCE
 274     struct kvm_x86_mce mce = {
 275         .bank = bank,
 276         .status = status,
 277         .mcg_status = mcg_status,
 278         .addr = addr,
 279         .misc = misc,
 280     };
 281     struct kvm_x86_mce_data data = {
 282             .env = cenv,
 283             .mce = &mce,
 284     };
 285
 286     if (!cenv->mcg_cap) {
 287         fprintf(stderr, "MCE support is not enabled!\n");
 288         return;
 289     }
 290
 291     run_on_cpu(cenv, kvm_do_inject_x86_mce, &data);
 292 #else
 293     if (abort_on_error)
 294         abort();
 295 #endif
 296 }
 297
 298 int kvm_arch_init_vcpu(CPUState *env)
 299 {
 300     struct {
 301         struct kvm_cpuid2 cpuid;
 302         struct kvm_cpuid_entry2 entries[100];
 303     } __attribute__((packed)) cpuid_data;
 304     uint32_t limit, i, j, cpuid_i;
 305     uint32_t unused;
 306     struct kvm_cpuid_entry2 *c;
 307 #ifdef KVM_CPUID_SIGNATURE
 308     uint32_t signature[3];
 309 #endif
 310
 311     env->mp_state = KVM_MP_STATE_RUNNABLE;
 312
 313     env->cpuid_features &= kvm_arch_get_supported_cpuid(env, 1, 0, R_EDX);
 314
 315     i = env->cpuid_ext_features & CPUID_EXT_HYPERVISOR;
 316     env->cpuid_ext_features &= kvm_arch_get_supported_cpuid(env, 1, 0, R_ECX);
 317     env->cpuid_ext_features |= i;
 318
 319     env->cpuid_ext2_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
 320                                                              0, R_EDX);
 321     env->cpuid_ext3_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
 322                                                              0, R_ECX);
 323     env->cpuid_svm_features  &= kvm_arch_get_supported_cpuid(env, 0x8000000A,
 324                                                              0, R_EDX);
 325
 326
 327     cpuid_i = 0;
 328
 329 #ifdef CONFIG_KVM_PARA
 330     /* Paravirtualization CPUIDs */
 331     memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 332     c = &cpuid_data.entries[cpuid_i++];
 333     memset(c, 0, sizeof(*c));
 334     c->function = KVM_CPUID_SIGNATURE;
 335     c->eax = 0;
 336     c->ebx = signature[0];
 337     c->ecx = signature[1];
 338     c->edx = signature[2];
 339
 340     c = &cpuid_data.entries[cpuid_i++];
 341     memset(c, 0, sizeof(*c));
 342     c->function = KVM_CPUID_FEATURES;
 343     c->eax = env->cpuid_kvm_features & get_para_features(env);
 344 #endif
 345
 346     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 347
 348     for (i = 0; i <= limit; i++) {
 349         c = &cpuid_data.entries[cpuid_i++];
 350
 351         switch (i) {
 352         case 2: {
 353             /* Keep reading function 2 till all the input is received */
 354             int times;
 355
 356             c->function = i;
 357             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 358                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 359             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 360             times = c->eax & 0xff;
 361
 362             for (j = 1; j < times; ++j) {
 363                 c = &cpuid_data.entries[cpuid_i++];
 364                 c->function = i;
 365                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 366                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 367             }
 368             break;
 369         }
 370         case 4:
 371         case 0xb:
 372         case 0xd:
 373             for (j = 0; ; j++) {
 374                 c->function = i;
 375                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 376                 c->index = j;
 377                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
 378
 379                 if (i == 4 && c->eax == 0)
 380                     break;
 381                 if (i == 0xb && !(c->ecx & 0xff00))
 382                     break;
 383                 if (i == 0xd && c->eax == 0)
 384                     break;
 385
 386                 c = &cpuid_data.entries[cpuid_i++];
 387             }
 388             break;
 389         default:
 390             c->function = i;
 391             c->flags = 0;
 392             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 393             break;
 394         }
 395     }
 396     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
 397
 398     for (i = 0x80000000; i <= limit; i++) {
 399         c = &cpuid_data.entries[cpuid_i++];
 400
 401         c->function = i;
 402         c->flags = 0;
 403         cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 404     }
 405
 406     cpuid_data.cpuid.nent = cpuid_i;
 407
 408 #ifdef KVM_CAP_MCE
 409     if (((env->cpuid_version >> 8)&0xF) >= 6
 410         && (env->cpuid_features&(CPUID_MCE|CPUID_MCA)) == (CPUID_MCE|CPUID_MCA)
 411         && kvm_check_extension(env->kvm_state, KVM_CAP_MCE) > 0) {
 412         uint64_t mcg_cap;
 413         int banks;
 414
 415         if (kvm_get_mce_cap_supported(env->kvm_state, &mcg_cap, &banks))
 416             perror("kvm_get_mce_cap_supported FAILED");
 417         else {
 418             if (banks > MCE_BANKS_DEF)
 419                 banks = MCE_BANKS_DEF;
 420             mcg_cap &= MCE_CAP_DEF;
 421             mcg_cap |= banks;
 422             if (kvm_setup_mce(env, &mcg_cap))
 423                 perror("kvm_setup_mce FAILED");
 424             else
 425                 env->mcg_cap = mcg_cap;
 426         }
 427     }
 428 #endif
 429
 430     return kvm_vcpu_ioctl(env, KVM_SET_CPUID2, &cpuid_data);
 431 }
 432
 433 void kvm_arch_reset_vcpu(CPUState *env)
 434 {
 435     env->exception_injected = -1;
 436     env->interrupt_injected = -1;
 437     env->nmi_injected = 0;
 438     env->nmi_pending = 0;
 439     if (kvm_irqchip_in_kernel()) {
 440         env->mp_state = cpu_is_bsp(env) ? KVM_MP_STATE_RUNNABLE :
 441                                           KVM_MP_STATE_UNINITIALIZED;
 442     } else {
 443         env->mp_state = KVM_MP_STATE_RUNNABLE;
 444     }
 445 }
 446
 447 int has_msr_star;
 448 int has_msr_hsave_pa;
 449
 450 static void kvm_supported_msrs(CPUState *env)
 451 {
 452     static int kvm_supported_msrs;
 453     int ret;
 454
 455     /* first time */
 456     if (kvm_supported_msrs == 0) {
 457         struct kvm_msr_list msr_list, *kvm_msr_list;
 458
 459         kvm_supported_msrs = -1;
 460
 461         /* Obtain MSR list from KVM.  These are the MSRs that we must
 462          * save/restore */
 463         msr_list.nmsrs = 0;
 464         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
 465         if (ret < 0 && ret != -E2BIG) {
 466             return;
 467         }
 468         /* Old kernel modules had a bug and could write beyond the provided
 469            memory. Allocate at least a safe amount of 1K. */
 470         kvm_msr_list = qemu_mallocz(MAX(1024, sizeof(msr_list) +
 471                                               msr_list.nmsrs *
 472                                               sizeof(msr_list.indices[0])));
 473
 474         kvm_msr_list->nmsrs = msr_list.nmsrs;
 475         ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
 476         if (ret >= 0) {
 477             int i;
 478
 479             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
 480                 if (kvm_msr_list->indices[i] == MSR_STAR) {
 481                     has_msr_star = 1;
 482                     continue;
 483                 }
 484                 if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA) {
 485                     has_msr_hsave_pa = 1;
 486                     continue;
 487                 }
 488             }
 489         }
 490
 491         free(kvm_msr_list);
 492     }
 493
 494     return;
 495 }
 496
 497 static int kvm_has_msr_hsave_pa(CPUState *env)
 498 {
 499     kvm_supported_msrs(env);
 500     return has_msr_hsave_pa;
 501 }
 502
 503 static int kvm_has_msr_star(CPUState *env)
 504 {
 505     kvm_supported_msrs(env);
 506     return has_msr_star;
 507 }
 508
 509 static int kvm_init_identity_map_page(KVMState *s)
 510 {
 511 #ifdef KVM_CAP_SET_IDENTITY_MAP_ADDR
 512     int ret;
 513     uint64_t addr = 0xfffbc000;
 514
 515     if (!kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
 516         return 0;
 517     }
 518
 519     ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &addr);
 520     if (ret < 0) {
 521         fprintf(stderr, "kvm_set_identity_map_addr: %s\n", strerror(ret));
 522         return ret;
 523     }
 524 #endif
 525     return 0;
 526 }
 527
 528 int kvm_arch_init(KVMState *s, int smp_cpus)
 529 {
 530     int ret;
 531
 532     struct utsname utsname;
 533
 534     uname(&utsname);
 535     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
 536
 537     /* create vm86 tss.  KVM uses vm86 mode to emulate 16-bit code
 538      * directly.  In order to use vm86 mode, a TSS is needed.  Since this
 539      * must be part of guest physical memory, we need to allocate it.  Older
 540      * versions of KVM just assumed that it would be at the end of physical
 541      * memory but that doesn't work with more than 4GB of memory.  We simply
 542      * refuse to work with those older versions of KVM. */
 543     ret = kvm_check_extension(s, KVM_CAP_SET_TSS_ADDR);
 544     if (ret <= 0) {
 545         fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
 546         return ret;
 547     }
 548
 549     /* this address is 3 pages before the bios, and the bios should present
 550      * as unavaible memory.  FIXME, need to ensure the e820 map deals with
 551      * this?
 552      */
 553     /*
 554      * Tell fw_cfg to notify the BIOS to reserve the range.
 555      */
 556     if (e820_add_entry(0xfffbc000, 0x4000, E820_RESERVED) < 0) {
 557         perror("e820_add_entry() table is full");
 558         exit(1);
 559     }
 560     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, 0xfffbd000);
 561     if (ret < 0) {
 562         return ret;
 563     }
 564
 565     return kvm_init_identity_map_page(s);
 566 }
 567
 568 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 569 {
 570     lhs->selector = rhs->selector;
 571     lhs->base = rhs->base;
 572     lhs->limit = rhs->limit;
 573     lhs->type = 3;
 574     lhs->present = 1;
 575     lhs->dpl = 3;
 576     lhs->db = 0;
 577     lhs->s = 1;
 578     lhs->l = 0;
 579     lhs->g = 0;
 580     lhs->avl = 0;
 581     lhs->unusable = 0;
 582 }
 583
 584 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 585 {
 586     unsigned flags = rhs->flags;
 587     lhs->selector = rhs->selector;
 588     lhs->base = rhs->base;
 589     lhs->limit = rhs->limit;
 590     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 591     lhs->present = (flags & DESC_P_MASK) != 0;
 592     lhs->dpl = rhs->selector & 3;
 593     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 594     lhs->s = (flags & DESC_S_MASK) != 0;
 595     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 596     lhs->g = (flags & DESC_G_MASK) != 0;
 597     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 598     lhs->unusable = 0;
 599 }
 600
 601 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 602 {
 603     lhs->selector = rhs->selector;
 604     lhs->base = rhs->base;
 605     lhs->limit = rhs->limit;
 606     lhs->flags =
 607         (rhs->type << DESC_TYPE_SHIFT)
 608         | (rhs->present * DESC_P_MASK)
 609         | (rhs->dpl << DESC_DPL_SHIFT)
 610         | (rhs->db << DESC_B_SHIFT)
 611         | (rhs->s * DESC_S_MASK)
 612         | (rhs->l << DESC_L_SHIFT)
 613         | (rhs->g * DESC_G_MASK)
 614         | (rhs->avl * DESC_AVL_MASK);
 615 }
 616
 617 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
 618 {
 619     if (set)
 620         *kvm_reg = *qemu_reg;
 621     else
 622         *qemu_reg = *kvm_reg;
 623 }
 624
 625 static int kvm_getput_regs(CPUState *env, int set)
 626 {
 627     struct kvm_regs regs;
 628     int ret = 0;
 629
 630     if (!set) {
 631         ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 632         if (ret < 0)
 633             return ret;
 634     }
 635
 636     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
 637     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
 638     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
 639     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
 640     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
 641     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
 642     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
 643     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
 644 #ifdef TARGET_X86_64
 645     kvm_getput_reg(&regs.r8, &env->regs[8], set);
 646     kvm_getput_reg(&regs.r9, &env->regs[9], set);
 647     kvm_getput_reg(&regs.r10, &env->regs[10], set);
 648     kvm_getput_reg(&regs.r11, &env->regs[11], set);
 649     kvm_getput_reg(&regs.r12, &env->regs[12], set);
 650     kvm_getput_reg(&regs.r13, &env->regs[13], set);
 651     kvm_getput_reg(&regs.r14, &env->regs[14], set);
 652     kvm_getput_reg(&regs.r15, &env->regs[15], set);
 653 #endif
 654
 655     kvm_getput_reg(&regs.rflags, &env->eflags, set);
 656     kvm_getput_reg(&regs.rip, &env->eip, set);
 657
 658     if (set)
 659         ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 660
 661     return ret;
 662 }
 663
 664 static int kvm_put_fpu(CPUState *env)
 665 {
 666     struct kvm_fpu fpu;
 667     int i;
 668
 669     memset(&fpu, 0, sizeof fpu);
 670     fpu.fsw = env->fpus & ~(7 << 11);
 671     fpu.fsw |= (env->fpstt & 7) << 11;
 672     fpu.fcw = env->fpuc;
 673     for (i = 0; i < 8; ++i)
 674         fpu.ftwx |= (!env->fptags[i]) << i;
 675     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 676     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 677     fpu.mxcsr = env->mxcsr;
 678
 679     return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
 680 }
 681
 682 #ifdef KVM_CAP_XSAVE
 683 #define XSAVE_CWD_RIP     2
 684 #define XSAVE_CWD_RDP     4
 685 #define XSAVE_MXCSR       6
 686 #define XSAVE_ST_SPACE    8
 687 #define XSAVE_XMM_SPACE   40
 688 #define XSAVE_XSTATE_BV   128
 689 #define XSAVE_YMMH_SPACE  144
 690 #endif
 691
 692 static int kvm_put_xsave(CPUState *env)
 693 {
 694 #ifdef KVM_CAP_XSAVE
 695     int i, r;
 696     struct kvm_xsave* xsave;
 697     uint16_t cwd, swd, twd, fop;
 698
 699     if (!kvm_has_xsave())
 700         return kvm_put_fpu(env);
 701
 702     xsave = qemu_memalign(4096, sizeof(struct kvm_xsave));
 703     memset(xsave, 0, sizeof(struct kvm_xsave));
 704     cwd = swd = twd = fop = 0;
 705     swd = env->fpus & ~(7 << 11);
 706     swd |= (env->fpstt & 7) << 11;
 707     cwd = env->fpuc;
 708     for (i = 0; i < 8; ++i)
 709         twd |= (!env->fptags[i]) << i;
 710     xsave->region[0] = (uint32_t)(swd << 16) + cwd;
 711     xsave->region[1] = (uint32_t)(fop << 16) + twd;
 712     memcpy(&xsave->region[XSAVE_ST_SPACE], env->fpregs,
 713             sizeof env->fpregs);
 714     memcpy(&xsave->region[XSAVE_XMM_SPACE], env->xmm_regs,
 715             sizeof env->xmm_regs);
 716     xsave->region[XSAVE_MXCSR] = env->mxcsr;
 717     *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV] = env->xstate_bv;
 718     memcpy(&xsave->region[XSAVE_YMMH_SPACE], env->ymmh_regs,
 719             sizeof env->ymmh_regs);
 720     r = kvm_vcpu_ioctl(env, KVM_SET_XSAVE, xsave);
 721     qemu_free(xsave);
 722     return r;
 723 #else
 724     return kvm_put_fpu(env);
 725 #endif
 726 }
 727
 728 static int kvm_put_xcrs(CPUState *env)
 729 {
 730 #ifdef KVM_CAP_XCRS
 731     struct kvm_xcrs xcrs;
 732
 733     if (!kvm_has_xcrs())
 734         return 0;
 735
 736     xcrs.nr_xcrs = 1;
 737     xcrs.flags = 0;
 738     xcrs.xcrs[0].xcr = 0;
 739     xcrs.xcrs[0].value = env->xcr0;
 740     return kvm_vcpu_ioctl(env, KVM_SET_XCRS, &xcrs);
 741 #else
 742     return 0;
 743 #endif
 744 }
 745
 746 static int kvm_put_sregs(CPUState *env)
 747 {
 748     struct kvm_sregs sregs;
 749
 750     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
 751     if (env->interrupt_injected >= 0) {
 752         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
 753                 (uint64_t)1 << (env->interrupt_injected % 64);
 754     }
 755
 756     if ((env->eflags & VM_MASK)) {
 757             set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
 758             set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
 759             set_v8086_seg(&sregs.es, &env->segs[R_ES]);
 760             set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
 761             set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
 762             set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
 763     } else {
 764             set_seg(&sregs.cs, &env->segs[R_CS]);
 765             set_seg(&sregs.ds, &env->segs[R_DS]);
 766             set_seg(&sregs.es, &env->segs[R_ES]);
 767             set_seg(&sregs.fs, &env->segs[R_FS]);
 768             set_seg(&sregs.gs, &env->segs[R_GS]);
 769             set_seg(&sregs.ss, &env->segs[R_SS]);
 770
 771             if (env->cr[0] & CR0_PE_MASK) {
 772                 /* force ss cpl to cs cpl */
 773                 sregs.ss.selector = (sregs.ss.selector & ~3) |
 774                         (sregs.cs.selector & 3);
 775                 sregs.ss.dpl = sregs.ss.selector & 3;
 776             }
 777     }
 778
 779     set_seg(&sregs.tr, &env->tr);
 780     set_seg(&sregs.ldt, &env->ldt);
 781
 782     sregs.idt.limit = env->idt.limit;
 783     sregs.idt.base = env->idt.base;
 784     sregs.gdt.limit = env->gdt.limit;
 785     sregs.gdt.base = env->gdt.base;
 786
 787     sregs.cr0 = env->cr[0];
 788     sregs.cr2 = env->cr[2];
 789     sregs.cr3 = env->cr[3];
 790     sregs.cr4 = env->cr[4];
 791
 792     sregs.cr8 = cpu_get_apic_tpr(env->apic_state);
 793     sregs.apic_base = cpu_get_apic_base(env->apic_state);
 794
 795     sregs.efer = env->efer;
 796
 797     return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
 798 }
 799
 800 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
 801                               uint32_t index, uint64_t value)
 802 {
 803     entry->index = index;
 804     entry->data = value;
 805 }
 806
 807 static int kvm_put_msrs(CPUState *env, int level)
 808 {
 809     struct {
 810         struct kvm_msrs info;
 811         struct kvm_msr_entry entries[100];
 812     } msr_data;
 813     struct kvm_msr_entry *msrs = msr_data.entries;
 814     int n = 0;
 815
 816     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
 817     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
 818     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
 819     if (kvm_has_msr_star(env))
 820         kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
 821     if (kvm_has_msr_hsave_pa(env))
 822         kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
 823 #ifdef TARGET_X86_64
 824     if (lm_capable_kernel) {
 825         kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
 826         kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
 827         kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
 828         kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
 829     }
 830 #endif
 831     if (level == KVM_PUT_FULL_STATE) {
 832         /*
 833          * KVM is yet unable to synchronize TSC values of multiple VCPUs on
 834          * writeback. Until this is fixed, we only write the offset to SMP
 835          * guests after migration, desynchronizing the VCPUs, but avoiding
 836          * huge jump-backs that would occur without any writeback at all.
 837          */
 838         if (smp_cpus == 1 || env->tsc != 0) {
 839             kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
 840         }
 841         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
 842                           env->system_time_msr);
 843         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
 844 #ifdef KVM_CAP_ASYNC_PF
 845         kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
 846 #endif
 847     }
 848 #ifdef KVM_CAP_MCE
 849     if (env->mcg_cap) {
 850         int i;
 851         if (level == KVM_PUT_RESET_STATE)
 852             kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
 853         else if (level == KVM_PUT_FULL_STATE) {
 854             kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
 855             kvm_msr_entry_set(&msrs[n++], MSR_MCG_CTL, env->mcg_ctl);
 856             for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++)
 857                 kvm_msr_entry_set(&msrs[n++], MSR_MC0_CTL + i, env->mce_banks[i]);
 858         }
 859     }
 860 #endif
 861
 862     msr_data.info.nmsrs = n;
 863
 864     return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
 865
 866 }
 867
 868
 869 static int kvm_get_fpu(CPUState *env)
 870 {
 871     struct kvm_fpu fpu;
 872     int i, ret;
 873
 874     ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
 875     if (ret < 0)
 876         return ret;
 877
 878     env->fpstt = (fpu.fsw >> 11) & 7;
 879     env->fpus = fpu.fsw;
 880     env->fpuc = fpu.fcw;
 881     for (i = 0; i < 8; ++i)
 882         env->fptags[i] = !((fpu.ftwx >> i) & 1);
 883     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
 884     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
 885     env->mxcsr = fpu.mxcsr;
 886
 887     return 0;
 888 }
 889
 890 static int kvm_get_xsave(CPUState *env)
 891 {
 892 #ifdef KVM_CAP_XSAVE
 893     struct kvm_xsave* xsave;
 894     int ret, i;
 895     uint16_t cwd, swd, twd, fop;
 896
 897     if (!kvm_has_xsave())
 898         return kvm_get_fpu(env);
 899
 900     xsave = qemu_memalign(4096, sizeof(struct kvm_xsave));
 901     ret = kvm_vcpu_ioctl(env, KVM_GET_XSAVE, xsave);
 902     if (ret < 0) {
 903         qemu_free(xsave);
 904         return ret;
 905     }
 906
 907     cwd = (uint16_t)xsave->region[0];
 908     swd = (uint16_t)(xsave->region[0] >> 16);
 909     twd = (uint16_t)xsave->region[1];
 910     fop = (uint16_t)(xsave->region[1] >> 16);
 911     env->fpstt = (swd >> 11) & 7;
 912     env->fpus = swd;
 913     env->fpuc = cwd;
 914     for (i = 0; i < 8; ++i)
 915         env->fptags[i] = !((twd >> i) & 1);
 916     env->mxcsr = xsave->region[XSAVE_MXCSR];
 917     memcpy(env->fpregs, &xsave->region[XSAVE_ST_SPACE],
 918             sizeof env->fpregs);
 919     memcpy(env->xmm_regs, &xsave->region[XSAVE_XMM_SPACE],
 920             sizeof env->xmm_regs);
 921     env->xstate_bv = *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV];
 922     memcpy(env->ymmh_regs, &xsave->region[XSAVE_YMMH_SPACE],
 923             sizeof env->ymmh_regs);
 924     qemu_free(xsave);
 925     return 0;
 926 #else
 927     return kvm_get_fpu(env);
 928 #endif
 929 }
 930
 931 static int kvm_get_xcrs(CPUState *env)
 932 {
 933 #ifdef KVM_CAP_XCRS
 934     int i, ret;
 935     struct kvm_xcrs xcrs;
 936
 937     if (!kvm_has_xcrs())
 938         return 0;
 939
 940     ret = kvm_vcpu_ioctl(env, KVM_GET_XCRS, &xcrs);
 941     if (ret < 0)
 942         return ret;
 943
 944     for (i = 0; i < xcrs.nr_xcrs; i++)
 945         /* Only support xcr0 now */
 946         if (xcrs.xcrs[0].xcr == 0) {
 947             env->xcr0 = xcrs.xcrs[0].value;
 948             break;
 949         }
 950     return 0;
 951 #else
 952     return 0;
 953 #endif
 954 }
 955
 956 static int kvm_get_sregs(CPUState *env)
 957 {
 958     struct kvm_sregs sregs;
 959     uint32_t hflags;
 960     int bit, i, ret;
 961
 962     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 963     if (ret < 0)
 964         return ret;
 965
 966     /* There can only be one pending IRQ set in the bitmap at a time, so try
 967        to find it and save its number instead (-1 for none). */
 968     env->interrupt_injected = -1;
 969     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
 970         if (sregs.interrupt_bitmap[i]) {
 971             bit = ctz64(sregs.interrupt_bitmap[i]);
 972             env->interrupt_injected = i * 64 + bit;
 973             break;
 974         }
 975     }
 976
 977     get_seg(&env->segs[R_CS], &sregs.cs);
 978     get_seg(&env->segs[R_DS], &sregs.ds);
 979     get_seg(&env->segs[R_ES], &sregs.es);
 980     get_seg(&env->segs[R_FS], &sregs.fs);
 981     get_seg(&env->segs[R_GS], &sregs.gs);
 982     get_seg(&env->segs[R_SS], &sregs.ss);
 983
 984     get_seg(&env->tr, &sregs.tr);
 985     get_seg(&env->ldt, &sregs.ldt);
 986
 987     env->idt.limit = sregs.idt.limit;
 988     env->idt.base = sregs.idt.base;
 989     env->gdt.limit = sregs.gdt.limit;
 990     env->gdt.base = sregs.gdt.base;
 991
 992     env->cr[0] = sregs.cr0;
 993     env->cr[2] = sregs.cr2;
 994     env->cr[3] = sregs.cr3;
 995     env->cr[4] = sregs.cr4;
 996
 997     cpu_set_apic_base(env->apic_state, sregs.apic_base);
 998
 999     env->efer = sregs.efer;
1000     //cpu_set_apic_tpr(env->apic_state, sregs.cr8);
1001
1002 #define HFLAG_COPY_MASK ~( \
1003                         HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1004                         HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1005                         HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1006                         HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1007
1008
1009
1010     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
1011     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
1012     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
1013             (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
1014     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
1015     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
1016             (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
1017
1018     if (env->efer & MSR_EFER_LMA) {
1019         hflags |= HF_LMA_MASK;
1020     }
1021
1022     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
1023         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
1024     } else {
1025         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
1026                 (DESC_B_SHIFT - HF_CS32_SHIFT);
1027         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
1028                 (DESC_B_SHIFT - HF_SS32_SHIFT);
1029         if (!(env->cr[0] & CR0_PE_MASK) ||
1030                    (env->eflags & VM_MASK) ||
1031                    !(hflags & HF_CS32_MASK)) {
1032                 hflags |= HF_ADDSEG_MASK;
1033             } else {
1034                 hflags |= ((env->segs[R_DS].base |
1035                                 env->segs[R_ES].base |
1036                                 env->segs[R_SS].base) != 0) <<
1037                     HF_ADDSEG_SHIFT;
1038             }
1039     }
1040     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
1041
1042     return 0;
1043 }
1044
1045 static int kvm_get_msrs(CPUState *env)
1046 {
1047     struct {
1048         struct kvm_msrs info;
1049         struct kvm_msr_entry entries[100];
1050     } msr_data;
1051     struct kvm_msr_entry *msrs = msr_data.entries;
1052     int ret, i, n;
1053
1054     n = 0;
1055     msrs[n++].index = MSR_IA32_SYSENTER_CS;
1056     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
1057     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
1058     if (kvm_has_msr_star(env))
1059         msrs[n++].index = MSR_STAR;
1060     if (kvm_has_msr_hsave_pa(env))
1061         msrs[n++].index = MSR_VM_HSAVE_PA;
1062     msrs[n++].index = MSR_IA32_TSC;
1063 #ifdef TARGET_X86_64
1064     if (lm_capable_kernel) {
1065         msrs[n++].index = MSR_CSTAR;
1066         msrs[n++].index = MSR_KERNELGSBASE;
1067         msrs[n++].index = MSR_FMASK;
1068         msrs[n++].index = MSR_LSTAR;
1069     }
1070 #endif
1071     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
1072     msrs[n++].index = MSR_KVM_WALL_CLOCK;
1073 #ifdef KVM_CAP_ASYNC_PF
1074     msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
1075 #endif
1076
1077 #ifdef KVM_CAP_MCE
1078     if (env->mcg_cap) {
1079         msrs[n++].index = MSR_MCG_STATUS;
1080         msrs[n++].index = MSR_MCG_CTL;
1081         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++)
1082             msrs[n++].index = MSR_MC0_CTL + i;
1083     }
1084 #endif
1085
1086     msr_data.info.nmsrs = n;
1087     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
1088     if (ret < 0)
1089         return ret;
1090
1091     for (i = 0; i < ret; i++) {
1092         switch (msrs[i].index) {
1093         case MSR_IA32_SYSENTER_CS:
1094             env->sysenter_cs = msrs[i].data;
1095             break;
1096         case MSR_IA32_SYSENTER_ESP:
1097             env->sysenter_esp = msrs[i].data;
1098             break;
1099         case MSR_IA32_SYSENTER_EIP:
1100             env->sysenter_eip = msrs[i].data;
1101             break;
1102         case MSR_STAR:
1103             env->star = msrs[i].data;
1104             break;
1105 #ifdef TARGET_X86_64
1106         case MSR_CSTAR:
1107             env->cstar = msrs[i].data;
1108             break;
1109         case MSR_KERNELGSBASE:
1110             env->kernelgsbase = msrs[i].data;
1111             break;
1112         case MSR_FMASK:
1113             env->fmask = msrs[i].data;
1114             break;
1115         case MSR_LSTAR:
1116             env->lstar = msrs[i].data;
1117             break;
1118 #endif
1119         case MSR_IA32_TSC:
1120             env->tsc = msrs[i].data;
1121             break;
1122         case MSR_VM_HSAVE_PA:
1123             env->vm_hsave = msrs[i].data;
1124             break;
1125         case MSR_KVM_SYSTEM_TIME:
1126             env->system_time_msr = msrs[i].data;
1127             break;
1128         case MSR_KVM_WALL_CLOCK:
1129             env->wall_clock_msr = msrs[i].data;
1130             break;
1131 #ifdef KVM_CAP_MCE
1132         case MSR_MCG_STATUS:
1133             env->mcg_status = msrs[i].data;
1134             break;
1135         case MSR_MCG_CTL:
1136             env->mcg_ctl = msrs[i].data;
1137             break;
1138 #endif
1139         default:
1140 #ifdef KVM_CAP_MCE
1141             if (msrs[i].index >= MSR_MC0_CTL &&
1142                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
1143                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
1144             }
1145 #endif
1146             break;
1147 #ifdef KVM_CAP_ASYNC_PF
1148         case MSR_KVM_ASYNC_PF_EN:
1149             env->async_pf_en_msr = msrs[i].data;
1150             break;
1151 #endif
1152         }
1153     }
1154
1155     return 0;
1156 }
1157
1158 static int kvm_put_mp_state(CPUState *env)
1159 {
1160     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
1161
1162     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
1163 }
1164
1165 static int kvm_get_mp_state(CPUState *env)
1166 {
1167     struct kvm_mp_state mp_state;
1168     int ret;
1169
1170     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
1171     if (ret < 0) {
1172         return ret;
1173     }
1174     env->mp_state = mp_state.mp_state;
1175     return 0;
1176 }
1177
1178 static int kvm_put_vcpu_events(CPUState *env, int level)
1179 {
1180 #ifdef KVM_CAP_VCPU_EVENTS
1181     struct kvm_vcpu_events events;
1182
1183     if (!kvm_has_vcpu_events()) {
1184         return 0;
1185     }
1186
1187     events.exception.injected = (env->exception_injected >= 0);
1188     events.exception.nr = env->exception_injected;
1189     events.exception.has_error_code = env->has_error_code;
1190     events.exception.error_code = env->error_code;
1191
1192     events.interrupt.injected = (env->interrupt_injected >= 0);
1193     events.interrupt.nr = env->interrupt_injected;
1194     events.interrupt.soft = env->soft_interrupt;
1195
1196     events.nmi.injected = env->nmi_injected;
1197     events.nmi.pending = env->nmi_pending;
1198     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
1199
1200     events.sipi_vector = env->sipi_vector;
1201
1202     events.flags = 0;
1203     if (level >= KVM_PUT_RESET_STATE) {
1204         events.flags |=
1205             KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;
1206     }
1207
1208     return kvm_vcpu_ioctl(env, KVM_SET_VCPU_EVENTS, &events);
1209 #else
1210     return 0;
1211 #endif
1212 }
1213
1214 static int kvm_get_vcpu_events(CPUState *env)
1215 {
1216 #ifdef KVM_CAP_VCPU_EVENTS
1217     struct kvm_vcpu_events events;
1218     int ret;
1219
1220     if (!kvm_has_vcpu_events()) {
1221         return 0;
1222     }
1223
1224     ret = kvm_vcpu_ioctl(env, KVM_GET_VCPU_EVENTS, &events);
1225     if (ret < 0) {
1226        return ret;
1227     }
1228     env->exception_injected =
1229        events.exception.injected ? events.exception.nr : -1;
1230     env->has_error_code = events.exception.has_error_code;
1231     env->error_code = events.exception.error_code;
1232
1233     env->interrupt_injected =
1234         events.interrupt.injected ? events.interrupt.nr : -1;
1235     env->soft_interrupt = events.interrupt.soft;
1236
1237     env->nmi_injected = events.nmi.injected;
1238     env->nmi_pending = events.nmi.pending;
1239     if (events.nmi.masked) {
1240         env->hflags2 |= HF2_NMI_MASK;
1241     } else {
1242         env->hflags2 &= ~HF2_NMI_MASK;
1243     }
1244
1245     env->sipi_vector = events.sipi_vector;
1246 #endif
1247
1248     return 0;
1249 }
1250
1251 static int kvm_guest_debug_workarounds(CPUState *env)
1252 {
1253     int ret = 0;
1254 #ifdef KVM_CAP_SET_GUEST_DEBUG
1255     unsigned long reinject_trap = 0;
1256
1257     if (!kvm_has_vcpu_events()) {
1258         if (env->exception_injected == 1) {
1259             reinject_trap = KVM_GUESTDBG_INJECT_DB;
1260         } else if (env->exception_injected == 3) {
1261             reinject_trap = KVM_GUESTDBG_INJECT_BP;
1262         }
1263         env->exception_injected = -1;
1264     }
1265
1266     /*
1267      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1268      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1269      * by updating the debug state once again if single-stepping is on.
1270      * Another reason to call kvm_update_guest_debug here is a pending debug
1271      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1272      * reinject them via SET_GUEST_DEBUG.
1273      */
1274     if (reinject_trap ||
1275         (!kvm_has_robust_singlestep() && env->singlestep_enabled)) {
1276         ret = kvm_update_guest_debug(env, reinject_trap);
1277     }
1278 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1279     return ret;
1280 }
1281
1282 static int kvm_put_debugregs(CPUState *env)
1283 {
1284 #ifdef KVM_CAP_DEBUGREGS
1285     struct kvm_debugregs dbgregs;
1286     int i;
1287
1288     if (!kvm_has_debugregs()) {
1289         return 0;
1290     }
1291
1292     for (i = 0; i < 4; i++) {
1293         dbgregs.db[i] = env->dr[i];
1294     }
1295     dbgregs.dr6 = env->dr[6];
1296     dbgregs.dr7 = env->dr[7];
1297     dbgregs.flags = 0;
1298
1299     return kvm_vcpu_ioctl(env, KVM_SET_DEBUGREGS, &dbgregs);
1300 #else
1301     return 0;
1302 #endif
1303 }
1304
1305 static int kvm_get_debugregs(CPUState *env)
1306 {
1307 #ifdef KVM_CAP_DEBUGREGS
1308     struct kvm_debugregs dbgregs;
1309     int i, ret;
1310
1311     if (!kvm_has_debugregs()) {
1312         return 0;
1313     }
1314
1315     ret = kvm_vcpu_ioctl(env, KVM_GET_DEBUGREGS, &dbgregs);
1316     if (ret < 0) {
1317        return ret;
1318     }
1319     for (i = 0; i < 4; i++) {
1320         env->dr[i] = dbgregs.db[i];
1321     }
1322     env->dr[4] = env->dr[6] = dbgregs.dr6;
1323     env->dr[5] = env->dr[7] = dbgregs.dr7;
1324 #endif
1325
1326     return 0;
1327 }
1328
1329 int kvm_arch_put_registers(CPUState *env, int level)
1330 {
1331     int ret;
1332
1333     assert(cpu_is_stopped(env) || qemu_cpu_self(env));
1334
1335     ret = kvm_getput_regs(env, 1);
1336     if (ret < 0)
1337         return ret;
1338
1339     ret = kvm_put_xsave(env);
1340     if (ret < 0)
1341         return ret;
1342
1343     ret = kvm_put_xcrs(env);
1344     if (ret < 0)
1345         return ret;
1346
1347     ret = kvm_put_sregs(env);
1348     if (ret < 0)
1349         return ret;
1350
1351     ret = kvm_put_msrs(env, level);
1352     if (ret < 0)
1353         return ret;
1354
1355     if (level >= KVM_PUT_RESET_STATE) {
1356         ret = kvm_put_mp_state(env);
1357         if (ret < 0)
1358             return ret;
1359     }
1360
1361     ret = kvm_put_vcpu_events(env, level);
1362     if (ret < 0)
1363         return ret;
1364
1365     /* must be last */
1366     ret = kvm_guest_debug_workarounds(env);
1367     if (ret < 0)
1368         return ret;
1369
1370     ret = kvm_put_debugregs(env);
1371     if (ret < 0)
1372         return ret;
1373
1374     return 0;
1375 }
1376
1377 int kvm_arch_get_registers(CPUState *env)
1378 {
1379     int ret;
1380
1381     assert(cpu_is_stopped(env) || qemu_cpu_self(env));
1382
1383     ret = kvm_getput_regs(env, 0);
1384     if (ret < 0)
1385         return ret;
1386
1387     ret = kvm_get_xsave(env);
1388     if (ret < 0)
1389         return ret;
1390
1391     ret = kvm_get_xcrs(env);
1392     if (ret < 0)
1393         return ret;
1394
1395     ret = kvm_get_sregs(env);
1396     if (ret < 0)
1397         return ret;
1398
1399     ret = kvm_get_msrs(env);
1400     if (ret < 0)
1401         return ret;
1402
1403     ret = kvm_get_mp_state(env);
1404     if (ret < 0)
1405         return ret;
1406
1407     ret = kvm_get_vcpu_events(env);
1408     if (ret < 0)
1409         return ret;
1410
1411     ret = kvm_get_debugregs(env);
1412     if (ret < 0)
1413         return ret;
1414
1415     return 0;
1416 }
1417
1418 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
1419 {
1420     /* Inject NMI */
1421     if (env->interrupt_request & CPU_INTERRUPT_NMI) {
1422         env->interrupt_request &= ~CPU_INTERRUPT_NMI;
1423         DPRINTF("injected NMI\n");
1424         kvm_vcpu_ioctl(env, KVM_NMI);
1425     }
1426
1427     /* Try to inject an interrupt if the guest can accept it */
1428     if (run->ready_for_interrupt_injection &&
1429         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
1430         (env->eflags & IF_MASK)) {
1431         int irq;
1432
1433         env->interrupt_request &= ~CPU_INTERRUPT_HARD;
1434         irq = cpu_get_pic_interrupt(env);
1435         if (irq >= 0) {
1436             struct kvm_interrupt intr;
1437             intr.irq = irq;
1438             /* FIXME: errors */
1439             DPRINTF("injected interrupt %d\n", irq);
1440             kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
1441         }
1442     }
1443
1444     /* If we have an interrupt but the guest is not ready to receive an
1445      * interrupt, request an interrupt window exit.  This will
1446      * cause a return to userspace as soon as the guest is ready to
1447      * receive interrupts. */
1448     if ((env->interrupt_request & CPU_INTERRUPT_HARD))
1449         run->request_interrupt_window = 1;
1450     else
1451         run->request_interrupt_window = 0;
1452
1453     DPRINTF("setting tpr\n");
1454     run->cr8 = cpu_get_apic_tpr(env->apic_state);
1455
1456     return 0;
1457 }
1458
1459 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
1460 {
1461     if (run->if_flag)
1462         env->eflags |= IF_MASK;
1463     else
1464         env->eflags &= ~IF_MASK;
1465
1466     cpu_set_apic_tpr(env->apic_state, run->cr8);
1467     cpu_set_apic_base(env->apic_state, run->apic_base);
1468
1469     return 0;
1470 }
1471
1472 int kvm_arch_process_irqchip_events(CPUState *env)
1473 {
1474     if (env->interrupt_request & CPU_INTERRUPT_INIT) {
1475         kvm_cpu_synchronize_state(env);
1476         do_cpu_init(env);
1477         env->exception_index = EXCP_HALTED;
1478     }
1479
1480     if (env->interrupt_request & CPU_INTERRUPT_SIPI) {
1481         kvm_cpu_synchronize_state(env);
1482         do_cpu_sipi(env);
1483     }
1484
1485     return env->halted;
1486 }
1487
1488 static int kvm_handle_halt(CPUState *env)
1489 {
1490     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1491           (env->eflags & IF_MASK)) &&
1492         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
1493         env->halted = 1;
1494         env->exception_index = EXCP_HLT;
1495         return 0;
1496     }
1497
1498     return 1;
1499 }
1500
1501 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
1502 {
1503     int ret = 0;
1504
1505     switch (run->exit_reason) {
1506     case KVM_EXIT_HLT:
1507         DPRINTF("handle_hlt\n");
1508         ret = kvm_handle_halt(env);
1509         break;
1510     }
1511
1512     return ret;
1513 }
1514
1515 #ifdef KVM_CAP_SET_GUEST_DEBUG
1516 int kvm_arch_insert_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1517 {
1518     static const uint8_t int3 = 0xcc;
1519
1520     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
1521         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1))
1522         return -EINVAL;
1523     return 0;
1524 }
1525
1526 int kvm_arch_remove_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1527 {
1528     uint8_t int3;
1529
1530     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
1531         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1))
1532         return -EINVAL;
1533     return 0;
1534 }
1535
1536 static struct {
1537     target_ulong addr;
1538     int len;
1539     int type;
1540 } hw_breakpoint[4];
1541
1542 static int nb_hw_breakpoint;
1543
1544 static int find_hw_breakpoint(target_ulong addr, int len, int type)
1545 {
1546     int n;
1547
1548     for (n = 0; n < nb_hw_breakpoint; n++)
1549         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
1550             (hw_breakpoint[n].len == len || len == -1))
1551             return n;
1552     return -1;
1553 }
1554
1555 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1556                                   target_ulong len, int type)
1557 {
1558     switch (type) {
1559     case GDB_BREAKPOINT_HW:
1560         len = 1;
1561         break;
1562     case GDB_WATCHPOINT_WRITE:
1563     case GDB_WATCHPOINT_ACCESS:
1564         switch (len) {
1565         case 1:
1566             break;
1567         case 2:
1568         case 4:
1569         case 8:
1570             if (addr & (len - 1))
1571                 return -EINVAL;
1572             break;
1573         default:
1574             return -EINVAL;
1575         }
1576         break;
1577     default:
1578         return -ENOSYS;
1579     }
1580
1581     if (nb_hw_breakpoint == 4)
1582         return -ENOBUFS;
1583
1584     if (find_hw_breakpoint(addr, len, type) >= 0)
1585         return -EEXIST;
1586
1587     hw_breakpoint[nb_hw_breakpoint].addr = addr;
1588     hw_breakpoint[nb_hw_breakpoint].len = len;
1589     hw_breakpoint[nb_hw_breakpoint].type = type;
1590     nb_hw_breakpoint++;
1591
1592     return 0;
1593 }
1594
1595 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1596                                   target_ulong len, int type)
1597 {
1598     int n;
1599
1600     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
1601     if (n < 0)
1602         return -ENOENT;
1603
1604     nb_hw_breakpoint--;
1605     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
1606
1607     return 0;
1608 }
1609
1610 void kvm_arch_remove_all_hw_breakpoints(void)
1611 {
1612     nb_hw_breakpoint = 0;
1613 }
1614
1615 static CPUWatchpoint hw_watchpoint;
1616
1617 int kvm_arch_debug(struct kvm_debug_exit_arch *arch_info)
1618 {
1619     int handle = 0;
1620     int n;
1621
1622     if (arch_info->exception == 1) {
1623         if (arch_info->dr6 & (1 << 14)) {
1624             if (cpu_single_env->singlestep_enabled)
1625                 handle = 1;
1626         } else {
1627             for (n = 0; n < 4; n++)
1628                 if (arch_info->dr6 & (1 << n))
1629                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
1630                     case 0x0:
1631                         handle = 1;
1632                         break;
1633                     case 0x1:
1634                         handle = 1;
1635                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1636                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1637                         hw_watchpoint.flags = BP_MEM_WRITE;
1638                         break;
1639                     case 0x3:
1640                         handle = 1;
1641                         cpu_single_env->watchpoint_hit = &hw_watchpoint;
1642                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1643                         hw_watchpoint.flags = BP_MEM_ACCESS;
1644                         break;
1645                     }
1646         }
1647     } else if (kvm_find_sw_breakpoint(cpu_single_env, arch_info->pc))
1648         handle = 1;
1649
1650     if (!handle) {
1651         cpu_synchronize_state(cpu_single_env);
1652         assert(cpu_single_env->exception_injected == -1);
1653
1654         cpu_single_env->exception_injected = arch_info->exception;
1655         cpu_single_env->has_error_code = 0;
1656     }
1657
1658     return handle;
1659 }
1660
1661 void kvm_arch_update_guest_debug(CPUState *env, struct kvm_guest_debug *dbg)
1662 {
1663     const uint8_t type_code[] = {
1664         [GDB_BREAKPOINT_HW] = 0x0,
1665         [GDB_WATCHPOINT_WRITE] = 0x1,
1666         [GDB_WATCHPOINT_ACCESS] = 0x3
1667     };
1668     const uint8_t len_code[] = {
1669         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1670     };
1671     int n;
1672
1673     if (kvm_sw_breakpoints_active(env))
1674         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1675
1676     if (nb_hw_breakpoint > 0) {
1677         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1678         dbg->arch.debugreg[7] = 0x0600;
1679         for (n = 0; n < nb_hw_breakpoint; n++) {
1680             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
1681             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
1682                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
1683                 (len_code[hw_breakpoint[n].len] << (18 + n*4));
1684         }
1685     }
1686     /* Legal xcr0 for loading */
1687     env->xcr0 = 1;
1688 }
1689 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1690
1691 bool kvm_arch_stop_on_emulation_error(CPUState *env)
1692 {
1693       return !(env->cr[0] & CR0_PE_MASK) ||
1694               ((env->segs[R_CS].selector  & 3) != 3);
1695 }
1696
1697 static void hardware_memory_error(void)
1698 {
1699     fprintf(stderr, "Hardware memory error!\n");
1700     exit(1);
1701 }
1702
1703 #ifdef KVM_CAP_MCE
1704 static void kvm_mce_broadcast_rest(CPUState *env)
1705 {
1706     CPUState *cenv;
1707     int family, model, cpuver = env->cpuid_version;
1708
1709     family = (cpuver >> 8) & 0xf;
1710     model = ((cpuver >> 12) & 0xf0) + ((cpuver >> 4) & 0xf);
1711
1712     /* Broadcast MCA signal for processor version 06H_EH and above */
1713     if ((family == 6 && model >= 14) || family > 6) {
1714         for (cenv = first_cpu; cenv != NULL; cenv = cenv->next_cpu) {
1715             if (cenv == env) {
1716                 continue;
1717             }
1718             kvm_inject_x86_mce(cenv, 1, MCI_STATUS_VAL | MCI_STATUS_UC,
1719                                MCG_STATUS_MCIP | MCG_STATUS_RIPV, 0, 0, 1);
1720         }
1721     }
1722 }
1723 #endif
1724
1725 int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr)
1726 {
1727 #if defined(KVM_CAP_MCE)
1728     struct kvm_x86_mce mce = {
1729             .bank = 9,
1730     };
1731     void *vaddr;
1732     ram_addr_t ram_addr;
1733     target_phys_addr_t paddr;
1734     int r;
1735
1736     if ((env->mcg_cap & MCG_SER_P) && addr
1737         && (code == BUS_MCEERR_AR
1738             || code == BUS_MCEERR_AO)) {
1739         if (code == BUS_MCEERR_AR) {
1740             /* Fake an Intel architectural Data Load SRAR UCR */
1741             mce.status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1742                 | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1743                 | MCI_STATUS_AR | 0x134;
1744             mce.misc = (MCM_ADDR_PHYS << 6) | 0xc;
1745             mce.mcg_status = MCG_STATUS_MCIP | MCG_STATUS_EIPV;
1746         } else {
1747             /*
1748              * If there is an MCE excpetion being processed, ignore
1749              * this SRAO MCE
1750              */
1751             r = kvm_mce_in_exception(env);
1752             if (r == -1) {
1753                 fprintf(stderr, "Failed to get MCE status\n");
1754             } else if (r) {
1755                 return 0;
1756             }
1757             /* Fake an Intel architectural Memory scrubbing UCR */
1758             mce.status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1759                 | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1760                 | 0xc0;
1761             mce.misc = (MCM_ADDR_PHYS << 6) | 0xc;
1762             mce.mcg_status = MCG_STATUS_MCIP | MCG_STATUS_RIPV;
1763         }
1764         vaddr = (void *)addr;
1765         if (qemu_ram_addr_from_host(vaddr, &ram_addr) ||
1766             !kvm_physical_memory_addr_from_ram(env->kvm_state, ram_addr, &paddr)) {
1767             fprintf(stderr, "Hardware memory error for memory used by "
1768                     "QEMU itself instead of guest system!\n");
1769             /* Hope we are lucky for AO MCE */
1770             if (code == BUS_MCEERR_AO) {
1771                 return 0;
1772             } else {
1773                 hardware_memory_error();
1774             }
1775         }
1776         mce.addr = paddr;
1777         r = kvm_set_mce(env, &mce);
1778         if (r < 0) {
1779             fprintf(stderr, "kvm_set_mce: %s\n", strerror(errno));
1780             abort();
1781         }
1782         kvm_mce_broadcast_rest(env);
1783     } else
1784 #endif
1785     {
1786         if (code == BUS_MCEERR_AO) {
1787             return 0;
1788         } else if (code == BUS_MCEERR_AR) {
1789             hardware_memory_error();
1790         } else {
1791             return 1;
1792         }
1793     }
1794     return 0;
1795 }
1796
1797 int kvm_on_sigbus(int code, void *addr)
1798 {
1799 #if defined(KVM_CAP_MCE)
1800     if ((first_cpu->mcg_cap & MCG_SER_P) && addr && code == BUS_MCEERR_AO) {
1801         uint64_t status;
1802         void *vaddr;
1803         ram_addr_t ram_addr;
1804         target_phys_addr_t paddr;
1805
1806         /* Hope we are lucky for AO MCE */
1807         vaddr = addr;
1808         if (qemu_ram_addr_from_host(vaddr, &ram_addr) ||
1809             !kvm_physical_memory_addr_from_ram(first_cpu->kvm_state, ram_addr, &paddr)) {
1810             fprintf(stderr, "Hardware memory error for memory used by "
1811                     "QEMU itself instead of guest system!: %p\n", addr);
1812             return 0;
1813         }
1814         status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1815             | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1816             | 0xc0;
1817         kvm_inject_x86_mce(first_cpu, 9, status,
1818                            MCG_STATUS_MCIP | MCG_STATUS_RIPV, paddr,
1819                            (MCM_ADDR_PHYS << 6) | 0xc, 1);
1820         kvm_mce_broadcast_rest(first_cpu);
1821     } else
1822 #endif
1823     {
1824         if (code == BUS_MCEERR_AO) {
1825             return 0;
1826         } else if (code == BUS_MCEERR_AR) {
1827             hardware_memory_error();
1828         } else {
1829             return 1;
1830         }
1831     }
1832     return 0;
1833 }