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/sysemu.h"
  25 #include "sysemu/kvm.h"
  26 #include "kvm_i386.h"
  27 #include "cpu.h"
  28 #include "exec/gdbstub.h"
  29 #include "qemu/host-utils.h"
  30 #include "qemu/config-file.h"
  31 #include "hw/pc.h"
  32 #include "hw/apic.h"
  33 #include "exec/ioport.h"
  34 #include "hyperv.h"
  35 #include "hw/pci/pci.h"
  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 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  58     KVM_CAP_INFO(SET_TSS_ADDR),
  59     KVM_CAP_INFO(EXT_CPUID),
  60     KVM_CAP_INFO(MP_STATE),
  61     KVM_CAP_LAST_INFO
  62 };
  63
  64 static bool has_msr_star;
  65 static bool has_msr_hsave_pa;
  66 static bool has_msr_tsc_adjust;
  67 static bool has_msr_tsc_deadline;
  68 static bool has_msr_async_pf_en;
  69 static bool has_msr_pv_eoi_en;
  70 static bool has_msr_misc_enable;
  71 static int lm_capable_kernel;
  72
  73 bool kvm_allows_irq0_override(void)
  74 {
  75     return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
  76 }
  77
  78 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
  79 {
  80     struct kvm_cpuid2 *cpuid;
  81     int r, size;
  82
  83     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
  84     cpuid = (struct kvm_cpuid2 *)g_malloc0(size);
  85     cpuid->nent = max;
  86     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
  87     if (r == 0 && cpuid->nent >= max) {
  88         r = -E2BIG;
  89     }
  90     if (r < 0) {
  91         if (r == -E2BIG) {
  92             g_free(cpuid);
  93             return NULL;
  94         } else {
  95             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
  96                     strerror(-r));
  97             exit(1);
  98         }
  99     }
 100     return cpuid;
 101 }
 102
 103 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
 104  * for all entries.
 105  */
 106 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
 107 {
 108     struct kvm_cpuid2 *cpuid;
 109     int max = 1;
 110     while ((cpuid = try_get_cpuid(s, max)) == NULL) {
 111         max *= 2;
 112     }
 113     return cpuid;
 114 }
 115
 116 struct kvm_para_features {
 117     int cap;
 118     int feature;
 119 } para_features[] = {
 120     { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 121     { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 122     { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 123     { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
 124     { -1, -1 }
 125 };
 126
 127 static int get_para_features(KVMState *s)
 128 {
 129     int i, features = 0;
 130
 131     for (i = 0; i < ARRAY_SIZE(para_features) - 1; i++) {
 132         if (kvm_check_extension(s, para_features[i].cap)) {
 133             features |= (1 << para_features[i].feature);
 134         }
 135     }
 136
 137     return features;
 138 }
 139
 140
 141 /* Returns the value for a specific register on the cpuid entry
 142  */
 143 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
 144 {
 145     uint32_t ret = 0;
 146     switch (reg) {
 147     case R_EAX:
 148         ret = entry->eax;
 149         break;
 150     case R_EBX:
 151         ret = entry->ebx;
 152         break;
 153     case R_ECX:
 154         ret = entry->ecx;
 155         break;
 156     case R_EDX:
 157         ret = entry->edx;
 158         break;
 159     }
 160     return ret;
 161 }
 162
 163 /* Find matching entry for function/index on kvm_cpuid2 struct
 164  */
 165 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
 166                                                  uint32_t function,
 167                                                  uint32_t index)
 168 {
 169     int i;
 170     for (i = 0; i < cpuid->nent; ++i) {
 171         if (cpuid->entries[i].function == function &&
 172             cpuid->entries[i].index == index) {
 173             return &cpuid->entries[i];
 174         }
 175     }
 176     /* not found: */
 177     return NULL;
 178 }
 179
 180 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
 181                                       uint32_t index, int reg)
 182 {
 183     struct kvm_cpuid2 *cpuid;
 184     uint32_t ret = 0;
 185     uint32_t cpuid_1_edx;
 186     bool found = false;
 187
 188     cpuid = get_supported_cpuid(s);
 189
 190     struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
 191     if (entry) {
 192         found = true;
 193         ret = cpuid_entry_get_reg(entry, reg);
 194     }
 195
 196     /* Fixups for the data returned by KVM, below */
 197
 198     if (function == 1 && reg == R_EDX) {
 199         /* KVM before 2.6.30 misreports the following features */
 200         ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
 201     } else if (function == 1 && reg == R_ECX) {
 202         /* We can set the hypervisor flag, even if KVM does not return it on
 203          * GET_SUPPORTED_CPUID
 204          */
 205         ret |= CPUID_EXT_HYPERVISOR;
 206         /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
 207          * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
 208          * and the irqchip is in the kernel.
 209          */
 210         if (kvm_irqchip_in_kernel() &&
 211                 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
 212             ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
 213         }
 214
 215         /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
 216          * without the in-kernel irqchip
 217          */
 218         if (!kvm_irqchip_in_kernel()) {
 219             ret &= ~CPUID_EXT_X2APIC;
 220         }
 221     } else if (function == 0x80000001 && reg == R_EDX) {
 222         /* On Intel, kvm returns cpuid according to the Intel spec,
 223          * so add missing bits according to the AMD spec:
 224          */
 225         cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
 226         ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
 227     }
 228
 229     g_free(cpuid);
 230
 231     /* fallback for older kernels */
 232     if ((function == KVM_CPUID_FEATURES) && !found) {
 233         ret = get_para_features(s);
 234     }
 235
 236     return ret;
 237 }
 238
 239 typedef struct HWPoisonPage {
 240     ram_addr_t ram_addr;
 241     QLIST_ENTRY(HWPoisonPage) list;
 242 } HWPoisonPage;
 243
 244 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
 245     QLIST_HEAD_INITIALIZER(hwpoison_page_list);
 246
 247 static void kvm_unpoison_all(void *param)
 248 {
 249     HWPoisonPage *page, *next_page;
 250
 251     QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
 252         QLIST_REMOVE(page, list);
 253         qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
 254         g_free(page);
 255     }
 256 }
 257
 258 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
 259 {
 260     HWPoisonPage *page;
 261
 262     QLIST_FOREACH(page, &hwpoison_page_list, list) {
 263         if (page->ram_addr == ram_addr) {
 264             return;
 265         }
 266     }
 267     page = g_malloc(sizeof(HWPoisonPage));
 268     page->ram_addr = ram_addr;
 269     QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
 270 }
 271
 272 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
 273                                      int *max_banks)
 274 {
 275     int r;
 276
 277     r = kvm_check_extension(s, KVM_CAP_MCE);
 278     if (r > 0) {
 279         *max_banks = r;
 280         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
 281     }
 282     return -ENOSYS;
 283 }
 284
 285 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
 286 {
 287     CPUX86State *env = &cpu->env;
 288     uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
 289                       MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
 290     uint64_t mcg_status = MCG_STATUS_MCIP;
 291
 292     if (code == BUS_MCEERR_AR) {
 293         status |= MCI_STATUS_AR | 0x134;
 294         mcg_status |= MCG_STATUS_EIPV;
 295     } else {
 296         status |= 0xc0;
 297         mcg_status |= MCG_STATUS_RIPV;
 298     }
 299     cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
 300                        (MCM_ADDR_PHYS << 6) | 0xc,
 301                        cpu_x86_support_mca_broadcast(env) ?
 302                        MCE_INJECT_BROADCAST : 0);
 303 }
 304
 305 static void hardware_memory_error(void)
 306 {
 307     fprintf(stderr, "Hardware memory error!\n");
 308     exit(1);
 309 }
 310
 311 int kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
 312 {
 313     X86CPU *cpu = X86_CPU(c);
 314     CPUX86State *env = &cpu->env;
 315     ram_addr_t ram_addr;
 316     hwaddr paddr;
 317
 318     if ((env->mcg_cap & MCG_SER_P) && addr
 319         && (code == BUS_MCEERR_AR || code == BUS_MCEERR_AO)) {
 320         if (qemu_ram_addr_from_host(addr, &ram_addr) ||
 321             !kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
 322             fprintf(stderr, "Hardware memory error for memory used by "
 323                     "QEMU itself instead of guest system!\n");
 324             /* Hope we are lucky for AO MCE */
 325             if (code == BUS_MCEERR_AO) {
 326                 return 0;
 327             } else {
 328                 hardware_memory_error();
 329             }
 330         }
 331         kvm_hwpoison_page_add(ram_addr);
 332         kvm_mce_inject(cpu, paddr, code);
 333     } else {
 334         if (code == BUS_MCEERR_AO) {
 335             return 0;
 336         } else if (code == BUS_MCEERR_AR) {
 337             hardware_memory_error();
 338         } else {
 339             return 1;
 340         }
 341     }
 342     return 0;
 343 }
 344
 345 int kvm_arch_on_sigbus(int code, void *addr)
 346 {
 347     if ((first_cpu->mcg_cap & MCG_SER_P) && addr && code == BUS_MCEERR_AO) {
 348         ram_addr_t ram_addr;
 349         hwaddr paddr;
 350
 351         /* Hope we are lucky for AO MCE */
 352         if (qemu_ram_addr_from_host(addr, &ram_addr) ||
 353             !kvm_physical_memory_addr_from_host(CPU(first_cpu)->kvm_state,
 354                                                 addr, &paddr)) {
 355             fprintf(stderr, "Hardware memory error for memory used by "
 356                     "QEMU itself instead of guest system!: %p\n", addr);
 357             return 0;
 358         }
 359         kvm_hwpoison_page_add(ram_addr);
 360         kvm_mce_inject(x86_env_get_cpu(first_cpu), paddr, code);
 361     } else {
 362         if (code == BUS_MCEERR_AO) {
 363             return 0;
 364         } else if (code == BUS_MCEERR_AR) {
 365             hardware_memory_error();
 366         } else {
 367             return 1;
 368         }
 369     }
 370     return 0;
 371 }
 372
 373 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
 374 {
 375     CPUX86State *env = &cpu->env;
 376
 377     if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
 378         unsigned int bank, bank_num = env->mcg_cap & 0xff;
 379         struct kvm_x86_mce mce;
 380
 381         env->exception_injected = -1;
 382
 383         /*
 384          * There must be at least one bank in use if an MCE is pending.
 385          * Find it and use its values for the event injection.
 386          */
 387         for (bank = 0; bank < bank_num; bank++) {
 388             if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
 389                 break;
 390             }
 391         }
 392         assert(bank < bank_num);
 393
 394         mce.bank = bank;
 395         mce.status = env->mce_banks[bank * 4 + 1];
 396         mce.mcg_status = env->mcg_status;
 397         mce.addr = env->mce_banks[bank * 4 + 2];
 398         mce.misc = env->mce_banks[bank * 4 + 3];
 399
 400         return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
 401     }
 402     return 0;
 403 }
 404
 405 static void cpu_update_state(void *opaque, int running, RunState state)
 406 {
 407     CPUX86State *env = opaque;
 408
 409     if (running) {
 410         env->tsc_valid = false;
 411     }
 412 }
 413
 414 unsigned long kvm_arch_vcpu_id(CPUState *cs)
 415 {
 416     X86CPU *cpu = X86_CPU(cs);
 417     return cpu->env.cpuid_apic_id;
 418 }
 419
 420 #define KVM_MAX_CPUID_ENTRIES  100
 421
 422 int kvm_arch_init_vcpu(CPUState *cs)
 423 {
 424     struct {
 425         struct kvm_cpuid2 cpuid;
 426         struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
 427     } QEMU_PACKED cpuid_data;
 428     X86CPU *cpu = X86_CPU(cs);
 429     CPUX86State *env = &cpu->env;
 430     uint32_t limit, i, j, cpuid_i;
 431     uint32_t unused;
 432     struct kvm_cpuid_entry2 *c;
 433     uint32_t signature[3];
 434     int r;
 435
 436     cpuid_i = 0;
 437
 438     /* Paravirtualization CPUIDs */
 439     c = &cpuid_data.entries[cpuid_i++];
 440     memset(c, 0, sizeof(*c));
 441     c->function = KVM_CPUID_SIGNATURE;
 442     if (!hyperv_enabled()) {
 443         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 444         c->eax = 0;
 445     } else {
 446         memcpy(signature, "Microsoft Hv", 12);
 447         c->eax = HYPERV_CPUID_MIN;
 448     }
 449     c->ebx = signature[0];
 450     c->ecx = signature[1];
 451     c->edx = signature[2];
 452
 453     c = &cpuid_data.entries[cpuid_i++];
 454     memset(c, 0, sizeof(*c));
 455     c->function = KVM_CPUID_FEATURES;
 456     c->eax = env->cpuid_kvm_features;
 457
 458     if (hyperv_enabled()) {
 459         memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
 460         c->eax = signature[0];
 461
 462         c = &cpuid_data.entries[cpuid_i++];
 463         memset(c, 0, sizeof(*c));
 464         c->function = HYPERV_CPUID_VERSION;
 465         c->eax = 0x00001bbc;
 466         c->ebx = 0x00060001;
 467
 468         c = &cpuid_data.entries[cpuid_i++];
 469         memset(c, 0, sizeof(*c));
 470         c->function = HYPERV_CPUID_FEATURES;
 471         if (hyperv_relaxed_timing_enabled()) {
 472             c->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
 473         }
 474         if (hyperv_vapic_recommended()) {
 475             c->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
 476             c->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
 477         }
 478
 479         c = &cpuid_data.entries[cpuid_i++];
 480         memset(c, 0, sizeof(*c));
 481         c->function = HYPERV_CPUID_ENLIGHTMENT_INFO;
 482         if (hyperv_relaxed_timing_enabled()) {
 483             c->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
 484         }
 485         if (hyperv_vapic_recommended()) {
 486             c->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
 487         }
 488         c->ebx = hyperv_get_spinlock_retries();
 489
 490         c = &cpuid_data.entries[cpuid_i++];
 491         memset(c, 0, sizeof(*c));
 492         c->function = HYPERV_CPUID_IMPLEMENT_LIMITS;
 493         c->eax = 0x40;
 494         c->ebx = 0x40;
 495
 496         c = &cpuid_data.entries[cpuid_i++];
 497         memset(c, 0, sizeof(*c));
 498         c->function = KVM_CPUID_SIGNATURE_NEXT;
 499         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 500         c->eax = 0;
 501         c->ebx = signature[0];
 502         c->ecx = signature[1];
 503         c->edx = signature[2];
 504     }
 505
 506     has_msr_async_pf_en = c->eax & (1 << KVM_FEATURE_ASYNC_PF);
 507
 508     has_msr_pv_eoi_en = c->eax & (1 << KVM_FEATURE_PV_EOI);
 509
 510     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 511
 512     for (i = 0; i <= limit; i++) {
 513         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 514             fprintf(stderr, "unsupported level value: 0x%x\n", limit);
 515             abort();
 516         }
 517         c = &cpuid_data.entries[cpuid_i++];
 518
 519         switch (i) {
 520         case 2: {
 521             /* Keep reading function 2 till all the input is received */
 522             int times;
 523
 524             c->function = i;
 525             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 526                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 527             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 528             times = c->eax & 0xff;
 529
 530             for (j = 1; j < times; ++j) {
 531                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 532                     fprintf(stderr, "cpuid_data is full, no space for "
 533                             "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
 534                     abort();
 535                 }
 536                 c = &cpuid_data.entries[cpuid_i++];
 537                 c->function = i;
 538                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 539                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 540             }
 541             break;
 542         }
 543         case 4:
 544         case 0xb:
 545         case 0xd:
 546             for (j = 0; ; j++) {
 547                 if (i == 0xd && j == 64) {
 548                     break;
 549                 }
 550                 c->function = i;
 551                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 552                 c->index = j;
 553                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
 554
 555                 if (i == 4 && c->eax == 0) {
 556                     break;
 557                 }
 558                 if (i == 0xb && !(c->ecx & 0xff00)) {
 559                     break;
 560                 }
 561                 if (i == 0xd && c->eax == 0) {
 562                     continue;
 563                 }
 564                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 565                     fprintf(stderr, "cpuid_data is full, no space for "
 566                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
 567                     abort();
 568                 }
 569                 c = &cpuid_data.entries[cpuid_i++];
 570             }
 571             break;
 572         default:
 573             c->function = i;
 574             c->flags = 0;
 575             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 576             break;
 577         }
 578     }
 579     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
 580
 581     for (i = 0x80000000; i <= limit; i++) {
 582         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 583             fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
 584             abort();
 585         }
 586         c = &cpuid_data.entries[cpuid_i++];
 587
 588         c->function = i;
 589         c->flags = 0;
 590         cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 591     }
 592
 593     /* Call Centaur's CPUID instructions they are supported. */
 594     if (env->cpuid_xlevel2 > 0) {
 595         cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
 596
 597         for (i = 0xC0000000; i <= limit; i++) {
 598             if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 599                 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
 600                 abort();
 601             }
 602             c = &cpuid_data.entries[cpuid_i++];
 603
 604             c->function = i;
 605             c->flags = 0;
 606             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 607         }
 608     }
 609
 610     cpuid_data.cpuid.nent = cpuid_i;
 611
 612     if (((env->cpuid_version >> 8)&0xF) >= 6
 613         && (env->cpuid_features&(CPUID_MCE|CPUID_MCA)) == (CPUID_MCE|CPUID_MCA)
 614         && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
 615         uint64_t mcg_cap;
 616         int banks;
 617         int ret;
 618
 619         ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
 620         if (ret < 0) {
 621             fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
 622             return ret;
 623         }
 624
 625         if (banks > MCE_BANKS_DEF) {
 626             banks = MCE_BANKS_DEF;
 627         }
 628         mcg_cap &= MCE_CAP_DEF;
 629         mcg_cap |= banks;
 630         ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &mcg_cap);
 631         if (ret < 0) {
 632             fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
 633             return ret;
 634         }
 635
 636         env->mcg_cap = mcg_cap;
 637     }
 638
 639     qemu_add_vm_change_state_handler(cpu_update_state, env);
 640
 641     cpuid_data.cpuid.padding = 0;
 642     r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
 643     if (r) {
 644         return r;
 645     }
 646
 647     r = kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL);
 648     if (r && env->tsc_khz) {
 649         r = kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz);
 650         if (r < 0) {
 651             fprintf(stderr, "KVM_SET_TSC_KHZ failed\n");
 652             return r;
 653         }
 654     }
 655
 656     if (kvm_has_xsave()) {
 657         env->kvm_xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
 658     }
 659
 660     return 0;
 661 }
 662
 663 void kvm_arch_reset_vcpu(CPUState *cs)
 664 {
 665     X86CPU *cpu = X86_CPU(cs);
 666     CPUX86State *env = &cpu->env;
 667
 668     env->exception_injected = -1;
 669     env->interrupt_injected = -1;
 670     env->xcr0 = 1;
 671     if (kvm_irqchip_in_kernel()) {
 672         env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
 673                                           KVM_MP_STATE_UNINITIALIZED;
 674     } else {
 675         env->mp_state = KVM_MP_STATE_RUNNABLE;
 676     }
 677 }
 678
 679 static int kvm_get_supported_msrs(KVMState *s)
 680 {
 681     static int kvm_supported_msrs;
 682     int ret = 0;
 683
 684     /* first time */
 685     if (kvm_supported_msrs == 0) {
 686         struct kvm_msr_list msr_list, *kvm_msr_list;
 687
 688         kvm_supported_msrs = -1;
 689
 690         /* Obtain MSR list from KVM.  These are the MSRs that we must
 691          * save/restore */
 692         msr_list.nmsrs = 0;
 693         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
 694         if (ret < 0 && ret != -E2BIG) {
 695             return ret;
 696         }
 697         /* Old kernel modules had a bug and could write beyond the provided
 698            memory. Allocate at least a safe amount of 1K. */
 699         kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
 700                                               msr_list.nmsrs *
 701                                               sizeof(msr_list.indices[0])));
 702
 703         kvm_msr_list->nmsrs = msr_list.nmsrs;
 704         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
 705         if (ret >= 0) {
 706             int i;
 707
 708             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
 709                 if (kvm_msr_list->indices[i] == MSR_STAR) {
 710                     has_msr_star = true;
 711                     continue;
 712                 }
 713                 if (kvm_msr_list->indices[i] == MSR_VM_HSAVE_PA) {
 714                     has_msr_hsave_pa = true;
 715                     continue;
 716                 }
 717                 if (kvm_msr_list->indices[i] == MSR_TSC_ADJUST) {
 718                     has_msr_tsc_adjust = true;
 719                     continue;
 720                 }
 721                 if (kvm_msr_list->indices[i] == MSR_IA32_TSCDEADLINE) {
 722                     has_msr_tsc_deadline = true;
 723                     continue;
 724                 }
 725                 if (kvm_msr_list->indices[i] == MSR_IA32_MISC_ENABLE) {
 726                     has_msr_misc_enable = true;
 727                     continue;
 728                 }
 729             }
 730         }
 731
 732         g_free(kvm_msr_list);
 733     }
 734
 735     return ret;
 736 }
 737
 738 int kvm_arch_init(KVMState *s)
 739 {
 740     QemuOptsList *list = qemu_find_opts("machine");
 741     uint64_t identity_base = 0xfffbc000;
 742     uint64_t shadow_mem;
 743     int ret;
 744     struct utsname utsname;
 745
 746     ret = kvm_get_supported_msrs(s);
 747     if (ret < 0) {
 748         return ret;
 749     }
 750
 751     uname(&utsname);
 752     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
 753
 754     /*
 755      * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
 756      * In order to use vm86 mode, an EPT identity map and a TSS  are needed.
 757      * Since these must be part of guest physical memory, we need to allocate
 758      * them, both by setting their start addresses in the kernel and by
 759      * creating a corresponding e820 entry. We need 4 pages before the BIOS.
 760      *
 761      * Older KVM versions may not support setting the identity map base. In
 762      * that case we need to stick with the default, i.e. a 256K maximum BIOS
 763      * size.
 764      */
 765     if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
 766         /* Allows up to 16M BIOSes. */
 767         identity_base = 0xfeffc000;
 768
 769         ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
 770         if (ret < 0) {
 771             return ret;
 772         }
 773     }
 774
 775     /* Set TSS base one page after EPT identity map. */
 776     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
 777     if (ret < 0) {
 778         return ret;
 779     }
 780
 781     /* Tell fw_cfg to notify the BIOS to reserve the range. */
 782     ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
 783     if (ret < 0) {
 784         fprintf(stderr, "e820_add_entry() table is full\n");
 785         return ret;
 786     }
 787     qemu_register_reset(kvm_unpoison_all, NULL);
 788
 789     if (!QTAILQ_EMPTY(&list->head)) {
 790         shadow_mem = qemu_opt_get_size(QTAILQ_FIRST(&list->head),
 791                                        "kvm_shadow_mem", -1);
 792         if (shadow_mem != -1) {
 793             shadow_mem /= 4096;
 794             ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
 795             if (ret < 0) {
 796                 return ret;
 797             }
 798         }
 799     }
 800     return 0;
 801 }
 802
 803 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 804 {
 805     lhs->selector = rhs->selector;
 806     lhs->base = rhs->base;
 807     lhs->limit = rhs->limit;
 808     lhs->type = 3;
 809     lhs->present = 1;
 810     lhs->dpl = 3;
 811     lhs->db = 0;
 812     lhs->s = 1;
 813     lhs->l = 0;
 814     lhs->g = 0;
 815     lhs->avl = 0;
 816     lhs->unusable = 0;
 817 }
 818
 819 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
 820 {
 821     unsigned flags = rhs->flags;
 822     lhs->selector = rhs->selector;
 823     lhs->base = rhs->base;
 824     lhs->limit = rhs->limit;
 825     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
 826     lhs->present = (flags & DESC_P_MASK) != 0;
 827     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
 828     lhs->db = (flags >> DESC_B_SHIFT) & 1;
 829     lhs->s = (flags & DESC_S_MASK) != 0;
 830     lhs->l = (flags >> DESC_L_SHIFT) & 1;
 831     lhs->g = (flags & DESC_G_MASK) != 0;
 832     lhs->avl = (flags & DESC_AVL_MASK) != 0;
 833     lhs->unusable = 0;
 834     lhs->padding = 0;
 835 }
 836
 837 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
 838 {
 839     lhs->selector = rhs->selector;
 840     lhs->base = rhs->base;
 841     lhs->limit = rhs->limit;
 842     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
 843                  (rhs->present * DESC_P_MASK) |
 844                  (rhs->dpl << DESC_DPL_SHIFT) |
 845                  (rhs->db << DESC_B_SHIFT) |
 846                  (rhs->s * DESC_S_MASK) |
 847                  (rhs->l << DESC_L_SHIFT) |
 848                  (rhs->g * DESC_G_MASK) |
 849                  (rhs->avl * DESC_AVL_MASK);
 850 }
 851
 852 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
 853 {
 854     if (set) {
 855         *kvm_reg = *qemu_reg;
 856     } else {
 857         *qemu_reg = *kvm_reg;
 858     }
 859 }
 860
 861 static int kvm_getput_regs(X86CPU *cpu, int set)
 862 {
 863     CPUX86State *env = &cpu->env;
 864     struct kvm_regs regs;
 865     int ret = 0;
 866
 867     if (!set) {
 868         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
 869         if (ret < 0) {
 870             return ret;
 871         }
 872     }
 873
 874     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
 875     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
 876     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
 877     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
 878     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
 879     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
 880     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
 881     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
 882 #ifdef TARGET_X86_64
 883     kvm_getput_reg(&regs.r8, &env->regs[8], set);
 884     kvm_getput_reg(&regs.r9, &env->regs[9], set);
 885     kvm_getput_reg(&regs.r10, &env->regs[10], set);
 886     kvm_getput_reg(&regs.r11, &env->regs[11], set);
 887     kvm_getput_reg(&regs.r12, &env->regs[12], set);
 888     kvm_getput_reg(&regs.r13, &env->regs[13], set);
 889     kvm_getput_reg(&regs.r14, &env->regs[14], set);
 890     kvm_getput_reg(&regs.r15, &env->regs[15], set);
 891 #endif
 892
 893     kvm_getput_reg(&regs.rflags, &env->eflags, set);
 894     kvm_getput_reg(&regs.rip, &env->eip, set);
 895
 896     if (set) {
 897         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
 898     }
 899
 900     return ret;
 901 }
 902
 903 static int kvm_put_fpu(X86CPU *cpu)
 904 {
 905     CPUX86State *env = &cpu->env;
 906     struct kvm_fpu fpu;
 907     int i;
 908
 909     memset(&fpu, 0, sizeof fpu);
 910     fpu.fsw = env->fpus & ~(7 << 11);
 911     fpu.fsw |= (env->fpstt & 7) << 11;
 912     fpu.fcw = env->fpuc;
 913     fpu.last_opcode = env->fpop;
 914     fpu.last_ip = env->fpip;
 915     fpu.last_dp = env->fpdp;
 916     for (i = 0; i < 8; ++i) {
 917         fpu.ftwx |= (!env->fptags[i]) << i;
 918     }
 919     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
 920     memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
 921     fpu.mxcsr = env->mxcsr;
 922
 923     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
 924 }
 925
 926 #define XSAVE_FCW_FSW     0
 927 #define XSAVE_FTW_FOP     1
 928 #define XSAVE_CWD_RIP     2
 929 #define XSAVE_CWD_RDP     4
 930 #define XSAVE_MXCSR       6
 931 #define XSAVE_ST_SPACE    8
 932 #define XSAVE_XMM_SPACE   40
 933 #define XSAVE_XSTATE_BV   128
 934 #define XSAVE_YMMH_SPACE  144
 935
 936 static int kvm_put_xsave(X86CPU *cpu)
 937 {
 938     CPUX86State *env = &cpu->env;
 939     struct kvm_xsave* xsave = env->kvm_xsave_buf;
 940     uint16_t cwd, swd, twd;
 941     int i, r;
 942
 943     if (!kvm_has_xsave()) {
 944         return kvm_put_fpu(cpu);
 945     }
 946
 947     memset(xsave, 0, sizeof(struct kvm_xsave));
 948     twd = 0;
 949     swd = env->fpus & ~(7 << 11);
 950     swd |= (env->fpstt & 7) << 11;
 951     cwd = env->fpuc;
 952     for (i = 0; i < 8; ++i) {
 953         twd |= (!env->fptags[i]) << i;
 954     }
 955     xsave->region[XSAVE_FCW_FSW] = (uint32_t)(swd << 16) + cwd;
 956     xsave->region[XSAVE_FTW_FOP] = (uint32_t)(env->fpop << 16) + twd;
 957     memcpy(&xsave->region[XSAVE_CWD_RIP], &env->fpip, sizeof(env->fpip));
 958     memcpy(&xsave->region[XSAVE_CWD_RDP], &env->fpdp, sizeof(env->fpdp));
 959     memcpy(&xsave->region[XSAVE_ST_SPACE], env->fpregs,
 960             sizeof env->fpregs);
 961     memcpy(&xsave->region[XSAVE_XMM_SPACE], env->xmm_regs,
 962             sizeof env->xmm_regs);
 963     xsave->region[XSAVE_MXCSR] = env->mxcsr;
 964     *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV] = env->xstate_bv;
 965     memcpy(&xsave->region[XSAVE_YMMH_SPACE], env->ymmh_regs,
 966             sizeof env->ymmh_regs);
 967     r = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
 968     return r;
 969 }
 970
 971 static int kvm_put_xcrs(X86CPU *cpu)
 972 {
 973     CPUX86State *env = &cpu->env;
 974     struct kvm_xcrs xcrs;
 975
 976     if (!kvm_has_xcrs()) {
 977         return 0;
 978     }
 979
 980     xcrs.nr_xcrs = 1;
 981     xcrs.flags = 0;
 982     xcrs.xcrs[0].xcr = 0;
 983     xcrs.xcrs[0].value = env->xcr0;
 984     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
 985 }
 986
 987 static int kvm_put_sregs(X86CPU *cpu)
 988 {
 989     CPUX86State *env = &cpu->env;
 990     struct kvm_sregs sregs;
 991
 992     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
 993     if (env->interrupt_injected >= 0) {
 994         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
 995                 (uint64_t)1 << (env->interrupt_injected % 64);
 996     }
 997
 998     if ((env->eflags & VM_MASK)) {
 999         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
1000         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
1001         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
1002         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
1003         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
1004         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
1005     } else {
1006         set_seg(&sregs.cs, &env->segs[R_CS]);
1007         set_seg(&sregs.ds, &env->segs[R_DS]);
1008         set_seg(&sregs.es, &env->segs[R_ES]);
1009         set_seg(&sregs.fs, &env->segs[R_FS]);
1010         set_seg(&sregs.gs, &env->segs[R_GS]);
1011         set_seg(&sregs.ss, &env->segs[R_SS]);
1012     }
1013
1014     set_seg(&sregs.tr, &env->tr);
1015     set_seg(&sregs.ldt, &env->ldt);
1016
1017     sregs.idt.limit = env->idt.limit;
1018     sregs.idt.base = env->idt.base;
1019     memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
1020     sregs.gdt.limit = env->gdt.limit;
1021     sregs.gdt.base = env->gdt.base;
1022     memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
1023
1024     sregs.cr0 = env->cr[0];
1025     sregs.cr2 = env->cr[2];
1026     sregs.cr3 = env->cr[3];
1027     sregs.cr4 = env->cr[4];
1028
1029     sregs.cr8 = cpu_get_apic_tpr(env->apic_state);
1030     sregs.apic_base = cpu_get_apic_base(env->apic_state);
1031
1032     sregs.efer = env->efer;
1033
1034     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
1035 }
1036
1037 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
1038                               uint32_t index, uint64_t value)
1039 {
1040     entry->index = index;
1041     entry->data = value;
1042 }
1043
1044 static int kvm_put_msrs(X86CPU *cpu, int level)
1045 {
1046     CPUX86State *env = &cpu->env;
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 n = 0;
1053
1054     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
1055     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
1056     kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
1057     kvm_msr_entry_set(&msrs[n++], MSR_PAT, env->pat);
1058     if (has_msr_star) {
1059         kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
1060     }
1061     if (has_msr_hsave_pa) {
1062         kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
1063     }
1064     if (has_msr_tsc_adjust) {
1065         kvm_msr_entry_set(&msrs[n++], MSR_TSC_ADJUST, env->tsc_adjust);
1066     }
1067     if (has_msr_tsc_deadline) {
1068         kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSCDEADLINE, env->tsc_deadline);
1069     }
1070     if (has_msr_misc_enable) {
1071         kvm_msr_entry_set(&msrs[n++], MSR_IA32_MISC_ENABLE,
1072                           env->msr_ia32_misc_enable);
1073     }
1074 #ifdef TARGET_X86_64
1075     if (lm_capable_kernel) {
1076         kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
1077         kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
1078         kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
1079         kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
1080     }
1081 #endif
1082     if (level == KVM_PUT_FULL_STATE) {
1083         /*
1084          * KVM is yet unable to synchronize TSC values of multiple VCPUs on
1085          * writeback. Until this is fixed, we only write the offset to SMP
1086          * guests after migration, desynchronizing the VCPUs, but avoiding
1087          * huge jump-backs that would occur without any writeback at all.
1088          */
1089         if (smp_cpus == 1 || env->tsc != 0) {
1090             kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
1091         }
1092     }
1093     /*
1094      * The following paravirtual MSRs have side effects on the guest or are
1095      * too heavy for normal writeback. Limit them to reset or full state
1096      * updates.
1097      */
1098     if (level >= KVM_PUT_RESET_STATE) {
1099         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
1100                           env->system_time_msr);
1101         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
1102         if (has_msr_async_pf_en) {
1103             kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN,
1104                               env->async_pf_en_msr);
1105         }
1106         if (has_msr_pv_eoi_en) {
1107             kvm_msr_entry_set(&msrs[n++], MSR_KVM_PV_EOI_EN,
1108                               env->pv_eoi_en_msr);
1109         }
1110         if (hyperv_hypercall_available()) {
1111             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_GUEST_OS_ID, 0);
1112             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_HYPERCALL, 0);
1113         }
1114         if (hyperv_vapic_recommended()) {
1115             kvm_msr_entry_set(&msrs[n++], HV_X64_MSR_APIC_ASSIST_PAGE, 0);
1116         }
1117     }
1118     if (env->mcg_cap) {
1119         int i;
1120
1121         kvm_msr_entry_set(&msrs[n++], MSR_MCG_STATUS, env->mcg_status);
1122         kvm_msr_entry_set(&msrs[n++], MSR_MCG_CTL, env->mcg_ctl);
1123         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1124             kvm_msr_entry_set(&msrs[n++], MSR_MC0_CTL + i, env->mce_banks[i]);
1125         }
1126     }
1127
1128     msr_data.info.nmsrs = n;
1129
1130     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, &msr_data);
1131
1132 }
1133
1134
1135 static int kvm_get_fpu(X86CPU *cpu)
1136 {
1137     CPUX86State *env = &cpu->env;
1138     struct kvm_fpu fpu;
1139     int i, ret;
1140
1141     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
1142     if (ret < 0) {
1143         return ret;
1144     }
1145
1146     env->fpstt = (fpu.fsw >> 11) & 7;
1147     env->fpus = fpu.fsw;
1148     env->fpuc = fpu.fcw;
1149     env->fpop = fpu.last_opcode;
1150     env->fpip = fpu.last_ip;
1151     env->fpdp = fpu.last_dp;
1152     for (i = 0; i < 8; ++i) {
1153         env->fptags[i] = !((fpu.ftwx >> i) & 1);
1154     }
1155     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
1156     memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
1157     env->mxcsr = fpu.mxcsr;
1158
1159     return 0;
1160 }
1161
1162 static int kvm_get_xsave(X86CPU *cpu)
1163 {
1164     CPUX86State *env = &cpu->env;
1165     struct kvm_xsave* xsave = env->kvm_xsave_buf;
1166     int ret, i;
1167     uint16_t cwd, swd, twd;
1168
1169     if (!kvm_has_xsave()) {
1170         return kvm_get_fpu(cpu);
1171     }
1172
1173     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
1174     if (ret < 0) {
1175         return ret;
1176     }
1177
1178     cwd = (uint16_t)xsave->region[XSAVE_FCW_FSW];
1179     swd = (uint16_t)(xsave->region[XSAVE_FCW_FSW] >> 16);
1180     twd = (uint16_t)xsave->region[XSAVE_FTW_FOP];
1181     env->fpop = (uint16_t)(xsave->region[XSAVE_FTW_FOP] >> 16);
1182     env->fpstt = (swd >> 11) & 7;
1183     env->fpus = swd;
1184     env->fpuc = cwd;
1185     for (i = 0; i < 8; ++i) {
1186         env->fptags[i] = !((twd >> i) & 1);
1187     }
1188     memcpy(&env->fpip, &xsave->region[XSAVE_CWD_RIP], sizeof(env->fpip));
1189     memcpy(&env->fpdp, &xsave->region[XSAVE_CWD_RDP], sizeof(env->fpdp));
1190     env->mxcsr = xsave->region[XSAVE_MXCSR];
1191     memcpy(env->fpregs, &xsave->region[XSAVE_ST_SPACE],
1192             sizeof env->fpregs);
1193     memcpy(env->xmm_regs, &xsave->region[XSAVE_XMM_SPACE],
1194             sizeof env->xmm_regs);
1195     env->xstate_bv = *(uint64_t *)&xsave->region[XSAVE_XSTATE_BV];
1196     memcpy(env->ymmh_regs, &xsave->region[XSAVE_YMMH_SPACE],
1197             sizeof env->ymmh_regs);
1198     return 0;
1199 }
1200
1201 static int kvm_get_xcrs(X86CPU *cpu)
1202 {
1203     CPUX86State *env = &cpu->env;
1204     int i, ret;
1205     struct kvm_xcrs xcrs;
1206
1207     if (!kvm_has_xcrs()) {
1208         return 0;
1209     }
1210
1211     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
1212     if (ret < 0) {
1213         return ret;
1214     }
1215
1216     for (i = 0; i < xcrs.nr_xcrs; i++) {
1217         /* Only support xcr0 now */
1218         if (xcrs.xcrs[0].xcr == 0) {
1219             env->xcr0 = xcrs.xcrs[0].value;
1220             break;
1221         }
1222     }
1223     return 0;
1224 }
1225
1226 static int kvm_get_sregs(X86CPU *cpu)
1227 {
1228     CPUX86State *env = &cpu->env;
1229     struct kvm_sregs sregs;
1230     uint32_t hflags;
1231     int bit, i, ret;
1232
1233     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1234     if (ret < 0) {
1235         return ret;
1236     }
1237
1238     /* There can only be one pending IRQ set in the bitmap at a time, so try
1239        to find it and save its number instead (-1 for none). */
1240     env->interrupt_injected = -1;
1241     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
1242         if (sregs.interrupt_bitmap[i]) {
1243             bit = ctz64(sregs.interrupt_bitmap[i]);
1244             env->interrupt_injected = i * 64 + bit;
1245             break;
1246         }
1247     }
1248
1249     get_seg(&env->segs[R_CS], &sregs.cs);
1250     get_seg(&env->segs[R_DS], &sregs.ds);
1251     get_seg(&env->segs[R_ES], &sregs.es);
1252     get_seg(&env->segs[R_FS], &sregs.fs);
1253     get_seg(&env->segs[R_GS], &sregs.gs);
1254     get_seg(&env->segs[R_SS], &sregs.ss);
1255
1256     get_seg(&env->tr, &sregs.tr);
1257     get_seg(&env->ldt, &sregs.ldt);
1258
1259     env->idt.limit = sregs.idt.limit;
1260     env->idt.base = sregs.idt.base;
1261     env->gdt.limit = sregs.gdt.limit;
1262     env->gdt.base = sregs.gdt.base;
1263
1264     env->cr[0] = sregs.cr0;
1265     env->cr[2] = sregs.cr2;
1266     env->cr[3] = sregs.cr3;
1267     env->cr[4] = sregs.cr4;
1268
1269     env->efer = sregs.efer;
1270
1271     /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
1272
1273 #define HFLAG_COPY_MASK \
1274     ~( HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
1275        HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
1276        HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
1277        HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
1278
1279     hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
1280     hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
1281     hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
1282                 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
1283     hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
1284     hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
1285                 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
1286
1287     if (env->efer & MSR_EFER_LMA) {
1288         hflags |= HF_LMA_MASK;
1289     }
1290
1291     if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
1292         hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
1293     } else {
1294         hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
1295                     (DESC_B_SHIFT - HF_CS32_SHIFT);
1296         hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
1297                     (DESC_B_SHIFT - HF_SS32_SHIFT);
1298         if (!(env->cr[0] & CR0_PE_MASK) || (env->eflags & VM_MASK) ||
1299             !(hflags & HF_CS32_MASK)) {
1300             hflags |= HF_ADDSEG_MASK;
1301         } else {
1302             hflags |= ((env->segs[R_DS].base | env->segs[R_ES].base |
1303                         env->segs[R_SS].base) != 0) << HF_ADDSEG_SHIFT;
1304         }
1305     }
1306     env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
1307
1308     return 0;
1309 }
1310
1311 static int kvm_get_msrs(X86CPU *cpu)
1312 {
1313     CPUX86State *env = &cpu->env;
1314     struct {
1315         struct kvm_msrs info;
1316         struct kvm_msr_entry entries[100];
1317     } msr_data;
1318     struct kvm_msr_entry *msrs = msr_data.entries;
1319     int ret, i, n;
1320
1321     n = 0;
1322     msrs[n++].index = MSR_IA32_SYSENTER_CS;
1323     msrs[n++].index = MSR_IA32_SYSENTER_ESP;
1324     msrs[n++].index = MSR_IA32_SYSENTER_EIP;
1325     msrs[n++].index = MSR_PAT;
1326     if (has_msr_star) {
1327         msrs[n++].index = MSR_STAR;
1328     }
1329     if (has_msr_hsave_pa) {
1330         msrs[n++].index = MSR_VM_HSAVE_PA;
1331     }
1332     if (has_msr_tsc_adjust) {
1333         msrs[n++].index = MSR_TSC_ADJUST;
1334     }
1335     if (has_msr_tsc_deadline) {
1336         msrs[n++].index = MSR_IA32_TSCDEADLINE;
1337     }
1338     if (has_msr_misc_enable) {
1339         msrs[n++].index = MSR_IA32_MISC_ENABLE;
1340     }
1341
1342     if (!env->tsc_valid) {
1343         msrs[n++].index = MSR_IA32_TSC;
1344         env->tsc_valid = !runstate_is_running();
1345     }
1346
1347 #ifdef TARGET_X86_64
1348     if (lm_capable_kernel) {
1349         msrs[n++].index = MSR_CSTAR;
1350         msrs[n++].index = MSR_KERNELGSBASE;
1351         msrs[n++].index = MSR_FMASK;
1352         msrs[n++].index = MSR_LSTAR;
1353     }
1354 #endif
1355     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
1356     msrs[n++].index = MSR_KVM_WALL_CLOCK;
1357     if (has_msr_async_pf_en) {
1358         msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
1359     }
1360     if (has_msr_pv_eoi_en) {
1361         msrs[n++].index = MSR_KVM_PV_EOI_EN;
1362     }
1363
1364     if (env->mcg_cap) {
1365         msrs[n++].index = MSR_MCG_STATUS;
1366         msrs[n++].index = MSR_MCG_CTL;
1367         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
1368             msrs[n++].index = MSR_MC0_CTL + i;
1369         }
1370     }
1371
1372     msr_data.info.nmsrs = n;
1373     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
1374     if (ret < 0) {
1375         return ret;
1376     }
1377
1378     for (i = 0; i < ret; i++) {
1379         switch (msrs[i].index) {
1380         case MSR_IA32_SYSENTER_CS:
1381             env->sysenter_cs = msrs[i].data;
1382             break;
1383         case MSR_IA32_SYSENTER_ESP:
1384             env->sysenter_esp = msrs[i].data;
1385             break;
1386         case MSR_IA32_SYSENTER_EIP:
1387             env->sysenter_eip = msrs[i].data;
1388             break;
1389         case MSR_PAT:
1390             env->pat = msrs[i].data;
1391             break;
1392         case MSR_STAR:
1393             env->star = msrs[i].data;
1394             break;
1395 #ifdef TARGET_X86_64
1396         case MSR_CSTAR:
1397             env->cstar = msrs[i].data;
1398             break;
1399         case MSR_KERNELGSBASE:
1400             env->kernelgsbase = msrs[i].data;
1401             break;
1402         case MSR_FMASK:
1403             env->fmask = msrs[i].data;
1404             break;
1405         case MSR_LSTAR:
1406             env->lstar = msrs[i].data;
1407             break;
1408 #endif
1409         case MSR_IA32_TSC:
1410             env->tsc = msrs[i].data;
1411             break;
1412         case MSR_TSC_ADJUST:
1413             env->tsc_adjust = msrs[i].data;
1414             break;
1415         case MSR_IA32_TSCDEADLINE:
1416             env->tsc_deadline = msrs[i].data;
1417             break;
1418         case MSR_VM_HSAVE_PA:
1419             env->vm_hsave = msrs[i].data;
1420             break;
1421         case MSR_KVM_SYSTEM_TIME:
1422             env->system_time_msr = msrs[i].data;
1423             break;
1424         case MSR_KVM_WALL_CLOCK:
1425             env->wall_clock_msr = msrs[i].data;
1426             break;
1427         case MSR_MCG_STATUS:
1428             env->mcg_status = msrs[i].data;
1429             break;
1430         case MSR_MCG_CTL:
1431             env->mcg_ctl = msrs[i].data;
1432             break;
1433         case MSR_IA32_MISC_ENABLE:
1434             env->msr_ia32_misc_enable = msrs[i].data;
1435             break;
1436         default:
1437             if (msrs[i].index >= MSR_MC0_CTL &&
1438                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
1439                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
1440             }
1441             break;
1442         case MSR_KVM_ASYNC_PF_EN:
1443             env->async_pf_en_msr = msrs[i].data;
1444             break;
1445         case MSR_KVM_PV_EOI_EN:
1446             env->pv_eoi_en_msr = msrs[i].data;
1447             break;
1448         }
1449     }
1450
1451     return 0;
1452 }
1453
1454 static int kvm_put_mp_state(X86CPU *cpu)
1455 {
1456     struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
1457
1458     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1459 }
1460
1461 static int kvm_get_mp_state(X86CPU *cpu)
1462 {
1463     CPUX86State *env = &cpu->env;
1464     struct kvm_mp_state mp_state;
1465     int ret;
1466
1467     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &mp_state);
1468     if (ret < 0) {
1469         return ret;
1470     }
1471     env->mp_state = mp_state.mp_state;
1472     if (kvm_irqchip_in_kernel()) {
1473         env->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
1474     }
1475     return 0;
1476 }
1477
1478 static int kvm_get_apic(X86CPU *cpu)
1479 {
1480     CPUX86State *env = &cpu->env;
1481     DeviceState *apic = env->apic_state;
1482     struct kvm_lapic_state kapic;
1483     int ret;
1484
1485     if (apic && kvm_irqchip_in_kernel()) {
1486         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
1487         if (ret < 0) {
1488             return ret;
1489         }
1490
1491         kvm_get_apic_state(apic, &kapic);
1492     }
1493     return 0;
1494 }
1495
1496 static int kvm_put_apic(X86CPU *cpu)
1497 {
1498     CPUX86State *env = &cpu->env;
1499     DeviceState *apic = env->apic_state;
1500     struct kvm_lapic_state kapic;
1501
1502     if (apic && kvm_irqchip_in_kernel()) {
1503         kvm_put_apic_state(apic, &kapic);
1504
1505         return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_LAPIC, &kapic);
1506     }
1507     return 0;
1508 }
1509
1510 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
1511 {
1512     CPUX86State *env = &cpu->env;
1513     struct kvm_vcpu_events events;
1514
1515     if (!kvm_has_vcpu_events()) {
1516         return 0;
1517     }
1518
1519     events.exception.injected = (env->exception_injected >= 0);
1520     events.exception.nr = env->exception_injected;
1521     events.exception.has_error_code = env->has_error_code;
1522     events.exception.error_code = env->error_code;
1523     events.exception.pad = 0;
1524
1525     events.interrupt.injected = (env->interrupt_injected >= 0);
1526     events.interrupt.nr = env->interrupt_injected;
1527     events.interrupt.soft = env->soft_interrupt;
1528
1529     events.nmi.injected = env->nmi_injected;
1530     events.nmi.pending = env->nmi_pending;
1531     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
1532     events.nmi.pad = 0;
1533
1534     events.sipi_vector = env->sipi_vector;
1535
1536     events.flags = 0;
1537     if (level >= KVM_PUT_RESET_STATE) {
1538         events.flags |=
1539             KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;
1540     }
1541
1542     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
1543 }
1544
1545 static int kvm_get_vcpu_events(X86CPU *cpu)
1546 {
1547     CPUX86State *env = &cpu->env;
1548     struct kvm_vcpu_events events;
1549     int ret;
1550
1551     if (!kvm_has_vcpu_events()) {
1552         return 0;
1553     }
1554
1555     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
1556     if (ret < 0) {
1557        return ret;
1558     }
1559     env->exception_injected =
1560        events.exception.injected ? events.exception.nr : -1;
1561     env->has_error_code = events.exception.has_error_code;
1562     env->error_code = events.exception.error_code;
1563
1564     env->interrupt_injected =
1565         events.interrupt.injected ? events.interrupt.nr : -1;
1566     env->soft_interrupt = events.interrupt.soft;
1567
1568     env->nmi_injected = events.nmi.injected;
1569     env->nmi_pending = events.nmi.pending;
1570     if (events.nmi.masked) {
1571         env->hflags2 |= HF2_NMI_MASK;
1572     } else {
1573         env->hflags2 &= ~HF2_NMI_MASK;
1574     }
1575
1576     env->sipi_vector = events.sipi_vector;
1577
1578     return 0;
1579 }
1580
1581 static int kvm_guest_debug_workarounds(X86CPU *cpu)
1582 {
1583     CPUX86State *env = &cpu->env;
1584     int ret = 0;
1585     unsigned long reinject_trap = 0;
1586
1587     if (!kvm_has_vcpu_events()) {
1588         if (env->exception_injected == 1) {
1589             reinject_trap = KVM_GUESTDBG_INJECT_DB;
1590         } else if (env->exception_injected == 3) {
1591             reinject_trap = KVM_GUESTDBG_INJECT_BP;
1592         }
1593         env->exception_injected = -1;
1594     }
1595
1596     /*
1597      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
1598      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
1599      * by updating the debug state once again if single-stepping is on.
1600      * Another reason to call kvm_update_guest_debug here is a pending debug
1601      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
1602      * reinject them via SET_GUEST_DEBUG.
1603      */
1604     if (reinject_trap ||
1605         (!kvm_has_robust_singlestep() && env->singlestep_enabled)) {
1606         ret = kvm_update_guest_debug(env, reinject_trap);
1607     }
1608     return ret;
1609 }
1610
1611 static int kvm_put_debugregs(X86CPU *cpu)
1612 {
1613     CPUX86State *env = &cpu->env;
1614     struct kvm_debugregs dbgregs;
1615     int i;
1616
1617     if (!kvm_has_debugregs()) {
1618         return 0;
1619     }
1620
1621     for (i = 0; i < 4; i++) {
1622         dbgregs.db[i] = env->dr[i];
1623     }
1624     dbgregs.dr6 = env->dr[6];
1625     dbgregs.dr7 = env->dr[7];
1626     dbgregs.flags = 0;
1627
1628     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
1629 }
1630
1631 static int kvm_get_debugregs(X86CPU *cpu)
1632 {
1633     CPUX86State *env = &cpu->env;
1634     struct kvm_debugregs dbgregs;
1635     int i, ret;
1636
1637     if (!kvm_has_debugregs()) {
1638         return 0;
1639     }
1640
1641     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
1642     if (ret < 0) {
1643         return ret;
1644     }
1645     for (i = 0; i < 4; i++) {
1646         env->dr[i] = dbgregs.db[i];
1647     }
1648     env->dr[4] = env->dr[6] = dbgregs.dr6;
1649     env->dr[5] = env->dr[7] = dbgregs.dr7;
1650
1651     return 0;
1652 }
1653
1654 int kvm_arch_put_registers(CPUState *cpu, int level)
1655 {
1656     X86CPU *x86_cpu = X86_CPU(cpu);
1657     int ret;
1658
1659     assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
1660
1661     ret = kvm_getput_regs(x86_cpu, 1);
1662     if (ret < 0) {
1663         return ret;
1664     }
1665     ret = kvm_put_xsave(x86_cpu);
1666     if (ret < 0) {
1667         return ret;
1668     }
1669     ret = kvm_put_xcrs(x86_cpu);
1670     if (ret < 0) {
1671         return ret;
1672     }
1673     ret = kvm_put_sregs(x86_cpu);
1674     if (ret < 0) {
1675         return ret;
1676     }
1677     /* must be before kvm_put_msrs */
1678     ret = kvm_inject_mce_oldstyle(x86_cpu);
1679     if (ret < 0) {
1680         return ret;
1681     }
1682     ret = kvm_put_msrs(x86_cpu, level);
1683     if (ret < 0) {
1684         return ret;
1685     }
1686     if (level >= KVM_PUT_RESET_STATE) {
1687         ret = kvm_put_mp_state(x86_cpu);
1688         if (ret < 0) {
1689             return ret;
1690         }
1691         ret = kvm_put_apic(x86_cpu);
1692         if (ret < 0) {
1693             return ret;
1694         }
1695     }
1696     ret = kvm_put_vcpu_events(x86_cpu, level);
1697     if (ret < 0) {
1698         return ret;
1699     }
1700     ret = kvm_put_debugregs(x86_cpu);
1701     if (ret < 0) {
1702         return ret;
1703     }
1704     /* must be last */
1705     ret = kvm_guest_debug_workarounds(x86_cpu);
1706     if (ret < 0) {
1707         return ret;
1708     }
1709     return 0;
1710 }
1711
1712 int kvm_arch_get_registers(CPUState *cs)
1713 {
1714     X86CPU *cpu = X86_CPU(cs);
1715     int ret;
1716
1717     assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
1718
1719     ret = kvm_getput_regs(cpu, 0);
1720     if (ret < 0) {
1721         return ret;
1722     }
1723     ret = kvm_get_xsave(cpu);
1724     if (ret < 0) {
1725         return ret;
1726     }
1727     ret = kvm_get_xcrs(cpu);
1728     if (ret < 0) {
1729         return ret;
1730     }
1731     ret = kvm_get_sregs(cpu);
1732     if (ret < 0) {
1733         return ret;
1734     }
1735     ret = kvm_get_msrs(cpu);
1736     if (ret < 0) {
1737         return ret;
1738     }
1739     ret = kvm_get_mp_state(cpu);
1740     if (ret < 0) {
1741         return ret;
1742     }
1743     ret = kvm_get_apic(cpu);
1744     if (ret < 0) {
1745         return ret;
1746     }
1747     ret = kvm_get_vcpu_events(cpu);
1748     if (ret < 0) {
1749         return ret;
1750     }
1751     ret = kvm_get_debugregs(cpu);
1752     if (ret < 0) {
1753         return ret;
1754     }
1755     return 0;
1756 }
1757
1758 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
1759 {
1760     X86CPU *x86_cpu = X86_CPU(cpu);
1761     CPUX86State *env = &x86_cpu->env;
1762     int ret;
1763
1764     /* Inject NMI */
1765     if (env->interrupt_request & CPU_INTERRUPT_NMI) {
1766         env->interrupt_request &= ~CPU_INTERRUPT_NMI;
1767         DPRINTF("injected NMI\n");
1768         ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
1769         if (ret < 0) {
1770             fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
1771                     strerror(-ret));
1772         }
1773     }
1774
1775     if (!kvm_irqchip_in_kernel()) {
1776         /* Force the VCPU out of its inner loop to process any INIT requests
1777          * or pending TPR access reports. */
1778         if (env->interrupt_request &
1779             (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
1780             cpu->exit_request = 1;
1781         }
1782
1783         /* Try to inject an interrupt if the guest can accept it */
1784         if (run->ready_for_interrupt_injection &&
1785             (env->interrupt_request & CPU_INTERRUPT_HARD) &&
1786             (env->eflags & IF_MASK)) {
1787             int irq;
1788
1789             env->interrupt_request &= ~CPU_INTERRUPT_HARD;
1790             irq = cpu_get_pic_interrupt(env);
1791             if (irq >= 0) {
1792                 struct kvm_interrupt intr;
1793
1794                 intr.irq = irq;
1795                 DPRINTF("injected interrupt %d\n", irq);
1796                 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
1797                 if (ret < 0) {
1798                     fprintf(stderr,
1799                             "KVM: injection failed, interrupt lost (%s)\n",
1800                             strerror(-ret));
1801                 }
1802             }
1803         }
1804
1805         /* If we have an interrupt but the guest is not ready to receive an
1806          * interrupt, request an interrupt window exit.  This will
1807          * cause a return to userspace as soon as the guest is ready to
1808          * receive interrupts. */
1809         if ((env->interrupt_request & CPU_INTERRUPT_HARD)) {
1810             run->request_interrupt_window = 1;
1811         } else {
1812             run->request_interrupt_window = 0;
1813         }
1814
1815         DPRINTF("setting tpr\n");
1816         run->cr8 = cpu_get_apic_tpr(env->apic_state);
1817     }
1818 }
1819
1820 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
1821 {
1822     X86CPU *x86_cpu = X86_CPU(cpu);
1823     CPUX86State *env = &x86_cpu->env;
1824
1825     if (run->if_flag) {
1826         env->eflags |= IF_MASK;
1827     } else {
1828         env->eflags &= ~IF_MASK;
1829     }
1830     cpu_set_apic_tpr(env->apic_state, run->cr8);
1831     cpu_set_apic_base(env->apic_state, run->apic_base);
1832 }
1833
1834 int kvm_arch_process_async_events(CPUState *cs)
1835 {
1836     X86CPU *cpu = X86_CPU(cs);
1837     CPUX86State *env = &cpu->env;
1838
1839     if (env->interrupt_request & CPU_INTERRUPT_MCE) {
1840         /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
1841         assert(env->mcg_cap);
1842
1843         env->interrupt_request &= ~CPU_INTERRUPT_MCE;
1844
1845         kvm_cpu_synchronize_state(env);
1846
1847         if (env->exception_injected == EXCP08_DBLE) {
1848             /* this means triple fault */
1849             qemu_system_reset_request();
1850             cs->exit_request = 1;
1851             return 0;
1852         }
1853         env->exception_injected = EXCP12_MCHK;
1854         env->has_error_code = 0;
1855
1856         env->halted = 0;
1857         if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
1858             env->mp_state = KVM_MP_STATE_RUNNABLE;
1859         }
1860     }
1861
1862     if (kvm_irqchip_in_kernel()) {
1863         return 0;
1864     }
1865
1866     if (env->interrupt_request & CPU_INTERRUPT_POLL) {
1867         env->interrupt_request &= ~CPU_INTERRUPT_POLL;
1868         apic_poll_irq(env->apic_state);
1869     }
1870     if (((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1871          (env->eflags & IF_MASK)) ||
1872         (env->interrupt_request & CPU_INTERRUPT_NMI)) {
1873         env->halted = 0;
1874     }
1875     if (env->interrupt_request & CPU_INTERRUPT_INIT) {
1876         kvm_cpu_synchronize_state(env);
1877         do_cpu_init(cpu);
1878     }
1879     if (env->interrupt_request & CPU_INTERRUPT_SIPI) {
1880         kvm_cpu_synchronize_state(env);
1881         do_cpu_sipi(cpu);
1882     }
1883     if (env->interrupt_request & CPU_INTERRUPT_TPR) {
1884         env->interrupt_request &= ~CPU_INTERRUPT_TPR;
1885         kvm_cpu_synchronize_state(env);
1886         apic_handle_tpr_access_report(env->apic_state, env->eip,
1887                                       env->tpr_access_type);
1888     }
1889
1890     return env->halted;
1891 }
1892
1893 static int kvm_handle_halt(X86CPU *cpu)
1894 {
1895     CPUX86State *env = &cpu->env;
1896
1897     if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1898           (env->eflags & IF_MASK)) &&
1899         !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
1900         env->halted = 1;
1901         return EXCP_HLT;
1902     }
1903
1904     return 0;
1905 }
1906
1907 static int kvm_handle_tpr_access(X86CPU *cpu)
1908 {
1909     CPUX86State *env = &cpu->env;
1910     CPUState *cs = CPU(cpu);
1911     struct kvm_run *run = cs->kvm_run;
1912
1913     apic_handle_tpr_access_report(env->apic_state, run->tpr_access.rip,
1914                                   run->tpr_access.is_write ? TPR_ACCESS_WRITE
1915                                                            : TPR_ACCESS_READ);
1916     return 1;
1917 }
1918
1919 int kvm_arch_insert_sw_breakpoint(CPUState *cpu, struct kvm_sw_breakpoint *bp)
1920 {
1921     CPUX86State *env = &X86_CPU(cpu)->env;
1922     static const uint8_t int3 = 0xcc;
1923
1924     if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
1925         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1)) {
1926         return -EINVAL;
1927     }
1928     return 0;
1929 }
1930
1931 int kvm_arch_remove_sw_breakpoint(CPUState *cpu, struct kvm_sw_breakpoint *bp)
1932 {
1933     CPUX86State *env = &X86_CPU(cpu)->env;
1934     uint8_t int3;
1935
1936     if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
1937         cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
1938         return -EINVAL;
1939     }
1940     return 0;
1941 }
1942
1943 static struct {
1944     target_ulong addr;
1945     int len;
1946     int type;
1947 } hw_breakpoint[4];
1948
1949 static int nb_hw_breakpoint;
1950
1951 static int find_hw_breakpoint(target_ulong addr, int len, int type)
1952 {
1953     int n;
1954
1955     for (n = 0; n < nb_hw_breakpoint; n++) {
1956         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
1957             (hw_breakpoint[n].len == len || len == -1)) {
1958             return n;
1959         }
1960     }
1961     return -1;
1962 }
1963
1964 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1965                                   target_ulong len, int type)
1966 {
1967     switch (type) {
1968     case GDB_BREAKPOINT_HW:
1969         len = 1;
1970         break;
1971     case GDB_WATCHPOINT_WRITE:
1972     case GDB_WATCHPOINT_ACCESS:
1973         switch (len) {
1974         case 1:
1975             break;
1976         case 2:
1977         case 4:
1978         case 8:
1979             if (addr & (len - 1)) {
1980                 return -EINVAL;
1981             }
1982             break;
1983         default:
1984             return -EINVAL;
1985         }
1986         break;
1987     default:
1988         return -ENOSYS;
1989     }
1990
1991     if (nb_hw_breakpoint == 4) {
1992         return -ENOBUFS;
1993     }
1994     if (find_hw_breakpoint(addr, len, type) >= 0) {
1995         return -EEXIST;
1996     }
1997     hw_breakpoint[nb_hw_breakpoint].addr = addr;
1998     hw_breakpoint[nb_hw_breakpoint].len = len;
1999     hw_breakpoint[nb_hw_breakpoint].type = type;
2000     nb_hw_breakpoint++;
2001
2002     return 0;
2003 }
2004
2005 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
2006                                   target_ulong len, int type)
2007 {
2008     int n;
2009
2010     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
2011     if (n < 0) {
2012         return -ENOENT;
2013     }
2014     nb_hw_breakpoint--;
2015     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
2016
2017     return 0;
2018 }
2019
2020 void kvm_arch_remove_all_hw_breakpoints(void)
2021 {
2022     nb_hw_breakpoint = 0;
2023 }
2024
2025 static CPUWatchpoint hw_watchpoint;
2026
2027 static int kvm_handle_debug(X86CPU *cpu,
2028                             struct kvm_debug_exit_arch *arch_info)
2029 {
2030     CPUX86State *env = &cpu->env;
2031     int ret = 0;
2032     int n;
2033
2034     if (arch_info->exception == 1) {
2035         if (arch_info->dr6 & (1 << 14)) {
2036             if (env->singlestep_enabled) {
2037                 ret = EXCP_DEBUG;
2038             }
2039         } else {
2040             for (n = 0; n < 4; n++) {
2041                 if (arch_info->dr6 & (1 << n)) {
2042                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
2043                     case 0x0:
2044                         ret = EXCP_DEBUG;
2045                         break;
2046                     case 0x1:
2047                         ret = EXCP_DEBUG;
2048                         env->watchpoint_hit = &hw_watchpoint;
2049                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
2050                         hw_watchpoint.flags = BP_MEM_WRITE;
2051                         break;
2052                     case 0x3:
2053                         ret = EXCP_DEBUG;
2054                         env->watchpoint_hit = &hw_watchpoint;
2055                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
2056                         hw_watchpoint.flags = BP_MEM_ACCESS;
2057                         break;
2058                     }
2059                 }
2060             }
2061         }
2062     } else if (kvm_find_sw_breakpoint(CPU(cpu), arch_info->pc)) {
2063         ret = EXCP_DEBUG;
2064     }
2065     if (ret == 0) {
2066         cpu_synchronize_state(env);
2067         assert(env->exception_injected == -1);
2068
2069         /* pass to guest */
2070         env->exception_injected = arch_info->exception;
2071         env->has_error_code = 0;
2072     }
2073
2074     return ret;
2075 }
2076
2077 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
2078 {
2079     const uint8_t type_code[] = {
2080         [GDB_BREAKPOINT_HW] = 0x0,
2081         [GDB_WATCHPOINT_WRITE] = 0x1,
2082         [GDB_WATCHPOINT_ACCESS] = 0x3
2083     };
2084     const uint8_t len_code[] = {
2085         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
2086     };
2087     int n;
2088
2089     if (kvm_sw_breakpoints_active(cpu)) {
2090         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
2091     }
2092     if (nb_hw_breakpoint > 0) {
2093         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
2094         dbg->arch.debugreg[7] = 0x0600;
2095         for (n = 0; n < nb_hw_breakpoint; n++) {
2096             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
2097             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
2098                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
2099                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
2100         }
2101     }
2102 }
2103
2104 static bool host_supports_vmx(void)
2105 {
2106     uint32_t ecx, unused;
2107
2108     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
2109     return ecx & CPUID_EXT_VMX;
2110 }
2111
2112 #define VMX_INVALID_GUEST_STATE 0x80000021
2113
2114 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2115 {
2116     X86CPU *cpu = X86_CPU(cs);
2117     uint64_t code;
2118     int ret;
2119
2120     switch (run->exit_reason) {
2121     case KVM_EXIT_HLT:
2122         DPRINTF("handle_hlt\n");
2123         ret = kvm_handle_halt(cpu);
2124         break;
2125     case KVM_EXIT_SET_TPR:
2126         ret = 0;
2127         break;
2128     case KVM_EXIT_TPR_ACCESS:
2129         ret = kvm_handle_tpr_access(cpu);
2130         break;
2131     case KVM_EXIT_FAIL_ENTRY:
2132         code = run->fail_entry.hardware_entry_failure_reason;
2133         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
2134                 code);
2135         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
2136             fprintf(stderr,
2137                     "\nIf you're running a guest on an Intel machine without "
2138                         "unrestricted mode\n"
2139                     "support, the failure can be most likely due to the guest "
2140                         "entering an invalid\n"
2141                     "state for Intel VT. For example, the guest maybe running "
2142                         "in big real mode\n"
2143                     "which is not supported on less recent Intel processors."
2144                         "\n\n");
2145         }
2146         ret = -1;
2147         break;
2148     case KVM_EXIT_EXCEPTION:
2149         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
2150                 run->ex.exception, run->ex.error_code);
2151         ret = -1;
2152         break;
2153     case KVM_EXIT_DEBUG:
2154         DPRINTF("kvm_exit_debug\n");
2155         ret = kvm_handle_debug(cpu, &run->debug.arch);
2156         break;
2157     default:
2158         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
2159         ret = -1;
2160         break;
2161     }
2162
2163     return ret;
2164 }
2165
2166 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
2167 {
2168     X86CPU *cpu = X86_CPU(cs);
2169     CPUX86State *env = &cpu->env;
2170
2171     kvm_cpu_synchronize_state(env);
2172     return !(env->cr[0] & CR0_PE_MASK) ||
2173            ((env->segs[R_CS].selector  & 3) != 3);
2174 }
2175
2176 void kvm_arch_init_irq_routing(KVMState *s)
2177 {
2178     if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2179         /* If kernel can't do irq routing, interrupt source
2180          * override 0->2 cannot be set up as required by HPET.
2181          * So we have to disable it.
2182          */
2183         no_hpet = 1;
2184     }
2185     /* We know at this point that we're using the in-kernel
2186      * irqchip, so we can use irqfds, and on x86 we know
2187      * we can use msi via irqfd and GSI routing.
2188      */
2189     kvm_irqfds_allowed = true;
2190     kvm_msi_via_irqfd_allowed = true;
2191     kvm_gsi_routing_allowed = true;
2192 }
2193
2194 /* Classic KVM device assignment interface. Will remain x86 only. */
2195 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
2196                           uint32_t flags, uint32_t *dev_id)
2197 {
2198     struct kvm_assigned_pci_dev dev_data = {
2199         .segnr = dev_addr->domain,
2200         .busnr = dev_addr->bus,
2201         .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
2202         .flags = flags,
2203     };
2204     int ret;
2205
2206     dev_data.assigned_dev_id =
2207         (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
2208
2209     ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
2210     if (ret < 0) {
2211         return ret;
2212     }
2213
2214     *dev_id = dev_data.assigned_dev_id;
2215
2216     return 0;
2217 }
2218
2219 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
2220 {
2221     struct kvm_assigned_pci_dev dev_data = {
2222         .assigned_dev_id = dev_id,
2223     };
2224
2225     return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
2226 }
2227
2228 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
2229                                    uint32_t irq_type, uint32_t guest_irq)
2230 {
2231     struct kvm_assigned_irq assigned_irq = {
2232         .assigned_dev_id = dev_id,
2233         .guest_irq = guest_irq,
2234         .flags = irq_type,
2235     };
2236
2237     if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
2238         return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
2239     } else {
2240         return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
2241     }
2242 }
2243
2244 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
2245                            uint32_t guest_irq)
2246 {
2247     uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
2248         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
2249
2250     return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
2251 }
2252
2253 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
2254 {
2255     struct kvm_assigned_pci_dev dev_data = {
2256         .assigned_dev_id = dev_id,
2257         .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
2258     };
2259
2260     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
2261 }
2262
2263 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
2264                                      uint32_t type)
2265 {
2266     struct kvm_assigned_irq assigned_irq = {
2267         .assigned_dev_id = dev_id,
2268         .flags = type,
2269     };
2270
2271     return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
2272 }
2273
2274 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
2275 {
2276     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
2277         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
2278 }
2279
2280 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
2281 {
2282     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
2283                                               KVM_DEV_IRQ_GUEST_MSI, virq);
2284 }
2285
2286 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
2287 {
2288     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
2289                                                 KVM_DEV_IRQ_HOST_MSI);
2290 }
2291
2292 bool kvm_device_msix_supported(KVMState *s)
2293 {
2294     /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
2295      * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
2296     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
2297 }
2298
2299 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
2300                                  uint32_t nr_vectors)
2301 {
2302     struct kvm_assigned_msix_nr msix_nr = {
2303         .assigned_dev_id = dev_id,
2304         .entry_nr = nr_vectors,
2305     };
2306
2307     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
2308 }
2309
2310 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
2311                                int virq)
2312 {
2313     struct kvm_assigned_msix_entry msix_entry = {
2314         .assigned_dev_id = dev_id,
2315         .gsi = virq,
2316         .entry = vector,
2317     };
2318
2319     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
2320 }
2321
2322 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
2323 {
2324     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
2325                                               KVM_DEV_IRQ_GUEST_MSIX, 0);
2326 }
2327
2328 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
2329 {
2330     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
2331                                                 KVM_DEV_IRQ_HOST_MSIX);
2332 }