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 "qemu/osdep.h"
  16 #include "qapi/error.h"
  17 #include <sys/ioctl.h>
  18 #include <sys/utsname.h>
  19
  20 #include <linux/kvm.h>
  21 #include "standard-headers/asm-x86/kvm_para.h"
  22
  23 #include "qemu-common.h"
  24 #include "cpu.h"
  25 #include "sysemu/sysemu.h"
  26 #include "sysemu/hw_accel.h"
  27 #include "sysemu/kvm_int.h"
  28 #include "kvm_i386.h"
  29 #include "hyperv.h"
  30 #include "hyperv-proto.h"
  31
  32 #include "exec/gdbstub.h"
  33 #include "qemu/host-utils.h"
  34 #include "qemu/config-file.h"
  35 #include "qemu/error-report.h"
  36 #include "hw/i386/pc.h"
  37 #include "hw/i386/apic.h"
  38 #include "hw/i386/apic_internal.h"
  39 #include "hw/i386/apic-msidef.h"
  40 #include "hw/i386/intel_iommu.h"
  41 #include "hw/i386/x86-iommu.h"
  42
  43 #include "hw/pci/pci.h"
  44 #include "hw/pci/msi.h"
  45 #include "hw/pci/msix.h"
  46 #include "migration/blocker.h"
  47 #include "exec/memattrs.h"
  48 #include "trace.h"
  49
  50 //#define DEBUG_KVM
  51
  52 #ifdef DEBUG_KVM
  53 #define DPRINTF(fmt, ...) \
  54     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  55 #else
  56 #define DPRINTF(fmt, ...) \
  57     do { } while (0)
  58 #endif
  59
  60 #define MSR_KVM_WALL_CLOCK  0x11
  61 #define MSR_KVM_SYSTEM_TIME 0x12
  62
  63 /* A 4096-byte buffer can hold the 8-byte kvm_msrs header, plus
  64  * 255 kvm_msr_entry structs */
  65 #define MSR_BUF_SIZE 4096
  66
  67 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  68     KVM_CAP_INFO(SET_TSS_ADDR),
  69     KVM_CAP_INFO(EXT_CPUID),
  70     KVM_CAP_INFO(MP_STATE),
  71     KVM_CAP_LAST_INFO
  72 };
  73
  74 static bool has_msr_star;
  75 static bool has_msr_hsave_pa;
  76 static bool has_msr_tsc_aux;
  77 static bool has_msr_tsc_adjust;
  78 static bool has_msr_tsc_deadline;
  79 static bool has_msr_feature_control;
  80 static bool has_msr_misc_enable;
  81 static bool has_msr_smbase;
  82 static bool has_msr_bndcfgs;
  83 static int lm_capable_kernel;
  84 static bool has_msr_hv_hypercall;
  85 static bool has_msr_hv_crash;
  86 static bool has_msr_hv_reset;
  87 static bool has_msr_hv_vpindex;
  88 static bool hv_vpindex_settable;
  89 static bool has_msr_hv_runtime;
  90 static bool has_msr_hv_synic;
  91 static bool has_msr_hv_stimer;
  92 static bool has_msr_hv_frequencies;
  93 static bool has_msr_hv_reenlightenment;
  94 static bool has_msr_xss;
  95 static bool has_msr_spec_ctrl;
  96 static bool has_msr_virt_ssbd;
  97 static bool has_msr_smi_count;
  98
  99 static uint32_t has_architectural_pmu_version;
 100 static uint32_t num_architectural_pmu_gp_counters;
 101 static uint32_t num_architectural_pmu_fixed_counters;
 102
 103 static int has_xsave;
 104 static int has_xcrs;
 105 static int has_pit_state2;
 106
 107 static bool has_msr_mcg_ext_ctl;
 108
 109 static struct kvm_cpuid2 *cpuid_cache;
 110
 111 int kvm_has_pit_state2(void)
 112 {
 113     return has_pit_state2;
 114 }
 115
 116 bool kvm_has_smm(void)
 117 {
 118     return kvm_check_extension(kvm_state, KVM_CAP_X86_SMM);
 119 }
 120
 121 bool kvm_has_adjust_clock_stable(void)
 122 {
 123     int ret = kvm_check_extension(kvm_state, KVM_CAP_ADJUST_CLOCK);
 124
 125     return (ret == KVM_CLOCK_TSC_STABLE);
 126 }
 127
 128 bool kvm_allows_irq0_override(void)
 129 {
 130     return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
 131 }
 132
 133 static bool kvm_x2apic_api_set_flags(uint64_t flags)
 134 {
 135     KVMState *s = KVM_STATE(current_machine->accelerator);
 136
 137     return !kvm_vm_enable_cap(s, KVM_CAP_X2APIC_API, 0, flags);
 138 }
 139
 140 #define MEMORIZE(fn, _result) \
 141     ({ \
 142         static bool _memorized; \
 143         \
 144         if (_memorized) { \
 145             return _result; \
 146         } \
 147         _memorized = true; \
 148         _result = fn; \
 149     })
 150
 151 static bool has_x2apic_api;
 152
 153 bool kvm_has_x2apic_api(void)
 154 {
 155     return has_x2apic_api;
 156 }
 157
 158 bool kvm_enable_x2apic(void)
 159 {
 160     return MEMORIZE(
 161              kvm_x2apic_api_set_flags(KVM_X2APIC_API_USE_32BIT_IDS |
 162                                       KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK),
 163              has_x2apic_api);
 164 }
 165
 166 bool kvm_hv_vpindex_settable(void)
 167 {
 168     return hv_vpindex_settable;
 169 }
 170
 171 static int kvm_get_tsc(CPUState *cs)
 172 {
 173     X86CPU *cpu = X86_CPU(cs);
 174     CPUX86State *env = &cpu->env;
 175     struct {
 176         struct kvm_msrs info;
 177         struct kvm_msr_entry entries[1];
 178     } msr_data;
 179     int ret;
 180
 181     if (env->tsc_valid) {
 182         return 0;
 183     }
 184
 185     msr_data.info.nmsrs = 1;
 186     msr_data.entries[0].index = MSR_IA32_TSC;
 187     env->tsc_valid = !runstate_is_running();
 188
 189     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
 190     if (ret < 0) {
 191         return ret;
 192     }
 193
 194     assert(ret == 1);
 195     env->tsc = msr_data.entries[0].data;
 196     return 0;
 197 }
 198
 199 static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
 200 {
 201     kvm_get_tsc(cpu);
 202 }
 203
 204 void kvm_synchronize_all_tsc(void)
 205 {
 206     CPUState *cpu;
 207
 208     if (kvm_enabled()) {
 209         CPU_FOREACH(cpu) {
 210             run_on_cpu(cpu, do_kvm_synchronize_tsc, RUN_ON_CPU_NULL);
 211         }
 212     }
 213 }
 214
 215 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
 216 {
 217     struct kvm_cpuid2 *cpuid;
 218     int r, size;
 219
 220     size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
 221     cpuid = g_malloc0(size);
 222     cpuid->nent = max;
 223     r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
 224     if (r == 0 && cpuid->nent >= max) {
 225         r = -E2BIG;
 226     }
 227     if (r < 0) {
 228         if (r == -E2BIG) {
 229             g_free(cpuid);
 230             return NULL;
 231         } else {
 232             fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
 233                     strerror(-r));
 234             exit(1);
 235         }
 236     }
 237     return cpuid;
 238 }
 239
 240 /* Run KVM_GET_SUPPORTED_CPUID ioctl(), allocating a buffer large enough
 241  * for all entries.
 242  */
 243 static struct kvm_cpuid2 *get_supported_cpuid(KVMState *s)
 244 {
 245     struct kvm_cpuid2 *cpuid;
 246     int max = 1;
 247
 248     if (cpuid_cache != NULL) {
 249         return cpuid_cache;
 250     }
 251     while ((cpuid = try_get_cpuid(s, max)) == NULL) {
 252         max *= 2;
 253     }
 254     cpuid_cache = cpuid;
 255     return cpuid;
 256 }
 257
 258 static const struct kvm_para_features {
 259     int cap;
 260     int feature;
 261 } para_features[] = {
 262     { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
 263     { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
 264     { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 265     { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
 266 };
 267
 268 static int get_para_features(KVMState *s)
 269 {
 270     int i, features = 0;
 271
 272     for (i = 0; i < ARRAY_SIZE(para_features); i++) {
 273         if (kvm_check_extension(s, para_features[i].cap)) {
 274             features |= (1 << para_features[i].feature);
 275         }
 276     }
 277
 278     return features;
 279 }
 280
 281 static bool host_tsx_blacklisted(void)
 282 {
 283     int family, model, stepping;\
 284     char vendor[CPUID_VENDOR_SZ + 1];
 285
 286     host_vendor_fms(vendor, &family, &model, &stepping);
 287
 288     /* Check if we are running on a Haswell host known to have broken TSX */
 289     return !strcmp(vendor, CPUID_VENDOR_INTEL) &&
 290            (family == 6) &&
 291            ((model == 63 && stepping < 4) ||
 292             model == 60 || model == 69 || model == 70);
 293 }
 294
 295 /* Returns the value for a specific register on the cpuid entry
 296  */
 297 static uint32_t cpuid_entry_get_reg(struct kvm_cpuid_entry2 *entry, int reg)
 298 {
 299     uint32_t ret = 0;
 300     switch (reg) {
 301     case R_EAX:
 302         ret = entry->eax;
 303         break;
 304     case R_EBX:
 305         ret = entry->ebx;
 306         break;
 307     case R_ECX:
 308         ret = entry->ecx;
 309         break;
 310     case R_EDX:
 311         ret = entry->edx;
 312         break;
 313     }
 314     return ret;
 315 }
 316
 317 /* Find matching entry for function/index on kvm_cpuid2 struct
 318  */
 319 static struct kvm_cpuid_entry2 *cpuid_find_entry(struct kvm_cpuid2 *cpuid,
 320                                                  uint32_t function,
 321                                                  uint32_t index)
 322 {
 323     int i;
 324     for (i = 0; i < cpuid->nent; ++i) {
 325         if (cpuid->entries[i].function == function &&
 326             cpuid->entries[i].index == index) {
 327             return &cpuid->entries[i];
 328         }
 329     }
 330     /* not found: */
 331     return NULL;
 332 }
 333
 334 uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
 335                                       uint32_t index, int reg)
 336 {
 337     struct kvm_cpuid2 *cpuid;
 338     uint32_t ret = 0;
 339     uint32_t cpuid_1_edx;
 340     bool found = false;
 341
 342     cpuid = get_supported_cpuid(s);
 343
 344     struct kvm_cpuid_entry2 *entry = cpuid_find_entry(cpuid, function, index);
 345     if (entry) {
 346         found = true;
 347         ret = cpuid_entry_get_reg(entry, reg);
 348     }
 349
 350     /* Fixups for the data returned by KVM, below */
 351
 352     if (function == 1 && reg == R_EDX) {
 353         /* KVM before 2.6.30 misreports the following features */
 354         ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
 355     } else if (function == 1 && reg == R_ECX) {
 356         /* We can set the hypervisor flag, even if KVM does not return it on
 357          * GET_SUPPORTED_CPUID
 358          */
 359         ret |= CPUID_EXT_HYPERVISOR;
 360         /* tsc-deadline flag is not returned by GET_SUPPORTED_CPUID, but it
 361          * can be enabled if the kernel has KVM_CAP_TSC_DEADLINE_TIMER,
 362          * and the irqchip is in the kernel.
 363          */
 364         if (kvm_irqchip_in_kernel() &&
 365                 kvm_check_extension(s, KVM_CAP_TSC_DEADLINE_TIMER)) {
 366             ret |= CPUID_EXT_TSC_DEADLINE_TIMER;
 367         }
 368
 369         /* x2apic is reported by GET_SUPPORTED_CPUID, but it can't be enabled
 370          * without the in-kernel irqchip
 371          */
 372         if (!kvm_irqchip_in_kernel()) {
 373             ret &= ~CPUID_EXT_X2APIC;
 374         }
 375
 376         if (enable_cpu_pm) {
 377             int disable_exits = kvm_check_extension(s,
 378                                                     KVM_CAP_X86_DISABLE_EXITS);
 379
 380             if (disable_exits & KVM_X86_DISABLE_EXITS_MWAIT) {
 381                 ret |= CPUID_EXT_MONITOR;
 382             }
 383         }
 384     } else if (function == 6 && reg == R_EAX) {
 385         ret |= CPUID_6_EAX_ARAT; /* safe to allow because of emulated APIC */
 386     } else if (function == 7 && index == 0 && reg == R_EBX) {
 387         if (host_tsx_blacklisted()) {
 388             ret &= ~(CPUID_7_0_EBX_RTM | CPUID_7_0_EBX_HLE);
 389         }
 390     } else if (function == 0x80000001 && reg == R_ECX) {
 391         /*
 392          * It's safe to enable TOPOEXT even if it's not returned by
 393          * GET_SUPPORTED_CPUID.  Unconditionally enabling TOPOEXT here allows
 394          * us to keep CPU models including TOPOEXT runnable on older kernels.
 395          */
 396         ret |= CPUID_EXT3_TOPOEXT;
 397     } else if (function == 0x80000001 && reg == R_EDX) {
 398         /* On Intel, kvm returns cpuid according to the Intel spec,
 399          * so add missing bits according to the AMD spec:
 400          */
 401         cpuid_1_edx = kvm_arch_get_supported_cpuid(s, 1, 0, R_EDX);
 402         ret |= cpuid_1_edx & CPUID_EXT2_AMD_ALIASES;
 403     } else if (function == KVM_CPUID_FEATURES && reg == R_EAX) {
 404         /* kvm_pv_unhalt is reported by GET_SUPPORTED_CPUID, but it can't
 405          * be enabled without the in-kernel irqchip
 406          */
 407         if (!kvm_irqchip_in_kernel()) {
 408             ret &= ~(1U << KVM_FEATURE_PV_UNHALT);
 409         }
 410     } else if (function == KVM_CPUID_FEATURES && reg == R_EDX) {
 411         ret |= 1U << KVM_HINTS_REALTIME;
 412         found = 1;
 413     }
 414
 415     /* fallback for older kernels */
 416     if ((function == KVM_CPUID_FEATURES) && !found) {
 417         ret = get_para_features(s);
 418     }
 419
 420     return ret;
 421 }
 422
 423 typedef struct HWPoisonPage {
 424     ram_addr_t ram_addr;
 425     QLIST_ENTRY(HWPoisonPage) list;
 426 } HWPoisonPage;
 427
 428 static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list =
 429     QLIST_HEAD_INITIALIZER(hwpoison_page_list);
 430
 431 static void kvm_unpoison_all(void *param)
 432 {
 433     HWPoisonPage *page, *next_page;
 434
 435     QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) {
 436         QLIST_REMOVE(page, list);
 437         qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE);
 438         g_free(page);
 439     }
 440 }
 441
 442 static void kvm_hwpoison_page_add(ram_addr_t ram_addr)
 443 {
 444     HWPoisonPage *page;
 445
 446     QLIST_FOREACH(page, &hwpoison_page_list, list) {
 447         if (page->ram_addr == ram_addr) {
 448             return;
 449         }
 450     }
 451     page = g_new(HWPoisonPage, 1);
 452     page->ram_addr = ram_addr;
 453     QLIST_INSERT_HEAD(&hwpoison_page_list, page, list);
 454 }
 455
 456 static int kvm_get_mce_cap_supported(KVMState *s, uint64_t *mce_cap,
 457                                      int *max_banks)
 458 {
 459     int r;
 460
 461     r = kvm_check_extension(s, KVM_CAP_MCE);
 462     if (r > 0) {
 463         *max_banks = r;
 464         return kvm_ioctl(s, KVM_X86_GET_MCE_CAP_SUPPORTED, mce_cap);
 465     }
 466     return -ENOSYS;
 467 }
 468
 469 static void kvm_mce_inject(X86CPU *cpu, hwaddr paddr, int code)
 470 {
 471     CPUState *cs = CPU(cpu);
 472     CPUX86State *env = &cpu->env;
 473     uint64_t status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN |
 474                       MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S;
 475     uint64_t mcg_status = MCG_STATUS_MCIP;
 476     int flags = 0;
 477
 478     if (code == BUS_MCEERR_AR) {
 479         status |= MCI_STATUS_AR | 0x134;
 480         mcg_status |= MCG_STATUS_EIPV;
 481     } else {
 482         status |= 0xc0;
 483         mcg_status |= MCG_STATUS_RIPV;
 484     }
 485
 486     flags = cpu_x86_support_mca_broadcast(env) ? MCE_INJECT_BROADCAST : 0;
 487     /* We need to read back the value of MSR_EXT_MCG_CTL that was set by the
 488      * guest kernel back into env->mcg_ext_ctl.
 489      */
 490     cpu_synchronize_state(cs);
 491     if (env->mcg_ext_ctl & MCG_EXT_CTL_LMCE_EN) {
 492         mcg_status |= MCG_STATUS_LMCE;
 493         flags = 0;
 494     }
 495
 496     cpu_x86_inject_mce(NULL, cpu, 9, status, mcg_status, paddr,
 497                        (MCM_ADDR_PHYS << 6) | 0xc, flags);
 498 }
 499
 500 static void hardware_memory_error(void)
 501 {
 502     fprintf(stderr, "Hardware memory error!\n");
 503     exit(1);
 504 }
 505
 506 void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
 507 {
 508     X86CPU *cpu = X86_CPU(c);
 509     CPUX86State *env = &cpu->env;
 510     ram_addr_t ram_addr;
 511     hwaddr paddr;
 512
 513     /* If we get an action required MCE, it has been injected by KVM
 514      * while the VM was running.  An action optional MCE instead should
 515      * be coming from the main thread, which qemu_init_sigbus identifies
 516      * as the "early kill" thread.
 517      */
 518     assert(code == BUS_MCEERR_AR || code == BUS_MCEERR_AO);
 519
 520     if ((env->mcg_cap & MCG_SER_P) && addr) {
 521         ram_addr = qemu_ram_addr_from_host(addr);
 522         if (ram_addr != RAM_ADDR_INVALID &&
 523             kvm_physical_memory_addr_from_host(c->kvm_state, addr, &paddr)) {
 524             kvm_hwpoison_page_add(ram_addr);
 525             kvm_mce_inject(cpu, paddr, code);
 526             return;
 527         }
 528
 529         fprintf(stderr, "Hardware memory error for memory used by "
 530                 "QEMU itself instead of guest system!\n");
 531     }
 532
 533     if (code == BUS_MCEERR_AR) {
 534         hardware_memory_error();
 535     }
 536
 537     /* Hope we are lucky for AO MCE */
 538 }
 539
 540 static int kvm_inject_mce_oldstyle(X86CPU *cpu)
 541 {
 542     CPUX86State *env = &cpu->env;
 543
 544     if (!kvm_has_vcpu_events() && env->exception_injected == EXCP12_MCHK) {
 545         unsigned int bank, bank_num = env->mcg_cap & 0xff;
 546         struct kvm_x86_mce mce;
 547
 548         env->exception_injected = -1;
 549
 550         /*
 551          * There must be at least one bank in use if an MCE is pending.
 552          * Find it and use its values for the event injection.
 553          */
 554         for (bank = 0; bank < bank_num; bank++) {
 555             if (env->mce_banks[bank * 4 + 1] & MCI_STATUS_VAL) {
 556                 break;
 557             }
 558         }
 559         assert(bank < bank_num);
 560
 561         mce.bank = bank;
 562         mce.status = env->mce_banks[bank * 4 + 1];
 563         mce.mcg_status = env->mcg_status;
 564         mce.addr = env->mce_banks[bank * 4 + 2];
 565         mce.misc = env->mce_banks[bank * 4 + 3];
 566
 567         return kvm_vcpu_ioctl(CPU(cpu), KVM_X86_SET_MCE, &mce);
 568     }
 569     return 0;
 570 }
 571
 572 static void cpu_update_state(void *opaque, int running, RunState state)
 573 {
 574     CPUX86State *env = opaque;
 575
 576     if (running) {
 577         env->tsc_valid = false;
 578     }
 579 }
 580
 581 unsigned long kvm_arch_vcpu_id(CPUState *cs)
 582 {
 583     X86CPU *cpu = X86_CPU(cs);
 584     return cpu->apic_id;
 585 }
 586
 587 #ifndef KVM_CPUID_SIGNATURE_NEXT
 588 #define KVM_CPUID_SIGNATURE_NEXT                0x40000100
 589 #endif
 590
 591 static bool hyperv_hypercall_available(X86CPU *cpu)
 592 {
 593     return cpu->hyperv_vapic ||
 594            (cpu->hyperv_spinlock_attempts != HYPERV_SPINLOCK_NEVER_RETRY);
 595 }
 596
 597 static bool hyperv_enabled(X86CPU *cpu)
 598 {
 599     CPUState *cs = CPU(cpu);
 600     return kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV) > 0 &&
 601            (hyperv_hypercall_available(cpu) ||
 602             cpu->hyperv_time  ||
 603             cpu->hyperv_relaxed_timing ||
 604             cpu->hyperv_crash ||
 605             cpu->hyperv_reset ||
 606             cpu->hyperv_vpindex ||
 607             cpu->hyperv_runtime ||
 608             cpu->hyperv_synic ||
 609             cpu->hyperv_stimer ||
 610             cpu->hyperv_reenlightenment ||
 611             cpu->hyperv_tlbflush ||
 612             cpu->hyperv_ipi);
 613 }
 614
 615 static int kvm_arch_set_tsc_khz(CPUState *cs)
 616 {
 617     X86CPU *cpu = X86_CPU(cs);
 618     CPUX86State *env = &cpu->env;
 619     int r;
 620
 621     if (!env->tsc_khz) {
 622         return 0;
 623     }
 624
 625     r = kvm_check_extension(cs->kvm_state, KVM_CAP_TSC_CONTROL) ?
 626         kvm_vcpu_ioctl(cs, KVM_SET_TSC_KHZ, env->tsc_khz) :
 627         -ENOTSUP;
 628     if (r < 0) {
 629         /* When KVM_SET_TSC_KHZ fails, it's an error only if the current
 630          * TSC frequency doesn't match the one we want.
 631          */
 632         int cur_freq = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
 633                        kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
 634                        -ENOTSUP;
 635         if (cur_freq <= 0 || cur_freq != env->tsc_khz) {
 636             warn_report("TSC frequency mismatch between "
 637                         "VM (%" PRId64 " kHz) and host (%d kHz), "
 638                         "and TSC scaling unavailable",
 639                         env->tsc_khz, cur_freq);
 640             return r;
 641         }
 642     }
 643
 644     return 0;
 645 }
 646
 647 static bool tsc_is_stable_and_known(CPUX86State *env)
 648 {
 649     if (!env->tsc_khz) {
 650         return false;
 651     }
 652     return (env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC)
 653         || env->user_tsc_khz;
 654 }
 655
 656 static int hyperv_handle_properties(CPUState *cs)
 657 {
 658     X86CPU *cpu = X86_CPU(cs);
 659     CPUX86State *env = &cpu->env;
 660
 661     if (cpu->hyperv_relaxed_timing) {
 662         env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
 663     }
 664     if (cpu->hyperv_vapic) {
 665         env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
 666         env->features[FEAT_HYPERV_EAX] |= HV_APIC_ACCESS_AVAILABLE;
 667     }
 668     if (cpu->hyperv_time) {
 669         if (kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_TIME) <= 0) {
 670             fprintf(stderr, "Hyper-V clocksources "
 671                     "(requested by 'hv-time' cpu flag) "
 672                     "are not supported by kernel\n");
 673             return -ENOSYS;
 674         }
 675         env->features[FEAT_HYPERV_EAX] |= HV_HYPERCALL_AVAILABLE;
 676         env->features[FEAT_HYPERV_EAX] |= HV_TIME_REF_COUNT_AVAILABLE;
 677         env->features[FEAT_HYPERV_EAX] |= HV_REFERENCE_TSC_AVAILABLE;
 678     }
 679     if (cpu->hyperv_frequencies) {
 680         if (!has_msr_hv_frequencies) {
 681             fprintf(stderr, "Hyper-V frequency MSRs "
 682                     "(requested by 'hv-frequencies' cpu flag) "
 683                     "are not supported by kernel\n");
 684             return -ENOSYS;
 685         }
 686         env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_FREQUENCY_MSRS;
 687         env->features[FEAT_HYPERV_EDX] |= HV_FREQUENCY_MSRS_AVAILABLE;
 688     }
 689     if (cpu->hyperv_crash) {
 690         if (!has_msr_hv_crash) {
 691             fprintf(stderr, "Hyper-V crash MSRs "
 692                     "(requested by 'hv-crash' cpu flag) "
 693                     "are not supported by kernel\n");
 694             return -ENOSYS;
 695         }
 696         env->features[FEAT_HYPERV_EDX] |= HV_GUEST_CRASH_MSR_AVAILABLE;
 697     }
 698     if (cpu->hyperv_reenlightenment) {
 699         if (!has_msr_hv_reenlightenment) {
 700             fprintf(stderr,
 701                     "Hyper-V Reenlightenment MSRs "
 702                     "(requested by 'hv-reenlightenment' cpu flag) "
 703                     "are not supported by kernel\n");
 704             return -ENOSYS;
 705         }
 706         env->features[FEAT_HYPERV_EAX] |= HV_ACCESS_REENLIGHTENMENTS_CONTROL;
 707     }
 708     env->features[FEAT_HYPERV_EDX] |= HV_CPU_DYNAMIC_PARTITIONING_AVAILABLE;
 709     if (cpu->hyperv_reset) {
 710         if (!has_msr_hv_reset) {
 711             fprintf(stderr, "Hyper-V reset MSR "
 712                     "(requested by 'hv-reset' cpu flag) "
 713                     "is not supported by kernel\n");
 714             return -ENOSYS;
 715         }
 716         env->features[FEAT_HYPERV_EAX] |= HV_RESET_AVAILABLE;
 717     }
 718     if (cpu->hyperv_vpindex) {
 719         if (!has_msr_hv_vpindex) {
 720             fprintf(stderr, "Hyper-V VP_INDEX MSR "
 721                     "(requested by 'hv-vpindex' cpu flag) "
 722                     "is not supported by kernel\n");
 723             return -ENOSYS;
 724         }
 725         env->features[FEAT_HYPERV_EAX] |= HV_VP_INDEX_AVAILABLE;
 726     }
 727     if (cpu->hyperv_runtime) {
 728         if (!has_msr_hv_runtime) {
 729             fprintf(stderr, "Hyper-V VP_RUNTIME MSR "
 730                     "(requested by 'hv-runtime' cpu flag) "
 731                     "is not supported by kernel\n");
 732             return -ENOSYS;
 733         }
 734         env->features[FEAT_HYPERV_EAX] |= HV_VP_RUNTIME_AVAILABLE;
 735     }
 736     if (cpu->hyperv_synic) {
 737         unsigned int cap = KVM_CAP_HYPERV_SYNIC;
 738         if (!cpu->hyperv_synic_kvm_only) {
 739             if (!cpu->hyperv_vpindex) {
 740                 fprintf(stderr, "Hyper-V SynIC "
 741                         "(requested by 'hv-synic' cpu flag) "
 742                         "requires Hyper-V VP_INDEX ('hv-vpindex')\n");
 743             return -ENOSYS;
 744             }
 745             cap = KVM_CAP_HYPERV_SYNIC2;
 746         }
 747
 748         if (!has_msr_hv_synic || !kvm_check_extension(cs->kvm_state, cap)) {
 749             fprintf(stderr, "Hyper-V SynIC (requested by 'hv-synic' cpu flag) "
 750                     "is not supported by kernel\n");
 751             return -ENOSYS;
 752         }
 753
 754         env->features[FEAT_HYPERV_EAX] |= HV_SYNIC_AVAILABLE;
 755     }
 756     if (cpu->hyperv_stimer) {
 757         if (!has_msr_hv_stimer) {
 758             fprintf(stderr, "Hyper-V timers aren't supported by kernel\n");
 759             return -ENOSYS;
 760         }
 761         env->features[FEAT_HYPERV_EAX] |= HV_SYNTIMERS_AVAILABLE;
 762     }
 763     return 0;
 764 }
 765
 766 static int hyperv_init_vcpu(X86CPU *cpu)
 767 {
 768     CPUState *cs = CPU(cpu);
 769     int ret;
 770
 771     if (cpu->hyperv_vpindex && !hv_vpindex_settable) {
 772         /*
 773          * the kernel doesn't support setting vp_index; assert that its value
 774          * is in sync
 775          */
 776         struct {
 777             struct kvm_msrs info;
 778             struct kvm_msr_entry entries[1];
 779         } msr_data = {
 780             .info.nmsrs = 1,
 781             .entries[0].index = HV_X64_MSR_VP_INDEX,
 782         };
 783
 784         ret = kvm_vcpu_ioctl(cs, KVM_GET_MSRS, &msr_data);
 785         if (ret < 0) {
 786             return ret;
 787         }
 788         assert(ret == 1);
 789
 790         if (msr_data.entries[0].data != hyperv_vp_index(CPU(cpu))) {
 791             error_report("kernel's vp_index != QEMU's vp_index");
 792             return -ENXIO;
 793         }
 794     }
 795
 796     if (cpu->hyperv_synic) {
 797         uint32_t synic_cap = cpu->hyperv_synic_kvm_only ?
 798             KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2;
 799         ret = kvm_vcpu_enable_cap(cs, synic_cap, 0);
 800         if (ret < 0) {
 801             error_report("failed to turn on HyperV SynIC in KVM: %s",
 802                          strerror(-ret));
 803             return ret;
 804         }
 805
 806         if (!cpu->hyperv_synic_kvm_only) {
 807             ret = hyperv_x86_synic_add(cpu);
 808             if (ret < 0) {
 809                 error_report("failed to create HyperV SynIC: %s",
 810                              strerror(-ret));
 811                 return ret;
 812             }
 813         }
 814     }
 815
 816     return 0;
 817 }
 818
 819 static Error *invtsc_mig_blocker;
 820
 821 #define KVM_MAX_CPUID_ENTRIES  100
 822
 823 int kvm_arch_init_vcpu(CPUState *cs)
 824 {
 825     struct {
 826         struct kvm_cpuid2 cpuid;
 827         struct kvm_cpuid_entry2 entries[KVM_MAX_CPUID_ENTRIES];
 828     } QEMU_PACKED cpuid_data;
 829     X86CPU *cpu = X86_CPU(cs);
 830     CPUX86State *env = &cpu->env;
 831     uint32_t limit, i, j, cpuid_i;
 832     uint32_t unused;
 833     struct kvm_cpuid_entry2 *c;
 834     uint32_t signature[3];
 835     int kvm_base = KVM_CPUID_SIGNATURE;
 836     int r;
 837     Error *local_err = NULL;
 838
 839     memset(&cpuid_data, 0, sizeof(cpuid_data));
 840
 841     cpuid_i = 0;
 842
 843     r = kvm_arch_set_tsc_khz(cs);
 844     if (r < 0) {
 845         goto fail;
 846     }
 847
 848     /* vcpu's TSC frequency is either specified by user, or following
 849      * the value used by KVM if the former is not present. In the
 850      * latter case, we query it from KVM and record in env->tsc_khz,
 851      * so that vcpu's TSC frequency can be migrated later via this field.
 852      */
 853     if (!env->tsc_khz) {
 854         r = kvm_check_extension(cs->kvm_state, KVM_CAP_GET_TSC_KHZ) ?
 855             kvm_vcpu_ioctl(cs, KVM_GET_TSC_KHZ) :
 856             -ENOTSUP;
 857         if (r > 0) {
 858             env->tsc_khz = r;
 859         }
 860     }
 861
 862     /* Paravirtualization CPUIDs */
 863     if (hyperv_enabled(cpu)) {
 864         c = &cpuid_data.entries[cpuid_i++];
 865         c->function = HV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
 866         if (!cpu->hyperv_vendor_id) {
 867             memcpy(signature, "Microsoft Hv", 12);
 868         } else {
 869             size_t len = strlen(cpu->hyperv_vendor_id);
 870
 871             if (len > 12) {
 872                 error_report("hv-vendor-id truncated to 12 characters");
 873                 len = 12;
 874             }
 875             memset(signature, 0, 12);
 876             memcpy(signature, cpu->hyperv_vendor_id, len);
 877         }
 878         c->eax = HV_CPUID_MIN;
 879         c->ebx = signature[0];
 880         c->ecx = signature[1];
 881         c->edx = signature[2];
 882
 883         c = &cpuid_data.entries[cpuid_i++];
 884         c->function = HV_CPUID_INTERFACE;
 885         memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
 886         c->eax = signature[0];
 887         c->ebx = 0;
 888         c->ecx = 0;
 889         c->edx = 0;
 890
 891         c = &cpuid_data.entries[cpuid_i++];
 892         c->function = HV_CPUID_VERSION;
 893         c->eax = 0x00001bbc;
 894         c->ebx = 0x00060001;
 895
 896         c = &cpuid_data.entries[cpuid_i++];
 897         c->function = HV_CPUID_FEATURES;
 898         r = hyperv_handle_properties(cs);
 899         if (r) {
 900             return r;
 901         }
 902         c->eax = env->features[FEAT_HYPERV_EAX];
 903         c->ebx = env->features[FEAT_HYPERV_EBX];
 904         c->edx = env->features[FEAT_HYPERV_EDX];
 905
 906         c = &cpuid_data.entries[cpuid_i++];
 907         c->function = HV_CPUID_ENLIGHTMENT_INFO;
 908         if (cpu->hyperv_relaxed_timing) {
 909             c->eax |= HV_RELAXED_TIMING_RECOMMENDED;
 910         }
 911         if (cpu->hyperv_vapic) {
 912             c->eax |= HV_APIC_ACCESS_RECOMMENDED;
 913         }
 914         if (cpu->hyperv_tlbflush) {
 915             if (kvm_check_extension(cs->kvm_state,
 916                                     KVM_CAP_HYPERV_TLBFLUSH) <= 0) {
 917                 fprintf(stderr, "Hyper-V TLB flush support "
 918                         "(requested by 'hv-tlbflush' cpu flag) "
 919                         " is not supported by kernel\n");
 920                 return -ENOSYS;
 921             }
 922             c->eax |= HV_REMOTE_TLB_FLUSH_RECOMMENDED;
 923             c->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
 924         }
 925         if (cpu->hyperv_ipi) {
 926             if (kvm_check_extension(cs->kvm_state,
 927                                     KVM_CAP_HYPERV_SEND_IPI) <= 0) {
 928                 fprintf(stderr, "Hyper-V IPI send support "
 929                         "(requested by 'hv-ipi' cpu flag) "
 930                         " is not supported by kernel\n");
 931                 return -ENOSYS;
 932             }
 933             c->eax |= HV_CLUSTER_IPI_RECOMMENDED;
 934             c->eax |= HV_EX_PROCESSOR_MASKS_RECOMMENDED;
 935         }
 936
 937         c->ebx = cpu->hyperv_spinlock_attempts;
 938
 939         c = &cpuid_data.entries[cpuid_i++];
 940         c->function = HV_CPUID_IMPLEMENT_LIMITS;
 941
 942         c->eax = cpu->hv_max_vps;
 943         c->ebx = 0x40;
 944
 945         kvm_base = KVM_CPUID_SIGNATURE_NEXT;
 946         has_msr_hv_hypercall = true;
 947     }
 948
 949     if (cpu->expose_kvm) {
 950         memcpy(signature, "KVMKVMKVM\0\0\0", 12);
 951         c = &cpuid_data.entries[cpuid_i++];
 952         c->function = KVM_CPUID_SIGNATURE | kvm_base;
 953         c->eax = KVM_CPUID_FEATURES | kvm_base;
 954         c->ebx = signature[0];
 955         c->ecx = signature[1];
 956         c->edx = signature[2];
 957
 958         c = &cpuid_data.entries[cpuid_i++];
 959         c->function = KVM_CPUID_FEATURES | kvm_base;
 960         c->eax = env->features[FEAT_KVM];
 961         c->edx = env->features[FEAT_KVM_HINTS];
 962     }
 963
 964     cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
 965
 966     for (i = 0; i <= limit; i++) {
 967         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 968             fprintf(stderr, "unsupported level value: 0x%x\n", limit);
 969             abort();
 970         }
 971         c = &cpuid_data.entries[cpuid_i++];
 972
 973         switch (i) {
 974         case 2: {
 975             /* Keep reading function 2 till all the input is received */
 976             int times;
 977
 978             c->function = i;
 979             c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
 980                        KVM_CPUID_FLAG_STATE_READ_NEXT;
 981             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 982             times = c->eax & 0xff;
 983
 984             for (j = 1; j < times; ++j) {
 985                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
 986                     fprintf(stderr, "cpuid_data is full, no space for "
 987                             "cpuid(eax:2):eax & 0xf = 0x%x\n", times);
 988                     abort();
 989                 }
 990                 c = &cpuid_data.entries[cpuid_i++];
 991                 c->function = i;
 992                 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
 993                 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
 994             }
 995             break;
 996         }
 997         case 4:
 998         case 0xb:
 999         case 0xd:
1000             for (j = 0; ; j++) {
1001                 if (i == 0xd && j == 64) {
1002                     break;
1003                 }
1004                 c->function = i;
1005                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1006                 c->index = j;
1007                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1008
1009                 if (i == 4 && c->eax == 0) {
1010                     break;
1011                 }
1012                 if (i == 0xb && !(c->ecx & 0xff00)) {
1013                     break;
1014                 }
1015                 if (i == 0xd && c->eax == 0) {
1016                     continue;
1017                 }
1018                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1019                     fprintf(stderr, "cpuid_data is full, no space for "
1020                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1021                     abort();
1022                 }
1023                 c = &cpuid_data.entries[cpuid_i++];
1024             }
1025             break;
1026         case 0x14: {
1027             uint32_t times;
1028
1029             c->function = i;
1030             c->index = 0;
1031             c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1032             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1033             times = c->eax;
1034
1035             for (j = 1; j <= times; ++j) {
1036                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1037                     fprintf(stderr, "cpuid_data is full, no space for "
1038                                 "cpuid(eax:0x14,ecx:0x%x)\n", j);
1039                     abort();
1040                 }
1041                 c = &cpuid_data.entries[cpuid_i++];
1042                 c->function = i;
1043                 c->index = j;
1044                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1045                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1046             }
1047             break;
1048         }
1049         default:
1050             c->function = i;
1051             c->flags = 0;
1052             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1053             break;
1054         }
1055     }
1056
1057     if (limit >= 0x0a) {
1058         uint32_t eax, edx;
1059
1060         cpu_x86_cpuid(env, 0x0a, 0, &eax, &unused, &unused, &edx);
1061
1062         has_architectural_pmu_version = eax & 0xff;
1063         if (has_architectural_pmu_version > 0) {
1064             num_architectural_pmu_gp_counters = (eax & 0xff00) >> 8;
1065
1066             /* Shouldn't be more than 32, since that's the number of bits
1067              * available in EBX to tell us _which_ counters are available.
1068              * Play it safe.
1069              */
1070             if (num_architectural_pmu_gp_counters > MAX_GP_COUNTERS) {
1071                 num_architectural_pmu_gp_counters = MAX_GP_COUNTERS;
1072             }
1073
1074             if (has_architectural_pmu_version > 1) {
1075                 num_architectural_pmu_fixed_counters = edx & 0x1f;
1076
1077                 if (num_architectural_pmu_fixed_counters > MAX_FIXED_COUNTERS) {
1078                     num_architectural_pmu_fixed_counters = MAX_FIXED_COUNTERS;
1079                 }
1080             }
1081         }
1082     }
1083
1084     cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
1085
1086     for (i = 0x80000000; i <= limit; i++) {
1087         if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1088             fprintf(stderr, "unsupported xlevel value: 0x%x\n", limit);
1089             abort();
1090         }
1091         c = &cpuid_data.entries[cpuid_i++];
1092
1093         switch (i) {
1094         case 0x8000001d:
1095             /* Query for all AMD cache information leaves */
1096             for (j = 0; ; j++) {
1097                 c->function = i;
1098                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1099                 c->index = j;
1100                 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
1101
1102                 if (c->eax == 0) {
1103                     break;
1104                 }
1105                 if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1106                     fprintf(stderr, "cpuid_data is full, no space for "
1107                             "cpuid(eax:0x%x,ecx:0x%x)\n", i, j);
1108                     abort();
1109                 }
1110                 c = &cpuid_data.entries[cpuid_i++];
1111             }
1112             break;
1113         default:
1114             c->function = i;
1115             c->flags = 0;
1116             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1117             break;
1118         }
1119     }
1120
1121     /* Call Centaur's CPUID instructions they are supported. */
1122     if (env->cpuid_xlevel2 > 0) {
1123         cpu_x86_cpuid(env, 0xC0000000, 0, &limit, &unused, &unused, &unused);
1124
1125         for (i = 0xC0000000; i <= limit; i++) {
1126             if (cpuid_i == KVM_MAX_CPUID_ENTRIES) {
1127                 fprintf(stderr, "unsupported xlevel2 value: 0x%x\n", limit);
1128                 abort();
1129             }
1130             c = &cpuid_data.entries[cpuid_i++];
1131
1132             c->function = i;
1133             c->flags = 0;
1134             cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
1135         }
1136     }
1137
1138     cpuid_data.cpuid.nent = cpuid_i;
1139
1140     if (((env->cpuid_version >> 8)&0xF) >= 6
1141         && (env->features[FEAT_1_EDX] & (CPUID_MCE | CPUID_MCA)) ==
1142            (CPUID_MCE | CPUID_MCA)
1143         && kvm_check_extension(cs->kvm_state, KVM_CAP_MCE) > 0) {
1144         uint64_t mcg_cap, unsupported_caps;
1145         int banks;
1146         int ret;
1147
1148         ret = kvm_get_mce_cap_supported(cs->kvm_state, &mcg_cap, &banks);
1149         if (ret < 0) {
1150             fprintf(stderr, "kvm_get_mce_cap_supported: %s", strerror(-ret));
1151             return ret;
1152         }
1153
1154         if (banks < (env->mcg_cap & MCG_CAP_BANKS_MASK)) {
1155             error_report("kvm: Unsupported MCE bank count (QEMU = %d, KVM = %d)",
1156                          (int)(env->mcg_cap & MCG_CAP_BANKS_MASK), banks);
1157             return -ENOTSUP;
1158         }
1159
1160         unsupported_caps = env->mcg_cap & ~(mcg_cap | MCG_CAP_BANKS_MASK);
1161         if (unsupported_caps) {
1162             if (unsupported_caps & MCG_LMCE_P) {
1163                 error_report("kvm: LMCE not supported");
1164                 return -ENOTSUP;
1165             }
1166             warn_report("Unsupported MCG_CAP bits: 0x%" PRIx64,
1167                         unsupported_caps);
1168         }
1169
1170         env->mcg_cap &= mcg_cap | MCG_CAP_BANKS_MASK;
1171         ret = kvm_vcpu_ioctl(cs, KVM_X86_SETUP_MCE, &env->mcg_cap);
1172         if (ret < 0) {
1173             fprintf(stderr, "KVM_X86_SETUP_MCE: %s", strerror(-ret));
1174             return ret;
1175         }
1176     }
1177
1178     qemu_add_vm_change_state_handler(cpu_update_state, env);
1179
1180     c = cpuid_find_entry(&cpuid_data.cpuid, 1, 0);
1181     if (c) {
1182         has_msr_feature_control = !!(c->ecx & CPUID_EXT_VMX) ||
1183                                   !!(c->ecx & CPUID_EXT_SMX);
1184     }
1185
1186     if (env->mcg_cap & MCG_LMCE_P) {
1187         has_msr_mcg_ext_ctl = has_msr_feature_control = true;
1188     }
1189
1190     if (!env->user_tsc_khz) {
1191         if ((env->features[FEAT_8000_0007_EDX] & CPUID_APM_INVTSC) &&
1192             invtsc_mig_blocker == NULL) {
1193             /* for migration */
1194             error_setg(&invtsc_mig_blocker,
1195                        "State blocked by non-migratable CPU device"
1196                        " (invtsc flag)");
1197             r = migrate_add_blocker(invtsc_mig_blocker, &local_err);
1198             if (local_err) {
1199                 error_report_err(local_err);
1200                 error_free(invtsc_mig_blocker);
1201                 goto fail;
1202             }
1203             /* for savevm */
1204             vmstate_x86_cpu.unmigratable = 1;
1205         }
1206     }
1207
1208     if (cpu->vmware_cpuid_freq
1209         /* Guests depend on 0x40000000 to detect this feature, so only expose
1210          * it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
1211         && cpu->expose_kvm
1212         && kvm_base == KVM_CPUID_SIGNATURE
1213         /* TSC clock must be stable and known for this feature. */
1214         && tsc_is_stable_and_known(env)) {
1215
1216         c = &cpuid_data.entries[cpuid_i++];
1217         c->function = KVM_CPUID_SIGNATURE | 0x10;
1218         c->eax = env->tsc_khz;
1219         /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
1220          * APIC_BUS_CYCLE_NS */
1221         c->ebx = 1000000;
1222         c->ecx = c->edx = 0;
1223
1224         c = cpuid_find_entry(&cpuid_data.cpuid, kvm_base, 0);
1225         c->eax = MAX(c->eax, KVM_CPUID_SIGNATURE | 0x10);
1226     }
1227
1228     cpuid_data.cpuid.nent = cpuid_i;
1229
1230     cpuid_data.cpuid.padding = 0;
1231     r = kvm_vcpu_ioctl(cs, KVM_SET_CPUID2, &cpuid_data);
1232     if (r) {
1233         goto fail;
1234     }
1235
1236     if (has_xsave) {
1237         env->xsave_buf = qemu_memalign(4096, sizeof(struct kvm_xsave));
1238     }
1239     cpu->kvm_msr_buf = g_malloc0(MSR_BUF_SIZE);
1240
1241     if (!(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_RDTSCP)) {
1242         has_msr_tsc_aux = false;
1243     }
1244
1245     r = hyperv_init_vcpu(cpu);
1246     if (r) {
1247         goto fail;
1248     }
1249
1250     return 0;
1251
1252  fail:
1253     migrate_del_blocker(invtsc_mig_blocker);
1254     return r;
1255 }
1256
1257 void kvm_arch_reset_vcpu(X86CPU *cpu)
1258 {
1259     CPUX86State *env = &cpu->env;
1260
1261     env->xcr0 = 1;
1262     if (kvm_irqchip_in_kernel()) {
1263         env->mp_state = cpu_is_bsp(cpu) ? KVM_MP_STATE_RUNNABLE :
1264                                           KVM_MP_STATE_UNINITIALIZED;
1265     } else {
1266         env->mp_state = KVM_MP_STATE_RUNNABLE;
1267     }
1268
1269     if (cpu->hyperv_synic) {
1270         int i;
1271         for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
1272             env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
1273         }
1274
1275         hyperv_x86_synic_reset(cpu);
1276     }
1277 }
1278
1279 void kvm_arch_do_init_vcpu(X86CPU *cpu)
1280 {
1281     CPUX86State *env = &cpu->env;
1282
1283     /* APs get directly into wait-for-SIPI state.  */
1284     if (env->mp_state == KVM_MP_STATE_UNINITIALIZED) {
1285         env->mp_state = KVM_MP_STATE_INIT_RECEIVED;
1286     }
1287 }
1288
1289 static int kvm_get_supported_msrs(KVMState *s)
1290 {
1291     static int kvm_supported_msrs;
1292     int ret = 0;
1293
1294     /* first time */
1295     if (kvm_supported_msrs == 0) {
1296         struct kvm_msr_list msr_list, *kvm_msr_list;
1297
1298         kvm_supported_msrs = -1;
1299
1300         /* Obtain MSR list from KVM.  These are the MSRs that we must
1301          * save/restore */
1302         msr_list.nmsrs = 0;
1303         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, &msr_list);
1304         if (ret < 0 && ret != -E2BIG) {
1305             return ret;
1306         }
1307         /* Old kernel modules had a bug and could write beyond the provided
1308            memory. Allocate at least a safe amount of 1K. */
1309         kvm_msr_list = g_malloc0(MAX(1024, sizeof(msr_list) +
1310                                               msr_list.nmsrs *
1311                                               sizeof(msr_list.indices[0])));
1312
1313         kvm_msr_list->nmsrs = msr_list.nmsrs;
1314         ret = kvm_ioctl(s, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
1315         if (ret >= 0) {
1316             int i;
1317
1318             for (i = 0; i < kvm_msr_list->nmsrs; i++) {
1319                 switch (kvm_msr_list->indices[i]) {
1320                 case MSR_STAR:
1321                     has_msr_star = true;
1322                     break;
1323                 case MSR_VM_HSAVE_PA:
1324                     has_msr_hsave_pa = true;
1325                     break;
1326                 case MSR_TSC_AUX:
1327                     has_msr_tsc_aux = true;
1328                     break;
1329                 case MSR_TSC_ADJUST:
1330                     has_msr_tsc_adjust = true;
1331                     break;
1332                 case MSR_IA32_TSCDEADLINE:
1333                     has_msr_tsc_deadline = true;
1334                     break;
1335                 case MSR_IA32_SMBASE:
1336                     has_msr_smbase = true;
1337                     break;
1338                 case MSR_SMI_COUNT:
1339                     has_msr_smi_count = true;
1340                     break;
1341                 case MSR_IA32_MISC_ENABLE:
1342                     has_msr_misc_enable = true;
1343                     break;
1344                 case MSR_IA32_BNDCFGS:
1345                     has_msr_bndcfgs = true;
1346                     break;
1347                 case MSR_IA32_XSS:
1348                     has_msr_xss = true;
1349                     break;
1350                 case HV_X64_MSR_CRASH_CTL:
1351                     has_msr_hv_crash = true;
1352                     break;
1353                 case HV_X64_MSR_RESET:
1354                     has_msr_hv_reset = true;
1355                     break;
1356                 case HV_X64_MSR_VP_INDEX:
1357                     has_msr_hv_vpindex = true;
1358                     break;
1359                 case HV_X64_MSR_VP_RUNTIME:
1360                     has_msr_hv_runtime = true;
1361                     break;
1362                 case HV_X64_MSR_SCONTROL:
1363                     has_msr_hv_synic = true;
1364                     break;
1365                 case HV_X64_MSR_STIMER0_CONFIG:
1366                     has_msr_hv_stimer = true;
1367                     break;
1368                 case HV_X64_MSR_TSC_FREQUENCY:
1369                     has_msr_hv_frequencies = true;
1370                     break;
1371                 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1372                     has_msr_hv_reenlightenment = true;
1373                     break;
1374                 case MSR_IA32_SPEC_CTRL:
1375                     has_msr_spec_ctrl = true;
1376                     break;
1377                 case MSR_VIRT_SSBD:
1378                     has_msr_virt_ssbd = true;
1379                     break;
1380                 }
1381             }
1382         }
1383
1384         g_free(kvm_msr_list);
1385     }
1386
1387     return ret;
1388 }
1389
1390 static Notifier smram_machine_done;
1391 static KVMMemoryListener smram_listener;
1392 static AddressSpace smram_address_space;
1393 static MemoryRegion smram_as_root;
1394 static MemoryRegion smram_as_mem;
1395
1396 static void register_smram_listener(Notifier *n, void *unused)
1397 {
1398     MemoryRegion *smram =
1399         (MemoryRegion *) object_resolve_path("/machine/smram", NULL);
1400
1401     /* Outer container... */
1402     memory_region_init(&smram_as_root, OBJECT(kvm_state), "mem-container-smram", ~0ull);
1403     memory_region_set_enabled(&smram_as_root, true);
1404
1405     /* ... with two regions inside: normal system memory with low
1406      * priority, and...
1407      */
1408     memory_region_init_alias(&smram_as_mem, OBJECT(kvm_state), "mem-smram",
1409                              get_system_memory(), 0, ~0ull);
1410     memory_region_add_subregion_overlap(&smram_as_root, 0, &smram_as_mem, 0);
1411     memory_region_set_enabled(&smram_as_mem, true);
1412
1413     if (smram) {
1414         /* ... SMRAM with higher priority */
1415         memory_region_add_subregion_overlap(&smram_as_root, 0, smram, 10);
1416         memory_region_set_enabled(smram, true);
1417     }
1418
1419     address_space_init(&smram_address_space, &smram_as_root, "KVM-SMRAM");
1420     kvm_memory_listener_register(kvm_state, &smram_listener,
1421                                  &smram_address_space, 1);
1422 }
1423
1424 int kvm_arch_init(MachineState *ms, KVMState *s)
1425 {
1426     uint64_t identity_base = 0xfffbc000;
1427     uint64_t shadow_mem;
1428     int ret;
1429     struct utsname utsname;
1430
1431     has_xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1432     has_xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1433     has_pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1434
1435     hv_vpindex_settable = kvm_check_extension(s, KVM_CAP_HYPERV_VP_INDEX);
1436
1437     ret = kvm_get_supported_msrs(s);
1438     if (ret < 0) {
1439         return ret;
1440     }
1441
1442     uname(&utsname);
1443     lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
1444
1445     /*
1446      * On older Intel CPUs, KVM uses vm86 mode to emulate 16-bit code directly.
1447      * In order to use vm86 mode, an EPT identity map and a TSS  are needed.
1448      * Since these must be part of guest physical memory, we need to allocate
1449      * them, both by setting their start addresses in the kernel and by
1450      * creating a corresponding e820 entry. We need 4 pages before the BIOS.
1451      *
1452      * Older KVM versions may not support setting the identity map base. In
1453      * that case we need to stick with the default, i.e. a 256K maximum BIOS
1454      * size.
1455      */
1456     if (kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
1457         /* Allows up to 16M BIOSes. */
1458         identity_base = 0xfeffc000;
1459
1460         ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &identity_base);
1461         if (ret < 0) {
1462             return ret;
1463         }
1464     }
1465
1466     /* Set TSS base one page after EPT identity map. */
1467     ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, identity_base + 0x1000);
1468     if (ret < 0) {
1469         return ret;
1470     }
1471
1472     /* Tell fw_cfg to notify the BIOS to reserve the range. */
1473     ret = e820_add_entry(identity_base, 0x4000, E820_RESERVED);
1474     if (ret < 0) {
1475         fprintf(stderr, "e820_add_entry() table is full\n");
1476         return ret;
1477     }
1478     qemu_register_reset(kvm_unpoison_all, NULL);
1479
1480     shadow_mem = machine_kvm_shadow_mem(ms);
1481     if (shadow_mem != -1) {
1482         shadow_mem /= 4096;
1483         ret = kvm_vm_ioctl(s, KVM_SET_NR_MMU_PAGES, shadow_mem);
1484         if (ret < 0) {
1485             return ret;
1486         }
1487     }
1488
1489     if (kvm_check_extension(s, KVM_CAP_X86_SMM) &&
1490         object_dynamic_cast(OBJECT(ms), TYPE_PC_MACHINE) &&
1491         pc_machine_is_smm_enabled(PC_MACHINE(ms))) {
1492         smram_machine_done.notify = register_smram_listener;
1493         qemu_add_machine_init_done_notifier(&smram_machine_done);
1494     }
1495
1496     if (enable_cpu_pm) {
1497         int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
1498         int ret;
1499
1500 /* Work around for kernel header with a typo. TODO: fix header and drop. */
1501 #if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
1502 #define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
1503 #endif
1504         if (disable_exits) {
1505             disable_exits &= (KVM_X86_DISABLE_EXITS_MWAIT |
1506                               KVM_X86_DISABLE_EXITS_HLT |
1507                               KVM_X86_DISABLE_EXITS_PAUSE);
1508         }
1509
1510         ret = kvm_vm_enable_cap(s, KVM_CAP_X86_DISABLE_EXITS, 0,
1511                                 disable_exits);
1512         if (ret < 0) {
1513             error_report("kvm: guest stopping CPU not supported: %s",
1514                          strerror(-ret));
1515         }
1516     }
1517
1518     return 0;
1519 }
1520
1521 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1522 {
1523     lhs->selector = rhs->selector;
1524     lhs->base = rhs->base;
1525     lhs->limit = rhs->limit;
1526     lhs->type = 3;
1527     lhs->present = 1;
1528     lhs->dpl = 3;
1529     lhs->db = 0;
1530     lhs->s = 1;
1531     lhs->l = 0;
1532     lhs->g = 0;
1533     lhs->avl = 0;
1534     lhs->unusable = 0;
1535 }
1536
1537 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
1538 {
1539     unsigned flags = rhs->flags;
1540     lhs->selector = rhs->selector;
1541     lhs->base = rhs->base;
1542     lhs->limit = rhs->limit;
1543     lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
1544     lhs->present = (flags & DESC_P_MASK) != 0;
1545     lhs->dpl = (flags >> DESC_DPL_SHIFT) & 3;
1546     lhs->db = (flags >> DESC_B_SHIFT) & 1;
1547     lhs->s = (flags & DESC_S_MASK) != 0;
1548     lhs->l = (flags >> DESC_L_SHIFT) & 1;
1549     lhs->g = (flags & DESC_G_MASK) != 0;
1550     lhs->avl = (flags & DESC_AVL_MASK) != 0;
1551     lhs->unusable = !lhs->present;
1552     lhs->padding = 0;
1553 }
1554
1555 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
1556 {
1557     lhs->selector = rhs->selector;
1558     lhs->base = rhs->base;
1559     lhs->limit = rhs->limit;
1560     lhs->flags = (rhs->type << DESC_TYPE_SHIFT) |
1561                  ((rhs->present && !rhs->unusable) * DESC_P_MASK) |
1562                  (rhs->dpl << DESC_DPL_SHIFT) |
1563                  (rhs->db << DESC_B_SHIFT) |
1564                  (rhs->s * DESC_S_MASK) |
1565                  (rhs->l << DESC_L_SHIFT) |
1566                  (rhs->g * DESC_G_MASK) |
1567                  (rhs->avl * DESC_AVL_MASK);
1568 }
1569
1570 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
1571 {
1572     if (set) {
1573         *kvm_reg = *qemu_reg;
1574     } else {
1575         *qemu_reg = *kvm_reg;
1576     }
1577 }
1578
1579 static int kvm_getput_regs(X86CPU *cpu, int set)
1580 {
1581     CPUX86State *env = &cpu->env;
1582     struct kvm_regs regs;
1583     int ret = 0;
1584
1585     if (!set) {
1586         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_REGS, &regs);
1587         if (ret < 0) {
1588             return ret;
1589         }
1590     }
1591
1592     kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
1593     kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
1594     kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
1595     kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
1596     kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
1597     kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
1598     kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
1599     kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
1600 #ifdef TARGET_X86_64
1601     kvm_getput_reg(&regs.r8, &env->regs[8], set);
1602     kvm_getput_reg(&regs.r9, &env->regs[9], set);
1603     kvm_getput_reg(&regs.r10, &env->regs[10], set);
1604     kvm_getput_reg(&regs.r11, &env->regs[11], set);
1605     kvm_getput_reg(&regs.r12, &env->regs[12], set);
1606     kvm_getput_reg(&regs.r13, &env->regs[13], set);
1607     kvm_getput_reg(&regs.r14, &env->regs[14], set);
1608     kvm_getput_reg(&regs.r15, &env->regs[15], set);
1609 #endif
1610
1611     kvm_getput_reg(&regs.rflags, &env->eflags, set);
1612     kvm_getput_reg(&regs.rip, &env->eip, set);
1613
1614     if (set) {
1615         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_REGS, &regs);
1616     }
1617
1618     return ret;
1619 }
1620
1621 static int kvm_put_fpu(X86CPU *cpu)
1622 {
1623     CPUX86State *env = &cpu->env;
1624     struct kvm_fpu fpu;
1625     int i;
1626
1627     memset(&fpu, 0, sizeof fpu);
1628     fpu.fsw = env->fpus & ~(7 << 11);
1629     fpu.fsw |= (env->fpstt & 7) << 11;
1630     fpu.fcw = env->fpuc;
1631     fpu.last_opcode = env->fpop;
1632     fpu.last_ip = env->fpip;
1633     fpu.last_dp = env->fpdp;
1634     for (i = 0; i < 8; ++i) {
1635         fpu.ftwx |= (!env->fptags[i]) << i;
1636     }
1637     memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
1638     for (i = 0; i < CPU_NB_REGS; i++) {
1639         stq_p(&fpu.xmm[i][0], env->xmm_regs[i].ZMM_Q(0));
1640         stq_p(&fpu.xmm[i][8], env->xmm_regs[i].ZMM_Q(1));
1641     }
1642     fpu.mxcsr = env->mxcsr;
1643
1644     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_FPU, &fpu);
1645 }
1646
1647 #define XSAVE_FCW_FSW     0
1648 #define XSAVE_FTW_FOP     1
1649 #define XSAVE_CWD_RIP     2
1650 #define XSAVE_CWD_RDP     4
1651 #define XSAVE_MXCSR       6
1652 #define XSAVE_ST_SPACE    8
1653 #define XSAVE_XMM_SPACE   40
1654 #define XSAVE_XSTATE_BV   128
1655 #define XSAVE_YMMH_SPACE  144
1656 #define XSAVE_BNDREGS     240
1657 #define XSAVE_BNDCSR      256
1658 #define XSAVE_OPMASK      272
1659 #define XSAVE_ZMM_Hi256   288
1660 #define XSAVE_Hi16_ZMM    416
1661 #define XSAVE_PKRU        672
1662
1663 #define XSAVE_BYTE_OFFSET(word_offset) \
1664     ((word_offset) * sizeof_field(struct kvm_xsave, region[0]))
1665
1666 #define ASSERT_OFFSET(word_offset, field) \
1667     QEMU_BUILD_BUG_ON(XSAVE_BYTE_OFFSET(word_offset) != \
1668                       offsetof(X86XSaveArea, field))
1669
1670 ASSERT_OFFSET(XSAVE_FCW_FSW, legacy.fcw);
1671 ASSERT_OFFSET(XSAVE_FTW_FOP, legacy.ftw);
1672 ASSERT_OFFSET(XSAVE_CWD_RIP, legacy.fpip);
1673 ASSERT_OFFSET(XSAVE_CWD_RDP, legacy.fpdp);
1674 ASSERT_OFFSET(XSAVE_MXCSR, legacy.mxcsr);
1675 ASSERT_OFFSET(XSAVE_ST_SPACE, legacy.fpregs);
1676 ASSERT_OFFSET(XSAVE_XMM_SPACE, legacy.xmm_regs);
1677 ASSERT_OFFSET(XSAVE_XSTATE_BV, header.xstate_bv);
1678 ASSERT_OFFSET(XSAVE_YMMH_SPACE, avx_state);
1679 ASSERT_OFFSET(XSAVE_BNDREGS, bndreg_state);
1680 ASSERT_OFFSET(XSAVE_BNDCSR, bndcsr_state);
1681 ASSERT_OFFSET(XSAVE_OPMASK, opmask_state);
1682 ASSERT_OFFSET(XSAVE_ZMM_Hi256, zmm_hi256_state);
1683 ASSERT_OFFSET(XSAVE_Hi16_ZMM, hi16_zmm_state);
1684 ASSERT_OFFSET(XSAVE_PKRU, pkru_state);
1685
1686 static int kvm_put_xsave(X86CPU *cpu)
1687 {
1688     CPUX86State *env = &cpu->env;
1689     X86XSaveArea *xsave = env->xsave_buf;
1690
1691     if (!has_xsave) {
1692         return kvm_put_fpu(cpu);
1693     }
1694     x86_cpu_xsave_all_areas(cpu, xsave);
1695
1696     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XSAVE, xsave);
1697 }
1698
1699 static int kvm_put_xcrs(X86CPU *cpu)
1700 {
1701     CPUX86State *env = &cpu->env;
1702     struct kvm_xcrs xcrs = {};
1703
1704     if (!has_xcrs) {
1705         return 0;
1706     }
1707
1708     xcrs.nr_xcrs = 1;
1709     xcrs.flags = 0;
1710     xcrs.xcrs[0].xcr = 0;
1711     xcrs.xcrs[0].value = env->xcr0;
1712     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_XCRS, &xcrs);
1713 }
1714
1715 static int kvm_put_sregs(X86CPU *cpu)
1716 {
1717     CPUX86State *env = &cpu->env;
1718     struct kvm_sregs sregs;
1719
1720     memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
1721     if (env->interrupt_injected >= 0) {
1722         sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
1723                 (uint64_t)1 << (env->interrupt_injected % 64);
1724     }
1725
1726     if ((env->eflags & VM_MASK)) {
1727         set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
1728         set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
1729         set_v8086_seg(&sregs.es, &env->segs[R_ES]);
1730         set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
1731         set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
1732         set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
1733     } else {
1734         set_seg(&sregs.cs, &env->segs[R_CS]);
1735         set_seg(&sregs.ds, &env->segs[R_DS]);
1736         set_seg(&sregs.es, &env->segs[R_ES]);
1737         set_seg(&sregs.fs, &env->segs[R_FS]);
1738         set_seg(&sregs.gs, &env->segs[R_GS]);
1739         set_seg(&sregs.ss, &env->segs[R_SS]);
1740     }
1741
1742     set_seg(&sregs.tr, &env->tr);
1743     set_seg(&sregs.ldt, &env->ldt);
1744
1745     sregs.idt.limit = env->idt.limit;
1746     sregs.idt.base = env->idt.base;
1747     memset(sregs.idt.padding, 0, sizeof sregs.idt.padding);
1748     sregs.gdt.limit = env->gdt.limit;
1749     sregs.gdt.base = env->gdt.base;
1750     memset(sregs.gdt.padding, 0, sizeof sregs.gdt.padding);
1751
1752     sregs.cr0 = env->cr[0];
1753     sregs.cr2 = env->cr[2];
1754     sregs.cr3 = env->cr[3];
1755     sregs.cr4 = env->cr[4];
1756
1757     sregs.cr8 = cpu_get_apic_tpr(cpu->apic_state);
1758     sregs.apic_base = cpu_get_apic_base(cpu->apic_state);
1759
1760     sregs.efer = env->efer;
1761
1762     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
1763 }
1764
1765 static void kvm_msr_buf_reset(X86CPU *cpu)
1766 {
1767     memset(cpu->kvm_msr_buf, 0, MSR_BUF_SIZE);
1768 }
1769
1770 static void kvm_msr_entry_add(X86CPU *cpu, uint32_t index, uint64_t value)
1771 {
1772     struct kvm_msrs *msrs = cpu->kvm_msr_buf;
1773     void *limit = ((void *)msrs) + MSR_BUF_SIZE;
1774     struct kvm_msr_entry *entry = &msrs->entries[msrs->nmsrs];
1775
1776     assert((void *)(entry + 1) <= limit);
1777
1778     entry->index = index;
1779     entry->reserved = 0;
1780     entry->data = value;
1781     msrs->nmsrs++;
1782 }
1783
1784 static int kvm_put_one_msr(X86CPU *cpu, int index, uint64_t value)
1785 {
1786     kvm_msr_buf_reset(cpu);
1787     kvm_msr_entry_add(cpu, index, value);
1788
1789     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
1790 }
1791
1792 void kvm_put_apicbase(X86CPU *cpu, uint64_t value)
1793 {
1794     int ret;
1795
1796     ret = kvm_put_one_msr(cpu, MSR_IA32_APICBASE, value);
1797     assert(ret == 1);
1798 }
1799
1800 static int kvm_put_tscdeadline_msr(X86CPU *cpu)
1801 {
1802     CPUX86State *env = &cpu->env;
1803     int ret;
1804
1805     if (!has_msr_tsc_deadline) {
1806         return 0;
1807     }
1808
1809     ret = kvm_put_one_msr(cpu, MSR_IA32_TSCDEADLINE, env->tsc_deadline);
1810     if (ret < 0) {
1811         return ret;
1812     }
1813
1814     assert(ret == 1);
1815     return 0;
1816 }
1817
1818 /*
1819  * Provide a separate write service for the feature control MSR in order to
1820  * kick the VCPU out of VMXON or even guest mode on reset. This has to be done
1821  * before writing any other state because forcibly leaving nested mode
1822  * invalidates the VCPU state.
1823  */
1824 static int kvm_put_msr_feature_control(X86CPU *cpu)
1825 {
1826     int ret;
1827
1828     if (!has_msr_feature_control) {
1829         return 0;
1830     }
1831
1832     ret = kvm_put_one_msr(cpu, MSR_IA32_FEATURE_CONTROL,
1833                           cpu->env.msr_ia32_feature_control);
1834     if (ret < 0) {
1835         return ret;
1836     }
1837
1838     assert(ret == 1);
1839     return 0;
1840 }
1841
1842 static int kvm_put_msrs(X86CPU *cpu, int level)
1843 {
1844     CPUX86State *env = &cpu->env;
1845     int i;
1846     int ret;
1847
1848     kvm_msr_buf_reset(cpu);
1849
1850     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, env->sysenter_cs);
1851     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
1852     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
1853     kvm_msr_entry_add(cpu, MSR_PAT, env->pat);
1854     if (has_msr_star) {
1855         kvm_msr_entry_add(cpu, MSR_STAR, env->star);
1856     }
1857     if (has_msr_hsave_pa) {
1858         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, env->vm_hsave);
1859     }
1860     if (has_msr_tsc_aux) {
1861         kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
1862     }
1863     if (has_msr_tsc_adjust) {
1864         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
1865     }
1866     if (has_msr_misc_enable) {
1867         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
1868                           env->msr_ia32_misc_enable);
1869     }
1870     if (has_msr_smbase) {
1871         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, env->smbase);
1872     }
1873     if (has_msr_smi_count) {
1874         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, env->msr_smi_count);
1875     }
1876     if (has_msr_bndcfgs) {
1877         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, env->msr_bndcfgs);
1878     }
1879     if (has_msr_xss) {
1880         kvm_msr_entry_add(cpu, MSR_IA32_XSS, env->xss);
1881     }
1882     if (has_msr_spec_ctrl) {
1883         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, env->spec_ctrl);
1884     }
1885     if (has_msr_virt_ssbd) {
1886         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, env->virt_ssbd);
1887     }
1888
1889 #ifdef TARGET_X86_64
1890     if (lm_capable_kernel) {
1891         kvm_msr_entry_add(cpu, MSR_CSTAR, env->cstar);
1892         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, env->kernelgsbase);
1893         kvm_msr_entry_add(cpu, MSR_FMASK, env->fmask);
1894         kvm_msr_entry_add(cpu, MSR_LSTAR, env->lstar);
1895     }
1896 #endif
1897
1898     /*
1899      * The following MSRs have side effects on the guest or are too heavy
1900      * for normal writeback. Limit them to reset or full state updates.
1901      */
1902     if (level >= KVM_PUT_RESET_STATE) {
1903         kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
1904         kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
1905         kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
1906         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
1907             kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
1908         }
1909         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
1910             kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, env->pv_eoi_en_msr);
1911         }
1912         if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
1913             kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, env->steal_time_msr);
1914         }
1915         if (has_architectural_pmu_version > 0) {
1916             if (has_architectural_pmu_version > 1) {
1917                 /* Stop the counter.  */
1918                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
1919                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
1920             }
1921
1922             /* Set the counter values.  */
1923             for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
1924                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i,
1925                                   env->msr_fixed_counters[i]);
1926             }
1927             for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
1928                 kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i,
1929                                   env->msr_gp_counters[i]);
1930                 kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i,
1931                                   env->msr_gp_evtsel[i]);
1932             }
1933             if (has_architectural_pmu_version > 1) {
1934                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS,
1935                                   env->msr_global_status);
1936                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1937                                   env->msr_global_ovf_ctrl);
1938
1939                 /* Now start the PMU.  */
1940                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL,
1941                                   env->msr_fixed_ctr_ctrl);
1942                 kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL,
1943                                   env->msr_global_ctrl);
1944             }
1945         }
1946         /*
1947          * Hyper-V partition-wide MSRs: to avoid clearing them on cpu hot-add,
1948          * only sync them to KVM on the first cpu
1949          */
1950         if (current_cpu == first_cpu) {
1951             if (has_msr_hv_hypercall) {
1952                 kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID,
1953                                   env->msr_hv_guest_os_id);
1954                 kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL,
1955                                   env->msr_hv_hypercall);
1956             }
1957             if (cpu->hyperv_time) {
1958                 kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC,
1959                                   env->msr_hv_tsc);
1960             }
1961             if (cpu->hyperv_reenlightenment) {
1962                 kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL,
1963                                   env->msr_hv_reenlightenment_control);
1964                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL,
1965                                   env->msr_hv_tsc_emulation_control);
1966                 kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS,
1967                                   env->msr_hv_tsc_emulation_status);
1968             }
1969         }
1970         if (cpu->hyperv_vapic) {
1971             kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE,
1972                               env->msr_hv_vapic);
1973         }
1974         if (has_msr_hv_crash) {
1975             int j;
1976
1977             for (j = 0; j < HV_CRASH_PARAMS; j++)
1978                 kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j,
1979                                   env->msr_hv_crash_params[j]);
1980
1981             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_NOTIFY);
1982         }
1983         if (has_msr_hv_runtime) {
1984             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, env->msr_hv_runtime);
1985         }
1986         if (cpu->hyperv_vpindex && hv_vpindex_settable) {
1987             kvm_msr_entry_add(cpu, HV_X64_MSR_VP_INDEX,
1988                               hyperv_vp_index(CPU(cpu)));
1989         }
1990         if (cpu->hyperv_synic) {
1991             int j;
1992
1993             kvm_msr_entry_add(cpu, HV_X64_MSR_SVERSION, HV_SYNIC_VERSION);
1994
1995             kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL,
1996                               env->msr_hv_synic_control);
1997             kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP,
1998                               env->msr_hv_synic_evt_page);
1999             kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP,
2000                               env->msr_hv_synic_msg_page);
2001
2002             for (j = 0; j < ARRAY_SIZE(env->msr_hv_synic_sint); j++) {
2003                 kvm_msr_entry_add(cpu, HV_X64_MSR_SINT0 + j,
2004                                   env->msr_hv_synic_sint[j]);
2005             }
2006         }
2007         if (has_msr_hv_stimer) {
2008             int j;
2009
2010             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_config); j++) {
2011                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_CONFIG + j * 2,
2012                                 env->msr_hv_stimer_config[j]);
2013             }
2014
2015             for (j = 0; j < ARRAY_SIZE(env->msr_hv_stimer_count); j++) {
2016                 kvm_msr_entry_add(cpu, HV_X64_MSR_STIMER0_COUNT + j * 2,
2017                                 env->msr_hv_stimer_count[j]);
2018             }
2019         }
2020         if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2021             uint64_t phys_mask = MAKE_64BIT_MASK(0, cpu->phys_bits);
2022
2023             kvm_msr_entry_add(cpu, MSR_MTRRdefType, env->mtrr_deftype);
2024             kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, env->mtrr_fixed[0]);
2025             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, env->mtrr_fixed[1]);
2026             kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, env->mtrr_fixed[2]);
2027             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, env->mtrr_fixed[3]);
2028             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, env->mtrr_fixed[4]);
2029             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, env->mtrr_fixed[5]);
2030             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, env->mtrr_fixed[6]);
2031             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, env->mtrr_fixed[7]);
2032             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, env->mtrr_fixed[8]);
2033             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, env->mtrr_fixed[9]);
2034             kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, env->mtrr_fixed[10]);
2035             for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2036                 /* The CPU GPs if we write to a bit above the physical limit of
2037                  * the host CPU (and KVM emulates that)
2038                  */
2039                 uint64_t mask = env->mtrr_var[i].mask;
2040                 mask &= phys_mask;
2041
2042                 kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i),
2043                                   env->mtrr_var[i].base);
2044                 kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), mask);
2045             }
2046         }
2047         if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2048             int addr_num = kvm_arch_get_supported_cpuid(kvm_state,
2049                                                     0x14, 1, R_EAX) & 0x7;
2050
2051             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL,
2052                             env->msr_rtit_ctrl);
2053             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS,
2054                             env->msr_rtit_status);
2055             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE,
2056                             env->msr_rtit_output_base);
2057             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK,
2058                             env->msr_rtit_output_mask);
2059             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH,
2060                             env->msr_rtit_cr3_match);
2061             for (i = 0; i < addr_num; i++) {
2062                 kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i,
2063                             env->msr_rtit_addrs[i]);
2064             }
2065         }
2066
2067         /* Note: MSR_IA32_FEATURE_CONTROL is written separately, see
2068          *       kvm_put_msr_feature_control. */
2069     }
2070     if (env->mcg_cap) {
2071         int i;
2072
2073         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, env->mcg_status);
2074         kvm_msr_entry_add(cpu, MSR_MCG_CTL, env->mcg_ctl);
2075         if (has_msr_mcg_ext_ctl) {
2076             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, env->mcg_ext_ctl);
2077         }
2078         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2079             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, env->mce_banks[i]);
2080         }
2081     }
2082
2083     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, cpu->kvm_msr_buf);
2084     if (ret < 0) {
2085         return ret;
2086     }
2087
2088     if (ret < cpu->kvm_msr_buf->nmsrs) {
2089         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2090         error_report("error: failed to set MSR 0x%" PRIx32 " to 0x%" PRIx64,
2091                      (uint32_t)e->index, (uint64_t)e->data);
2092     }
2093
2094     assert(ret == cpu->kvm_msr_buf->nmsrs);
2095     return 0;
2096 }
2097
2098
2099 static int kvm_get_fpu(X86CPU *cpu)
2100 {
2101     CPUX86State *env = &cpu->env;
2102     struct kvm_fpu fpu;
2103     int i, ret;
2104
2105     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_FPU, &fpu);
2106     if (ret < 0) {
2107         return ret;
2108     }
2109
2110     env->fpstt = (fpu.fsw >> 11) & 7;
2111     env->fpus = fpu.fsw;
2112     env->fpuc = fpu.fcw;
2113     env->fpop = fpu.last_opcode;
2114     env->fpip = fpu.last_ip;
2115     env->fpdp = fpu.last_dp;
2116     for (i = 0; i < 8; ++i) {
2117         env->fptags[i] = !((fpu.ftwx >> i) & 1);
2118     }
2119     memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
2120     for (i = 0; i < CPU_NB_REGS; i++) {
2121         env->xmm_regs[i].ZMM_Q(0) = ldq_p(&fpu.xmm[i][0]);
2122         env->xmm_regs[i].ZMM_Q(1) = ldq_p(&fpu.xmm[i][8]);
2123     }
2124     env->mxcsr = fpu.mxcsr;
2125
2126     return 0;
2127 }
2128
2129 static int kvm_get_xsave(X86CPU *cpu)
2130 {
2131     CPUX86State *env = &cpu->env;
2132     X86XSaveArea *xsave = env->xsave_buf;
2133     int ret;
2134
2135     if (!has_xsave) {
2136         return kvm_get_fpu(cpu);
2137     }
2138
2139     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XSAVE, xsave);
2140     if (ret < 0) {
2141         return ret;
2142     }
2143     x86_cpu_xrstor_all_areas(cpu, xsave);
2144
2145     return 0;
2146 }
2147
2148 static int kvm_get_xcrs(X86CPU *cpu)
2149 {
2150     CPUX86State *env = &cpu->env;
2151     int i, ret;
2152     struct kvm_xcrs xcrs;
2153
2154     if (!has_xcrs) {
2155         return 0;
2156     }
2157
2158     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_XCRS, &xcrs);
2159     if (ret < 0) {
2160         return ret;
2161     }
2162
2163     for (i = 0; i < xcrs.nr_xcrs; i++) {
2164         /* Only support xcr0 now */
2165         if (xcrs.xcrs[i].xcr == 0) {
2166             env->xcr0 = xcrs.xcrs[i].value;
2167             break;
2168         }
2169     }
2170     return 0;
2171 }
2172
2173 static int kvm_get_sregs(X86CPU *cpu)
2174 {
2175     CPUX86State *env = &cpu->env;
2176     struct kvm_sregs sregs;
2177     int bit, i, ret;
2178
2179     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
2180     if (ret < 0) {
2181         return ret;
2182     }
2183
2184     /* There can only be one pending IRQ set in the bitmap at a time, so try
2185        to find it and save its number instead (-1 for none). */
2186     env->interrupt_injected = -1;
2187     for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
2188         if (sregs.interrupt_bitmap[i]) {
2189             bit = ctz64(sregs.interrupt_bitmap[i]);
2190             env->interrupt_injected = i * 64 + bit;
2191             break;
2192         }
2193     }
2194
2195     get_seg(&env->segs[R_CS], &sregs.cs);
2196     get_seg(&env->segs[R_DS], &sregs.ds);
2197     get_seg(&env->segs[R_ES], &sregs.es);
2198     get_seg(&env->segs[R_FS], &sregs.fs);
2199     get_seg(&env->segs[R_GS], &sregs.gs);
2200     get_seg(&env->segs[R_SS], &sregs.ss);
2201
2202     get_seg(&env->tr, &sregs.tr);
2203     get_seg(&env->ldt, &sregs.ldt);
2204
2205     env->idt.limit = sregs.idt.limit;
2206     env->idt.base = sregs.idt.base;
2207     env->gdt.limit = sregs.gdt.limit;
2208     env->gdt.base = sregs.gdt.base;
2209
2210     env->cr[0] = sregs.cr0;
2211     env->cr[2] = sregs.cr2;
2212     env->cr[3] = sregs.cr3;
2213     env->cr[4] = sregs.cr4;
2214
2215     env->efer = sregs.efer;
2216
2217     /* changes to apic base and cr8/tpr are read back via kvm_arch_post_run */
2218     x86_update_hflags(env);
2219
2220     return 0;
2221 }
2222
2223 static int kvm_get_msrs(X86CPU *cpu)
2224 {
2225     CPUX86State *env = &cpu->env;
2226     struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
2227     int ret, i;
2228     uint64_t mtrr_top_bits;
2229
2230     kvm_msr_buf_reset(cpu);
2231
2232     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_CS, 0);
2233     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_ESP, 0);
2234     kvm_msr_entry_add(cpu, MSR_IA32_SYSENTER_EIP, 0);
2235     kvm_msr_entry_add(cpu, MSR_PAT, 0);
2236     if (has_msr_star) {
2237         kvm_msr_entry_add(cpu, MSR_STAR, 0);
2238     }
2239     if (has_msr_hsave_pa) {
2240         kvm_msr_entry_add(cpu, MSR_VM_HSAVE_PA, 0);
2241     }
2242     if (has_msr_tsc_aux) {
2243         kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
2244     }
2245     if (has_msr_tsc_adjust) {
2246         kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
2247     }
2248     if (has_msr_tsc_deadline) {
2249         kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
2250     }
2251     if (has_msr_misc_enable) {
2252         kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE, 0);
2253     }
2254     if (has_msr_smbase) {
2255         kvm_msr_entry_add(cpu, MSR_IA32_SMBASE, 0);
2256     }
2257     if (has_msr_smi_count) {
2258         kvm_msr_entry_add(cpu, MSR_SMI_COUNT, 0);
2259     }
2260     if (has_msr_feature_control) {
2261         kvm_msr_entry_add(cpu, MSR_IA32_FEATURE_CONTROL, 0);
2262     }
2263     if (has_msr_bndcfgs) {
2264         kvm_msr_entry_add(cpu, MSR_IA32_BNDCFGS, 0);
2265     }
2266     if (has_msr_xss) {
2267         kvm_msr_entry_add(cpu, MSR_IA32_XSS, 0);
2268     }
2269     if (has_msr_spec_ctrl) {
2270         kvm_msr_entry_add(cpu, MSR_IA32_SPEC_CTRL, 0);
2271     }
2272     if (has_msr_virt_ssbd) {
2273         kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
2274     }
2275     if (!env->tsc_valid) {
2276         kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
2277         env->tsc_valid = !runstate_is_running();
2278     }
2279
2280 #ifdef TARGET_X86_64
2281     if (lm_capable_kernel) {
2282         kvm_msr_entry_add(cpu, MSR_CSTAR, 0);
2283         kvm_msr_entry_add(cpu, MSR_KERNELGSBASE, 0);
2284         kvm_msr_entry_add(cpu, MSR_FMASK, 0);
2285         kvm_msr_entry_add(cpu, MSR_LSTAR, 0);
2286     }
2287 #endif
2288     kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, 0);
2289     kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, 0);
2290     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF)) {
2291         kvm_msr_entry_add(cpu, MSR_KVM_ASYNC_PF_EN, 0);
2292     }
2293     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_PV_EOI)) {
2294         kvm_msr_entry_add(cpu, MSR_KVM_PV_EOI_EN, 0);
2295     }
2296     if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_STEAL_TIME)) {
2297         kvm_msr_entry_add(cpu, MSR_KVM_STEAL_TIME, 0);
2298     }
2299     if (has_architectural_pmu_version > 0) {
2300         if (has_architectural_pmu_version > 1) {
2301             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
2302             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_CTRL, 0);
2303             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_STATUS, 0);
2304             kvm_msr_entry_add(cpu, MSR_CORE_PERF_GLOBAL_OVF_CTRL, 0);
2305         }
2306         for (i = 0; i < num_architectural_pmu_fixed_counters; i++) {
2307             kvm_msr_entry_add(cpu, MSR_CORE_PERF_FIXED_CTR0 + i, 0);
2308         }
2309         for (i = 0; i < num_architectural_pmu_gp_counters; i++) {
2310             kvm_msr_entry_add(cpu, MSR_P6_PERFCTR0 + i, 0);
2311             kvm_msr_entry_add(cpu, MSR_P6_EVNTSEL0 + i, 0);
2312         }
2313     }
2314
2315     if (env->mcg_cap) {
2316         kvm_msr_entry_add(cpu, MSR_MCG_STATUS, 0);
2317         kvm_msr_entry_add(cpu, MSR_MCG_CTL, 0);
2318         if (has_msr_mcg_ext_ctl) {
2319             kvm_msr_entry_add(cpu, MSR_MCG_EXT_CTL, 0);
2320         }
2321         for (i = 0; i < (env->mcg_cap & 0xff) * 4; i++) {
2322             kvm_msr_entry_add(cpu, MSR_MC0_CTL + i, 0);
2323         }
2324     }
2325
2326     if (has_msr_hv_hypercall) {
2327         kvm_msr_entry_add(cpu, HV_X64_MSR_HYPERCALL, 0);
2328         kvm_msr_entry_add(cpu, HV_X64_MSR_GUEST_OS_ID, 0);
2329     }
2330     if (cpu->hyperv_vapic) {
2331         kvm_msr_entry_add(cpu, HV_X64_MSR_APIC_ASSIST_PAGE, 0);
2332     }
2333     if (cpu->hyperv_time) {
2334         kvm_msr_entry_add(cpu, HV_X64_MSR_REFERENCE_TSC, 0);
2335     }
2336     if (cpu->hyperv_reenlightenment) {
2337         kvm_msr_entry_add(cpu, HV_X64_MSR_REENLIGHTENMENT_CONTROL, 0);
2338         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_CONTROL, 0);
2339         kvm_msr_entry_add(cpu, HV_X64_MSR_TSC_EMULATION_STATUS, 0);
2340     }
2341     if (has_msr_hv_crash) {
2342         int j;
2343
2344         for (j = 0; j < HV_CRASH_PARAMS; j++) {
2345             kvm_msr_entry_add(cpu, HV_X64_MSR_CRASH_P0 + j, 0);
2346         }
2347     }
2348     if (has_msr_hv_runtime) {
2349         kvm_msr_entry_add(cpu, HV_X64_MSR_VP_RUNTIME, 0);
2350     }
2351     if (cpu->hyperv_synic) {
2352         uint32_t msr;
2353
2354         kvm_msr_entry_add(cpu, HV_X64_MSR_SCONTROL, 0);
2355         kvm_msr_entry_add(cpu, HV_X64_MSR_SIEFP, 0);
2356         kvm_msr_entry_add(cpu, HV_X64_MSR_SIMP, 0);
2357         for (msr = HV_X64_MSR_SINT0; msr <= HV_X64_MSR_SINT15; msr++) {
2358             kvm_msr_entry_add(cpu, msr, 0);
2359         }
2360     }
2361     if (has_msr_hv_stimer) {
2362         uint32_t msr;
2363
2364         for (msr = HV_X64_MSR_STIMER0_CONFIG; msr <= HV_X64_MSR_STIMER3_COUNT;
2365              msr++) {
2366             kvm_msr_entry_add(cpu, msr, 0);
2367         }
2368     }
2369     if (env->features[FEAT_1_EDX] & CPUID_MTRR) {
2370         kvm_msr_entry_add(cpu, MSR_MTRRdefType, 0);
2371         kvm_msr_entry_add(cpu, MSR_MTRRfix64K_00000, 0);
2372         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_80000, 0);
2373         kvm_msr_entry_add(cpu, MSR_MTRRfix16K_A0000, 0);
2374         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C0000, 0);
2375         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_C8000, 0);
2376         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D0000, 0);
2377         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_D8000, 0);
2378         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E0000, 0);
2379         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_E8000, 0);
2380         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F0000, 0);
2381         kvm_msr_entry_add(cpu, MSR_MTRRfix4K_F8000, 0);
2382         for (i = 0; i < MSR_MTRRcap_VCNT; i++) {
2383             kvm_msr_entry_add(cpu, MSR_MTRRphysBase(i), 0);
2384             kvm_msr_entry_add(cpu, MSR_MTRRphysMask(i), 0);
2385         }
2386     }
2387
2388     if (env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_INTEL_PT) {
2389         int addr_num =
2390             kvm_arch_get_supported_cpuid(kvm_state, 0x14, 1, R_EAX) & 0x7;
2391
2392         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CTL, 0);
2393         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_STATUS, 0);
2394         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_BASE, 0);
2395         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_OUTPUT_MASK, 0);
2396         kvm_msr_entry_add(cpu, MSR_IA32_RTIT_CR3_MATCH, 0);
2397         for (i = 0; i < addr_num; i++) {
2398             kvm_msr_entry_add(cpu, MSR_IA32_RTIT_ADDR0_A + i, 0);
2399         }
2400     }
2401
2402     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, cpu->kvm_msr_buf);
2403     if (ret < 0) {
2404         return ret;
2405     }
2406
2407     if (ret < cpu->kvm_msr_buf->nmsrs) {
2408         struct kvm_msr_entry *e = &cpu->kvm_msr_buf->entries[ret];
2409         error_report("error: failed to get MSR 0x%" PRIx32,
2410                      (uint32_t)e->index);
2411     }
2412
2413     assert(ret == cpu->kvm_msr_buf->nmsrs);
2414     /*
2415      * MTRR masks: Each mask consists of 5 parts
2416      * a  10..0: must be zero
2417      * b  11   : valid bit
2418      * c n-1.12: actual mask bits
2419      * d  51..n: reserved must be zero
2420      * e  63.52: reserved must be zero
2421      *
2422      * 'n' is the number of physical bits supported by the CPU and is
2423      * apparently always <= 52.   We know our 'n' but don't know what
2424      * the destinations 'n' is; it might be smaller, in which case
2425      * it masks (c) on loading. It might be larger, in which case
2426      * we fill 'd' so that d..c is consistent irrespetive of the 'n'
2427      * we're migrating to.
2428      */
2429
2430     if (cpu->fill_mtrr_mask) {
2431         QEMU_BUILD_BUG_ON(TARGET_PHYS_ADDR_SPACE_BITS > 52);
2432         assert(cpu->phys_bits <= TARGET_PHYS_ADDR_SPACE_BITS);
2433         mtrr_top_bits = MAKE_64BIT_MASK(cpu->phys_bits, 52 - cpu->phys_bits);
2434     } else {
2435         mtrr_top_bits = 0;
2436     }
2437
2438     for (i = 0; i < ret; i++) {
2439         uint32_t index = msrs[i].index;
2440         switch (index) {
2441         case MSR_IA32_SYSENTER_CS:
2442             env->sysenter_cs = msrs[i].data;
2443             break;
2444         case MSR_IA32_SYSENTER_ESP:
2445             env->sysenter_esp = msrs[i].data;
2446             break;
2447         case MSR_IA32_SYSENTER_EIP:
2448             env->sysenter_eip = msrs[i].data;
2449             break;
2450         case MSR_PAT:
2451             env->pat = msrs[i].data;
2452             break;
2453         case MSR_STAR:
2454             env->star = msrs[i].data;
2455             break;
2456 #ifdef TARGET_X86_64
2457         case MSR_CSTAR:
2458             env->cstar = msrs[i].data;
2459             break;
2460         case MSR_KERNELGSBASE:
2461             env->kernelgsbase = msrs[i].data;
2462             break;
2463         case MSR_FMASK:
2464             env->fmask = msrs[i].data;
2465             break;
2466         case MSR_LSTAR:
2467             env->lstar = msrs[i].data;
2468             break;
2469 #endif
2470         case MSR_IA32_TSC:
2471             env->tsc = msrs[i].data;
2472             break;
2473         case MSR_TSC_AUX:
2474             env->tsc_aux = msrs[i].data;
2475             break;
2476         case MSR_TSC_ADJUST:
2477             env->tsc_adjust = msrs[i].data;
2478             break;
2479         case MSR_IA32_TSCDEADLINE:
2480             env->tsc_deadline = msrs[i].data;
2481             break;
2482         case MSR_VM_HSAVE_PA:
2483             env->vm_hsave = msrs[i].data;
2484             break;
2485         case MSR_KVM_SYSTEM_TIME:
2486             env->system_time_msr = msrs[i].data;
2487             break;
2488         case MSR_KVM_WALL_CLOCK:
2489             env->wall_clock_msr = msrs[i].data;
2490             break;
2491         case MSR_MCG_STATUS:
2492             env->mcg_status = msrs[i].data;
2493             break;
2494         case MSR_MCG_CTL:
2495             env->mcg_ctl = msrs[i].data;
2496             break;
2497         case MSR_MCG_EXT_CTL:
2498             env->mcg_ext_ctl = msrs[i].data;
2499             break;
2500         case MSR_IA32_MISC_ENABLE:
2501             env->msr_ia32_misc_enable = msrs[i].data;
2502             break;
2503         case MSR_IA32_SMBASE:
2504             env->smbase = msrs[i].data;
2505             break;
2506         case MSR_SMI_COUNT:
2507             env->msr_smi_count = msrs[i].data;
2508             break;
2509         case MSR_IA32_FEATURE_CONTROL:
2510             env->msr_ia32_feature_control = msrs[i].data;
2511             break;
2512         case MSR_IA32_BNDCFGS:
2513             env->msr_bndcfgs = msrs[i].data;
2514             break;
2515         case MSR_IA32_XSS:
2516             env->xss = msrs[i].data;
2517             break;
2518         default:
2519             if (msrs[i].index >= MSR_MC0_CTL &&
2520                 msrs[i].index < MSR_MC0_CTL + (env->mcg_cap & 0xff) * 4) {
2521                 env->mce_banks[msrs[i].index - MSR_MC0_CTL] = msrs[i].data;
2522             }
2523             break;
2524         case MSR_KVM_ASYNC_PF_EN:
2525             env->async_pf_en_msr = msrs[i].data;
2526             break;
2527         case MSR_KVM_PV_EOI_EN:
2528             env->pv_eoi_en_msr = msrs[i].data;
2529             break;
2530         case MSR_KVM_STEAL_TIME:
2531             env->steal_time_msr = msrs[i].data;
2532             break;
2533         case MSR_CORE_PERF_FIXED_CTR_CTRL:
2534             env->msr_fixed_ctr_ctrl = msrs[i].data;
2535             break;
2536         case MSR_CORE_PERF_GLOBAL_CTRL:
2537             env->msr_global_ctrl = msrs[i].data;
2538             break;
2539         case MSR_CORE_PERF_GLOBAL_STATUS:
2540             env->msr_global_status = msrs[i].data;
2541             break;
2542         case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
2543             env->msr_global_ovf_ctrl = msrs[i].data;
2544             break;
2545         case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR0 + MAX_FIXED_COUNTERS - 1:
2546             env->msr_fixed_counters[index - MSR_CORE_PERF_FIXED_CTR0] = msrs[i].data;
2547             break;
2548         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR0 + MAX_GP_COUNTERS - 1:
2549             env->msr_gp_counters[index - MSR_P6_PERFCTR0] = msrs[i].data;
2550             break;
2551         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL0 + MAX_GP_COUNTERS - 1:
2552             env->msr_gp_evtsel[index - MSR_P6_EVNTSEL0] = msrs[i].data;
2553             break;
2554         case HV_X64_MSR_HYPERCALL:
2555             env->msr_hv_hypercall = msrs[i].data;
2556             break;
2557         case HV_X64_MSR_GUEST_OS_ID:
2558             env->msr_hv_guest_os_id = msrs[i].data;
2559             break;
2560         case HV_X64_MSR_APIC_ASSIST_PAGE:
2561             env->msr_hv_vapic = msrs[i].data;
2562             break;
2563         case HV_X64_MSR_REFERENCE_TSC:
2564             env->msr_hv_tsc = msrs[i].data;
2565             break;
2566         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2567             env->msr_hv_crash_params[index - HV_X64_MSR_CRASH_P0] = msrs[i].data;
2568             break;
2569         case HV_X64_MSR_VP_RUNTIME:
2570             env->msr_hv_runtime = msrs[i].data;
2571             break;
2572         case HV_X64_MSR_SCONTROL:
2573             env->msr_hv_synic_control = msrs[i].data;
2574             break;
2575         case HV_X64_MSR_SIEFP:
2576             env->msr_hv_synic_evt_page = msrs[i].data;
2577             break;
2578         case HV_X64_MSR_SIMP:
2579             env->msr_hv_synic_msg_page = msrs[i].data;
2580             break;
2581         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
2582             env->msr_hv_synic_sint[index - HV_X64_MSR_SINT0] = msrs[i].data;
2583             break;
2584         case HV_X64_MSR_STIMER0_CONFIG:
2585         case HV_X64_MSR_STIMER1_CONFIG:
2586         case HV_X64_MSR_STIMER2_CONFIG:
2587         case HV_X64_MSR_STIMER3_CONFIG:
2588             env->msr_hv_stimer_config[(index - HV_X64_MSR_STIMER0_CONFIG)/2] =
2589                                 msrs[i].data;
2590             break;
2591         case HV_X64_MSR_STIMER0_COUNT:
2592         case HV_X64_MSR_STIMER1_COUNT:
2593         case HV_X64_MSR_STIMER2_COUNT:
2594         case HV_X64_MSR_STIMER3_COUNT:
2595             env->msr_hv_stimer_count[(index - HV_X64_MSR_STIMER0_COUNT)/2] =
2596                                 msrs[i].data;
2597             break;
2598         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2599             env->msr_hv_reenlightenment_control = msrs[i].data;
2600             break;
2601         case HV_X64_MSR_TSC_EMULATION_CONTROL:
2602             env->msr_hv_tsc_emulation_control = msrs[i].data;
2603             break;
2604         case HV_X64_MSR_TSC_EMULATION_STATUS:
2605             env->msr_hv_tsc_emulation_status = msrs[i].data;
2606             break;
2607         case MSR_MTRRdefType:
2608             env->mtrr_deftype = msrs[i].data;
2609             break;
2610         case MSR_MTRRfix64K_00000:
2611             env->mtrr_fixed[0] = msrs[i].data;
2612             break;
2613         case MSR_MTRRfix16K_80000:
2614             env->mtrr_fixed[1] = msrs[i].data;
2615             break;
2616         case MSR_MTRRfix16K_A0000:
2617             env->mtrr_fixed[2] = msrs[i].data;
2618             break;
2619         case MSR_MTRRfix4K_C0000:
2620             env->mtrr_fixed[3] = msrs[i].data;
2621             break;
2622         case MSR_MTRRfix4K_C8000:
2623             env->mtrr_fixed[4] = msrs[i].data;
2624             break;
2625         case MSR_MTRRfix4K_D0000:
2626             env->mtrr_fixed[5] = msrs[i].data;
2627             break;
2628         case MSR_MTRRfix4K_D8000:
2629             env->mtrr_fixed[6] = msrs[i].data;
2630             break;
2631         case MSR_MTRRfix4K_E0000:
2632             env->mtrr_fixed[7] = msrs[i].data;
2633             break;
2634         case MSR_MTRRfix4K_E8000:
2635             env->mtrr_fixed[8] = msrs[i].data;
2636             break;
2637         case MSR_MTRRfix4K_F0000:
2638             env->mtrr_fixed[9] = msrs[i].data;
2639             break;
2640         case MSR_MTRRfix4K_F8000:
2641             env->mtrr_fixed[10] = msrs[i].data;
2642             break;
2643         case MSR_MTRRphysBase(0) ... MSR_MTRRphysMask(MSR_MTRRcap_VCNT - 1):
2644             if (index & 1) {
2645                 env->mtrr_var[MSR_MTRRphysIndex(index)].mask = msrs[i].data |
2646                                                                mtrr_top_bits;
2647             } else {
2648                 env->mtrr_var[MSR_MTRRphysIndex(index)].base = msrs[i].data;
2649             }
2650             break;
2651         case MSR_IA32_SPEC_CTRL:
2652             env->spec_ctrl = msrs[i].data;
2653             break;
2654         case MSR_VIRT_SSBD:
2655             env->virt_ssbd = msrs[i].data;
2656             break;
2657         case MSR_IA32_RTIT_CTL:
2658             env->msr_rtit_ctrl = msrs[i].data;
2659             break;
2660         case MSR_IA32_RTIT_STATUS:
2661             env->msr_rtit_status = msrs[i].data;
2662             break;
2663         case MSR_IA32_RTIT_OUTPUT_BASE:
2664             env->msr_rtit_output_base = msrs[i].data;
2665             break;
2666         case MSR_IA32_RTIT_OUTPUT_MASK:
2667             env->msr_rtit_output_mask = msrs[i].data;
2668             break;
2669         case MSR_IA32_RTIT_CR3_MATCH:
2670             env->msr_rtit_cr3_match = msrs[i].data;
2671             break;
2672         case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2673             env->msr_rtit_addrs[index - MSR_IA32_RTIT_ADDR0_A] = msrs[i].data;
2674             break;
2675         }
2676     }
2677
2678     return 0;
2679 }
2680
2681 static int kvm_put_mp_state(X86CPU *cpu)
2682 {
2683     struct kvm_mp_state mp_state = { .mp_state = cpu->env.mp_state };
2684
2685     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2686 }
2687
2688 static int kvm_get_mp_state(X86CPU *cpu)
2689 {
2690     CPUState *cs = CPU(cpu);
2691     CPUX86State *env = &cpu->env;
2692     struct kvm_mp_state mp_state;
2693     int ret;
2694
2695     ret = kvm_vcpu_ioctl(cs, KVM_GET_MP_STATE, &mp_state);
2696     if (ret < 0) {
2697         return ret;
2698     }
2699     env->mp_state = mp_state.mp_state;
2700     if (kvm_irqchip_in_kernel()) {
2701         cs->halted = (mp_state.mp_state == KVM_MP_STATE_HALTED);
2702     }
2703     return 0;
2704 }
2705
2706 static int kvm_get_apic(X86CPU *cpu)
2707 {
2708     DeviceState *apic = cpu->apic_state;
2709     struct kvm_lapic_state kapic;
2710     int ret;
2711
2712     if (apic && kvm_irqchip_in_kernel()) {
2713         ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_LAPIC, &kapic);
2714         if (ret < 0) {
2715             return ret;
2716         }
2717
2718         kvm_get_apic_state(apic, &kapic);
2719     }
2720     return 0;
2721 }
2722
2723 static int kvm_put_vcpu_events(X86CPU *cpu, int level)
2724 {
2725     CPUState *cs = CPU(cpu);
2726     CPUX86State *env = &cpu->env;
2727     struct kvm_vcpu_events events = {};
2728
2729     if (!kvm_has_vcpu_events()) {
2730         return 0;
2731     }
2732
2733     events.exception.injected = (env->exception_injected >= 0);
2734     events.exception.nr = env->exception_injected;
2735     events.exception.has_error_code = env->has_error_code;
2736     events.exception.error_code = env->error_code;
2737
2738     events.interrupt.injected = (env->interrupt_injected >= 0);
2739     events.interrupt.nr = env->interrupt_injected;
2740     events.interrupt.soft = env->soft_interrupt;
2741
2742     events.nmi.injected = env->nmi_injected;
2743     events.nmi.pending = env->nmi_pending;
2744     events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
2745
2746     events.sipi_vector = env->sipi_vector;
2747     events.flags = 0;
2748
2749     if (has_msr_smbase) {
2750         events.smi.smm = !!(env->hflags & HF_SMM_MASK);
2751         events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);
2752         if (kvm_irqchip_in_kernel()) {
2753             /* As soon as these are moved to the kernel, remove them
2754              * from cs->interrupt_request.
2755              */
2756             events.smi.pending = cs->interrupt_request & CPU_INTERRUPT_SMI;
2757             events.smi.latched_init = cs->interrupt_request & CPU_INTERRUPT_INIT;
2758             cs->interrupt_request &= ~(CPU_INTERRUPT_INIT | CPU_INTERRUPT_SMI);
2759         } else {
2760             /* Keep these in cs->interrupt_request.  */
2761             events.smi.pending = 0;
2762             events.smi.latched_init = 0;
2763         }
2764         /* Stop SMI delivery on old machine types to avoid a reboot
2765          * on an inward migration of an old VM.
2766          */
2767         if (!cpu->kvm_no_smi_migration) {
2768             events.flags |= KVM_VCPUEVENT_VALID_SMM;
2769         }
2770     }
2771
2772     if (level >= KVM_PUT_RESET_STATE) {
2773         events.flags |= KVM_VCPUEVENT_VALID_NMI_PENDING;
2774         if (env->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
2775             events.flags |= KVM_VCPUEVENT_VALID_SIPI_VECTOR;
2776         }
2777     }
2778
2779     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
2780 }
2781
2782 static int kvm_get_vcpu_events(X86CPU *cpu)
2783 {
2784     CPUX86State *env = &cpu->env;
2785     struct kvm_vcpu_events events;
2786     int ret;
2787
2788     if (!kvm_has_vcpu_events()) {
2789         return 0;
2790     }
2791
2792     memset(&events, 0, sizeof(events));
2793     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
2794     if (ret < 0) {
2795        return ret;
2796     }
2797     env->exception_injected =
2798        events.exception.injected ? events.exception.nr : -1;
2799     env->has_error_code = events.exception.has_error_code;
2800     env->error_code = events.exception.error_code;
2801
2802     env->interrupt_injected =
2803         events.interrupt.injected ? events.interrupt.nr : -1;
2804     env->soft_interrupt = events.interrupt.soft;
2805
2806     env->nmi_injected = events.nmi.injected;
2807     env->nmi_pending = events.nmi.pending;
2808     if (events.nmi.masked) {
2809         env->hflags2 |= HF2_NMI_MASK;
2810     } else {
2811         env->hflags2 &= ~HF2_NMI_MASK;
2812     }
2813
2814     if (events.flags & KVM_VCPUEVENT_VALID_SMM) {
2815         if (events.smi.smm) {
2816             env->hflags |= HF_SMM_MASK;
2817         } else {
2818             env->hflags &= ~HF_SMM_MASK;
2819         }
2820         if (events.smi.pending) {
2821             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2822         } else {
2823             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_SMI);
2824         }
2825         if (events.smi.smm_inside_nmi) {
2826             env->hflags2 |= HF2_SMM_INSIDE_NMI_MASK;
2827         } else {
2828             env->hflags2 &= ~HF2_SMM_INSIDE_NMI_MASK;
2829         }
2830         if (events.smi.latched_init) {
2831             cpu_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2832         } else {
2833             cpu_reset_interrupt(CPU(cpu), CPU_INTERRUPT_INIT);
2834         }
2835     }
2836
2837     env->sipi_vector = events.sipi_vector;
2838
2839     return 0;
2840 }
2841
2842 static int kvm_guest_debug_workarounds(X86CPU *cpu)
2843 {
2844     CPUState *cs = CPU(cpu);
2845     CPUX86State *env = &cpu->env;
2846     int ret = 0;
2847     unsigned long reinject_trap = 0;
2848
2849     if (!kvm_has_vcpu_events()) {
2850         if (env->exception_injected == 1) {
2851             reinject_trap = KVM_GUESTDBG_INJECT_DB;
2852         } else if (env->exception_injected == 3) {
2853             reinject_trap = KVM_GUESTDBG_INJECT_BP;
2854         }
2855         env->exception_injected = -1;
2856     }
2857
2858     /*
2859      * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
2860      * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
2861      * by updating the debug state once again if single-stepping is on.
2862      * Another reason to call kvm_update_guest_debug here is a pending debug
2863      * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
2864      * reinject them via SET_GUEST_DEBUG.
2865      */
2866     if (reinject_trap ||
2867         (!kvm_has_robust_singlestep() && cs->singlestep_enabled)) {
2868         ret = kvm_update_guest_debug(cs, reinject_trap);
2869     }
2870     return ret;
2871 }
2872
2873 static int kvm_put_debugregs(X86CPU *cpu)
2874 {
2875     CPUX86State *env = &cpu->env;
2876     struct kvm_debugregs dbgregs;
2877     int i;
2878
2879     if (!kvm_has_debugregs()) {
2880         return 0;
2881     }
2882
2883     for (i = 0; i < 4; i++) {
2884         dbgregs.db[i] = env->dr[i];
2885     }
2886     dbgregs.dr6 = env->dr[6];
2887     dbgregs.dr7 = env->dr[7];
2888     dbgregs.flags = 0;
2889
2890     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_DEBUGREGS, &dbgregs);
2891 }
2892
2893 static int kvm_get_debugregs(X86CPU *cpu)
2894 {
2895     CPUX86State *env = &cpu->env;
2896     struct kvm_debugregs dbgregs;
2897     int i, ret;
2898
2899     if (!kvm_has_debugregs()) {
2900         return 0;
2901     }
2902
2903     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_DEBUGREGS, &dbgregs);
2904     if (ret < 0) {
2905         return ret;
2906     }
2907     for (i = 0; i < 4; i++) {
2908         env->dr[i] = dbgregs.db[i];
2909     }
2910     env->dr[4] = env->dr[6] = dbgregs.dr6;
2911     env->dr[5] = env->dr[7] = dbgregs.dr7;
2912
2913     return 0;
2914 }
2915
2916 int kvm_arch_put_registers(CPUState *cpu, int level)
2917 {
2918     X86CPU *x86_cpu = X86_CPU(cpu);
2919     int ret;
2920
2921     assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
2922
2923     if (level >= KVM_PUT_RESET_STATE) {
2924         ret = kvm_put_msr_feature_control(x86_cpu);
2925         if (ret < 0) {
2926             return ret;
2927         }
2928     }
2929
2930     if (level == KVM_PUT_FULL_STATE) {
2931         /* We don't check for kvm_arch_set_tsc_khz() errors here,
2932          * because TSC frequency mismatch shouldn't abort migration,
2933          * unless the user explicitly asked for a more strict TSC
2934          * setting (e.g. using an explicit "tsc-freq" option).
2935          */
2936         kvm_arch_set_tsc_khz(cpu);
2937     }
2938
2939     ret = kvm_getput_regs(x86_cpu, 1);
2940     if (ret < 0) {
2941         return ret;
2942     }
2943     ret = kvm_put_xsave(x86_cpu);
2944     if (ret < 0) {
2945         return ret;
2946     }
2947     ret = kvm_put_xcrs(x86_cpu);
2948     if (ret < 0) {
2949         return ret;
2950     }
2951     ret = kvm_put_sregs(x86_cpu);
2952     if (ret < 0) {
2953         return ret;
2954     }
2955     /* must be before kvm_put_msrs */
2956     ret = kvm_inject_mce_oldstyle(x86_cpu);
2957     if (ret < 0) {
2958         return ret;
2959     }
2960     ret = kvm_put_msrs(x86_cpu, level);
2961     if (ret < 0) {
2962         return ret;
2963     }
2964     ret = kvm_put_vcpu_events(x86_cpu, level);
2965     if (ret < 0) {
2966         return ret;
2967     }
2968     if (level >= KVM_PUT_RESET_STATE) {
2969         ret = kvm_put_mp_state(x86_cpu);
2970         if (ret < 0) {
2971             return ret;
2972         }
2973     }
2974
2975     ret = kvm_put_tscdeadline_msr(x86_cpu);
2976     if (ret < 0) {
2977         return ret;
2978     }
2979     ret = kvm_put_debugregs(x86_cpu);
2980     if (ret < 0) {
2981         return ret;
2982     }
2983     /* must be last */
2984     ret = kvm_guest_debug_workarounds(x86_cpu);
2985     if (ret < 0) {
2986         return ret;
2987     }
2988     return 0;
2989 }
2990
2991 int kvm_arch_get_registers(CPUState *cs)
2992 {
2993     X86CPU *cpu = X86_CPU(cs);
2994     int ret;
2995
2996     assert(cpu_is_stopped(cs) || qemu_cpu_is_self(cs));
2997
2998     ret = kvm_get_vcpu_events(cpu);
2999     if (ret < 0) {
3000         goto out;
3001     }
3002     /*
3003      * KVM_GET_MPSTATE can modify CS and RIP, call it before
3004      * KVM_GET_REGS and KVM_GET_SREGS.
3005      */
3006     ret = kvm_get_mp_state(cpu);
3007     if (ret < 0) {
3008         goto out;
3009     }
3010     ret = kvm_getput_regs(cpu, 0);
3011     if (ret < 0) {
3012         goto out;
3013     }
3014     ret = kvm_get_xsave(cpu);
3015     if (ret < 0) {
3016         goto out;
3017     }
3018     ret = kvm_get_xcrs(cpu);
3019     if (ret < 0) {
3020         goto out;
3021     }
3022     ret = kvm_get_sregs(cpu);
3023     if (ret < 0) {
3024         goto out;
3025     }
3026     ret = kvm_get_msrs(cpu);
3027     if (ret < 0) {
3028         goto out;
3029     }
3030     ret = kvm_get_apic(cpu);
3031     if (ret < 0) {
3032         goto out;
3033     }
3034     ret = kvm_get_debugregs(cpu);
3035     if (ret < 0) {
3036         goto out;
3037     }
3038     ret = 0;
3039  out:
3040     cpu_sync_bndcs_hflags(&cpu->env);
3041     return ret;
3042 }
3043
3044 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
3045 {
3046     X86CPU *x86_cpu = X86_CPU(cpu);
3047     CPUX86State *env = &x86_cpu->env;
3048     int ret;
3049
3050     /* Inject NMI */
3051     if (cpu->interrupt_request & (CPU_INTERRUPT_NMI | CPU_INTERRUPT_SMI)) {
3052         if (cpu->interrupt_request & CPU_INTERRUPT_NMI) {
3053             qemu_mutex_lock_iothread();
3054             cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
3055             qemu_mutex_unlock_iothread();
3056             DPRINTF("injected NMI\n");
3057             ret = kvm_vcpu_ioctl(cpu, KVM_NMI);
3058             if (ret < 0) {
3059                 fprintf(stderr, "KVM: injection failed, NMI lost (%s)\n",
3060                         strerror(-ret));
3061             }
3062         }
3063         if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
3064             qemu_mutex_lock_iothread();
3065             cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
3066             qemu_mutex_unlock_iothread();
3067             DPRINTF("injected SMI\n");
3068             ret = kvm_vcpu_ioctl(cpu, KVM_SMI);
3069             if (ret < 0) {
3070                 fprintf(stderr, "KVM: injection failed, SMI lost (%s)\n",
3071                         strerror(-ret));
3072             }
3073         }
3074     }
3075
3076     if (!kvm_pic_in_kernel()) {
3077         qemu_mutex_lock_iothread();
3078     }
3079
3080     /* Force the VCPU out of its inner loop to process any INIT requests
3081      * or (for userspace APIC, but it is cheap to combine the checks here)
3082      * pending TPR access reports.
3083      */
3084     if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
3085         if ((cpu->interrupt_request & CPU_INTERRUPT_INIT) &&
3086             !(env->hflags & HF_SMM_MASK)) {
3087             cpu->exit_request = 1;
3088         }
3089         if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
3090             cpu->exit_request = 1;
3091         }
3092     }
3093
3094     if (!kvm_pic_in_kernel()) {
3095         /* Try to inject an interrupt if the guest can accept it */
3096         if (run->ready_for_interrupt_injection &&
3097             (cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
3098             (env->eflags & IF_MASK)) {
3099             int irq;
3100
3101             cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
3102             irq = cpu_get_pic_interrupt(env);
3103             if (irq >= 0) {
3104                 struct kvm_interrupt intr;
3105
3106                 intr.irq = irq;
3107                 DPRINTF("injected interrupt %d\n", irq);
3108                 ret = kvm_vcpu_ioctl(cpu, KVM_INTERRUPT, &intr);
3109                 if (ret < 0) {
3110                     fprintf(stderr,
3111                             "KVM: injection failed, interrupt lost (%s)\n",
3112                             strerror(-ret));
3113                 }
3114             }
3115         }
3116
3117         /* If we have an interrupt but the guest is not ready to receive an
3118          * interrupt, request an interrupt window exit.  This will
3119          * cause a return to userspace as soon as the guest is ready to
3120          * receive interrupts. */
3121         if ((cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
3122             run->request_interrupt_window = 1;
3123         } else {
3124             run->request_interrupt_window = 0;
3125         }
3126
3127         DPRINTF("setting tpr\n");
3128         run->cr8 = cpu_get_apic_tpr(x86_cpu->apic_state);
3129
3130         qemu_mutex_unlock_iothread();
3131     }
3132 }
3133
3134 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
3135 {
3136     X86CPU *x86_cpu = X86_CPU(cpu);
3137     CPUX86State *env = &x86_cpu->env;
3138
3139     if (run->flags & KVM_RUN_X86_SMM) {
3140         env->hflags |= HF_SMM_MASK;
3141     } else {
3142         env->hflags &= ~HF_SMM_MASK;
3143     }
3144     if (run->if_flag) {
3145         env->eflags |= IF_MASK;
3146     } else {
3147         env->eflags &= ~IF_MASK;
3148     }
3149
3150     /* We need to protect the apic state against concurrent accesses from
3151      * different threads in case the userspace irqchip is used. */
3152     if (!kvm_irqchip_in_kernel()) {
3153         qemu_mutex_lock_iothread();
3154     }
3155     cpu_set_apic_tpr(x86_cpu->apic_state, run->cr8);
3156     cpu_set_apic_base(x86_cpu->apic_state, run->apic_base);
3157     if (!kvm_irqchip_in_kernel()) {
3158         qemu_mutex_unlock_iothread();
3159     }
3160     return cpu_get_mem_attrs(env);
3161 }
3162
3163 int kvm_arch_process_async_events(CPUState *cs)
3164 {
3165     X86CPU *cpu = X86_CPU(cs);
3166     CPUX86State *env = &cpu->env;
3167
3168     if (cs->interrupt_request & CPU_INTERRUPT_MCE) {
3169         /* We must not raise CPU_INTERRUPT_MCE if it's not supported. */
3170         assert(env->mcg_cap);
3171
3172         cs->interrupt_request &= ~CPU_INTERRUPT_MCE;
3173
3174         kvm_cpu_synchronize_state(cs);
3175
3176         if (env->exception_injected == EXCP08_DBLE) {
3177             /* this means triple fault */
3178             qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
3179             cs->exit_request = 1;
3180             return 0;
3181         }
3182         env->exception_injected = EXCP12_MCHK;
3183         env->has_error_code = 0;
3184
3185         cs->halted = 0;
3186         if (kvm_irqchip_in_kernel() && env->mp_state == KVM_MP_STATE_HALTED) {
3187             env->mp_state = KVM_MP_STATE_RUNNABLE;
3188         }
3189     }
3190
3191     if ((cs->interrupt_request & CPU_INTERRUPT_INIT) &&
3192         !(env->hflags & HF_SMM_MASK)) {
3193         kvm_cpu_synchronize_state(cs);
3194         do_cpu_init(cpu);
3195     }
3196
3197     if (kvm_irqchip_in_kernel()) {
3198         return 0;
3199     }
3200
3201     if (cs->interrupt_request & CPU_INTERRUPT_POLL) {
3202         cs->interrupt_request &= ~CPU_INTERRUPT_POLL;
3203         apic_poll_irq(cpu->apic_state);
3204     }
3205     if (((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3206          (env->eflags & IF_MASK)) ||
3207         (cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3208         cs->halted = 0;
3209     }
3210     if (cs->interrupt_request & CPU_INTERRUPT_SIPI) {
3211         kvm_cpu_synchronize_state(cs);
3212         do_cpu_sipi(cpu);
3213     }
3214     if (cs->interrupt_request & CPU_INTERRUPT_TPR) {
3215         cs->interrupt_request &= ~CPU_INTERRUPT_TPR;
3216         kvm_cpu_synchronize_state(cs);
3217         apic_handle_tpr_access_report(cpu->apic_state, env->eip,
3218                                       env->tpr_access_type);
3219     }
3220
3221     return cs->halted;
3222 }
3223
3224 static int kvm_handle_halt(X86CPU *cpu)
3225 {
3226     CPUState *cs = CPU(cpu);
3227     CPUX86State *env = &cpu->env;
3228
3229     if (!((cs->interrupt_request & CPU_INTERRUPT_HARD) &&
3230           (env->eflags & IF_MASK)) &&
3231         !(cs->interrupt_request & CPU_INTERRUPT_NMI)) {
3232         cs->halted = 1;
3233         return EXCP_HLT;
3234     }
3235
3236     return 0;
3237 }
3238
3239 static int kvm_handle_tpr_access(X86CPU *cpu)
3240 {
3241     CPUState *cs = CPU(cpu);
3242     struct kvm_run *run = cs->kvm_run;
3243
3244     apic_handle_tpr_access_report(cpu->apic_state, run->tpr_access.rip,
3245                                   run->tpr_access.is_write ? TPR_ACCESS_WRITE
3246                                                            : TPR_ACCESS_READ);
3247     return 1;
3248 }
3249
3250 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3251 {
3252     static const uint8_t int3 = 0xcc;
3253
3254     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
3255         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&int3, 1, 1)) {
3256         return -EINVAL;
3257     }
3258     return 0;
3259 }
3260
3261 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
3262 {
3263     uint8_t int3;
3264
3265     if (cpu_memory_rw_debug(cs, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
3266         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
3267         return -EINVAL;
3268     }
3269     return 0;
3270 }
3271
3272 static struct {
3273     target_ulong addr;
3274     int len;
3275     int type;
3276 } hw_breakpoint[4];
3277
3278 static int nb_hw_breakpoint;
3279
3280 static int find_hw_breakpoint(target_ulong addr, int len, int type)
3281 {
3282     int n;
3283
3284     for (n = 0; n < nb_hw_breakpoint; n++) {
3285         if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
3286             (hw_breakpoint[n].len == len || len == -1)) {
3287             return n;
3288         }
3289     }
3290     return -1;
3291 }
3292
3293 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
3294                                   target_ulong len, int type)
3295 {
3296     switch (type) {
3297     case GDB_BREAKPOINT_HW:
3298         len = 1;
3299         break;
3300     case GDB_WATCHPOINT_WRITE:
3301     case GDB_WATCHPOINT_ACCESS:
3302         switch (len) {
3303         case 1:
3304             break;
3305         case 2:
3306         case 4:
3307         case 8:
3308             if (addr & (len - 1)) {
3309                 return -EINVAL;
3310             }
3311             break;
3312         default:
3313             return -EINVAL;
3314         }
3315         break;
3316     default:
3317         return -ENOSYS;
3318     }
3319
3320     if (nb_hw_breakpoint == 4) {
3321         return -ENOBUFS;
3322     }
3323     if (find_hw_breakpoint(addr, len, type) >= 0) {
3324         return -EEXIST;
3325     }
3326     hw_breakpoint[nb_hw_breakpoint].addr = addr;
3327     hw_breakpoint[nb_hw_breakpoint].len = len;
3328     hw_breakpoint[nb_hw_breakpoint].type = type;
3329     nb_hw_breakpoint++;
3330
3331     return 0;
3332 }
3333
3334 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
3335                                   target_ulong len, int type)
3336 {
3337     int n;
3338
3339     n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
3340     if (n < 0) {
3341         return -ENOENT;
3342     }
3343     nb_hw_breakpoint--;
3344     hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
3345
3346     return 0;
3347 }
3348
3349 void kvm_arch_remove_all_hw_breakpoints(void)
3350 {
3351     nb_hw_breakpoint = 0;
3352 }
3353
3354 static CPUWatchpoint hw_watchpoint;
3355
3356 static int kvm_handle_debug(X86CPU *cpu,
3357                             struct kvm_debug_exit_arch *arch_info)
3358 {
3359     CPUState *cs = CPU(cpu);
3360     CPUX86State *env = &cpu->env;
3361     int ret = 0;
3362     int n;
3363
3364     if (arch_info->exception == 1) {
3365         if (arch_info->dr6 & (1 << 14)) {
3366             if (cs->singlestep_enabled) {
3367                 ret = EXCP_DEBUG;
3368             }
3369         } else {
3370             for (n = 0; n < 4; n++) {
3371                 if (arch_info->dr6 & (1 << n)) {
3372                     switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
3373                     case 0x0:
3374                         ret = EXCP_DEBUG;
3375                         break;
3376                     case 0x1:
3377                         ret = EXCP_DEBUG;
3378                         cs->watchpoint_hit = &hw_watchpoint;
3379                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3380                         hw_watchpoint.flags = BP_MEM_WRITE;
3381                         break;
3382                     case 0x3:
3383                         ret = EXCP_DEBUG;
3384                         cs->watchpoint_hit = &hw_watchpoint;
3385                         hw_watchpoint.vaddr = hw_breakpoint[n].addr;
3386                         hw_watchpoint.flags = BP_MEM_ACCESS;
3387                         break;
3388                     }
3389                 }
3390             }
3391         }
3392     } else if (kvm_find_sw_breakpoint(cs, arch_info->pc)) {
3393         ret = EXCP_DEBUG;
3394     }
3395     if (ret == 0) {
3396         cpu_synchronize_state(cs);
3397         assert(env->exception_injected == -1);
3398
3399         /* pass to guest */
3400         env->exception_injected = arch_info->exception;
3401         env->has_error_code = 0;
3402     }
3403
3404     return ret;
3405 }
3406
3407 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
3408 {
3409     const uint8_t type_code[] = {
3410         [GDB_BREAKPOINT_HW] = 0x0,
3411         [GDB_WATCHPOINT_WRITE] = 0x1,
3412         [GDB_WATCHPOINT_ACCESS] = 0x3
3413     };
3414     const uint8_t len_code[] = {
3415         [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
3416     };
3417     int n;
3418
3419     if (kvm_sw_breakpoints_active(cpu)) {
3420         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
3421     }
3422     if (nb_hw_breakpoint > 0) {
3423         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
3424         dbg->arch.debugreg[7] = 0x0600;
3425         for (n = 0; n < nb_hw_breakpoint; n++) {
3426             dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
3427             dbg->arch.debugreg[7] |= (2 << (n * 2)) |
3428                 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
3429                 ((uint32_t)len_code[hw_breakpoint[n].len] << (18 + n*4));
3430         }
3431     }
3432 }
3433
3434 static bool host_supports_vmx(void)
3435 {
3436     uint32_t ecx, unused;
3437
3438     host_cpuid(1, 0, &unused, &unused, &ecx, &unused);
3439     return ecx & CPUID_EXT_VMX;
3440 }
3441
3442 #define VMX_INVALID_GUEST_STATE 0x80000021
3443
3444 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
3445 {
3446     X86CPU *cpu = X86_CPU(cs);
3447     uint64_t code;
3448     int ret;
3449
3450     switch (run->exit_reason) {
3451     case KVM_EXIT_HLT:
3452         DPRINTF("handle_hlt\n");
3453         qemu_mutex_lock_iothread();
3454         ret = kvm_handle_halt(cpu);
3455         qemu_mutex_unlock_iothread();
3456         break;
3457     case KVM_EXIT_SET_TPR:
3458         ret = 0;
3459         break;
3460     case KVM_EXIT_TPR_ACCESS:
3461         qemu_mutex_lock_iothread();
3462         ret = kvm_handle_tpr_access(cpu);
3463         qemu_mutex_unlock_iothread();
3464         break;
3465     case KVM_EXIT_FAIL_ENTRY:
3466         code = run->fail_entry.hardware_entry_failure_reason;
3467         fprintf(stderr, "KVM: entry failed, hardware error 0x%" PRIx64 "\n",
3468                 code);
3469         if (host_supports_vmx() && code == VMX_INVALID_GUEST_STATE) {
3470             fprintf(stderr,
3471                     "\nIf you're running a guest on an Intel machine without "
3472                         "unrestricted mode\n"
3473                     "support, the failure can be most likely due to the guest "
3474                         "entering an invalid\n"
3475                     "state for Intel VT. For example, the guest maybe running "
3476                         "in big real mode\n"
3477                     "which is not supported on less recent Intel processors."
3478                         "\n\n");
3479         }
3480         ret = -1;
3481         break;
3482     case KVM_EXIT_EXCEPTION:
3483         fprintf(stderr, "KVM: exception %d exit (error code 0x%x)\n",
3484                 run->ex.exception, run->ex.error_code);
3485         ret = -1;
3486         break;
3487     case KVM_EXIT_DEBUG:
3488         DPRINTF("kvm_exit_debug\n");
3489         qemu_mutex_lock_iothread();
3490         ret = kvm_handle_debug(cpu, &run->debug.arch);
3491         qemu_mutex_unlock_iothread();
3492         break;
3493     case KVM_EXIT_HYPERV:
3494         ret = kvm_hv_handle_exit(cpu, &run->hyperv);
3495         break;
3496     case KVM_EXIT_IOAPIC_EOI:
3497         ioapic_eoi_broadcast(run->eoi.vector);
3498         ret = 0;
3499         break;
3500     default:
3501         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
3502         ret = -1;
3503         break;
3504     }
3505
3506     return ret;
3507 }
3508
3509 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
3510 {
3511     X86CPU *cpu = X86_CPU(cs);
3512     CPUX86State *env = &cpu->env;
3513
3514     kvm_cpu_synchronize_state(cs);
3515     return !(env->cr[0] & CR0_PE_MASK) ||
3516            ((env->segs[R_CS].selector  & 3) != 3);
3517 }
3518
3519 void kvm_arch_init_irq_routing(KVMState *s)
3520 {
3521     if (!kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
3522         /* If kernel can't do irq routing, interrupt source
3523          * override 0->2 cannot be set up as required by HPET.
3524          * So we have to disable it.
3525          */
3526         no_hpet = 1;
3527     }
3528     /* We know at this point that we're using the in-kernel
3529      * irqchip, so we can use irqfds, and on x86 we know
3530      * we can use msi via irqfd and GSI routing.
3531      */
3532     kvm_msi_via_irqfd_allowed = true;
3533     kvm_gsi_routing_allowed = true;
3534
3535     if (kvm_irqchip_is_split()) {
3536         int i;
3537
3538         /* If the ioapic is in QEMU and the lapics are in KVM, reserve
3539            MSI routes for signaling interrupts to the local apics. */
3540         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
3541             if (kvm_irqchip_add_msi_route(s, 0, NULL) < 0) {
3542                 error_report("Could not enable split IRQ mode.");
3543                 exit(1);
3544             }
3545         }
3546     }
3547 }
3548
3549 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
3550 {
3551     int ret;
3552     if (machine_kernel_irqchip_split(ms)) {
3553         ret = kvm_vm_enable_cap(s, KVM_CAP_SPLIT_IRQCHIP, 0, 24);
3554         if (ret) {
3555             error_report("Could not enable split irqchip mode: %s",
3556                          strerror(-ret));
3557             exit(1);
3558         } else {
3559             DPRINTF("Enabled KVM_CAP_SPLIT_IRQCHIP\n");
3560             kvm_split_irqchip = true;
3561             return 1;
3562         }
3563     } else {
3564         return 0;
3565     }
3566 }
3567
3568 /* Classic KVM device assignment interface. Will remain x86 only. */
3569 int kvm_device_pci_assign(KVMState *s, PCIHostDeviceAddress *dev_addr,
3570                           uint32_t flags, uint32_t *dev_id)
3571 {
3572     struct kvm_assigned_pci_dev dev_data = {
3573         .segnr = dev_addr->domain,
3574         .busnr = dev_addr->bus,
3575         .devfn = PCI_DEVFN(dev_addr->slot, dev_addr->function),
3576         .flags = flags,
3577     };
3578     int ret;
3579
3580     dev_data.assigned_dev_id =
3581         (dev_addr->domain << 16) | (dev_addr->bus << 8) | dev_data.devfn;
3582
3583     ret = kvm_vm_ioctl(s, KVM_ASSIGN_PCI_DEVICE, &dev_data);
3584     if (ret < 0) {
3585         return ret;
3586     }
3587
3588     *dev_id = dev_data.assigned_dev_id;
3589
3590     return 0;
3591 }
3592
3593 int kvm_device_pci_deassign(KVMState *s, uint32_t dev_id)
3594 {
3595     struct kvm_assigned_pci_dev dev_data = {
3596         .assigned_dev_id = dev_id,
3597     };
3598
3599     return kvm_vm_ioctl(s, KVM_DEASSIGN_PCI_DEVICE, &dev_data);
3600 }
3601
3602 static int kvm_assign_irq_internal(KVMState *s, uint32_t dev_id,
3603                                    uint32_t irq_type, uint32_t guest_irq)
3604 {
3605     struct kvm_assigned_irq assigned_irq = {
3606         .assigned_dev_id = dev_id,
3607         .guest_irq = guest_irq,
3608         .flags = irq_type,
3609     };
3610
3611     if (kvm_check_extension(s, KVM_CAP_ASSIGN_DEV_IRQ)) {
3612         return kvm_vm_ioctl(s, KVM_ASSIGN_DEV_IRQ, &assigned_irq);
3613     } else {
3614         return kvm_vm_ioctl(s, KVM_ASSIGN_IRQ, &assigned_irq);
3615     }
3616 }
3617
3618 int kvm_device_intx_assign(KVMState *s, uint32_t dev_id, bool use_host_msi,
3619                            uint32_t guest_irq)
3620 {
3621     uint32_t irq_type = KVM_DEV_IRQ_GUEST_INTX |
3622         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX);
3623
3624     return kvm_assign_irq_internal(s, dev_id, irq_type, guest_irq);
3625 }
3626
3627 int kvm_device_intx_set_mask(KVMState *s, uint32_t dev_id, bool masked)
3628 {
3629     struct kvm_assigned_pci_dev dev_data = {
3630         .assigned_dev_id = dev_id,
3631         .flags = masked ? KVM_DEV_ASSIGN_MASK_INTX : 0,
3632     };
3633
3634     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_INTX_MASK, &dev_data);
3635 }
3636
3637 static int kvm_deassign_irq_internal(KVMState *s, uint32_t dev_id,
3638                                      uint32_t type)
3639 {
3640     struct kvm_assigned_irq assigned_irq = {
3641         .assigned_dev_id = dev_id,
3642         .flags = type,
3643     };
3644
3645     return kvm_vm_ioctl(s, KVM_DEASSIGN_DEV_IRQ, &assigned_irq);
3646 }
3647
3648 int kvm_device_intx_deassign(KVMState *s, uint32_t dev_id, bool use_host_msi)
3649 {
3650     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_INTX |
3651         (use_host_msi ? KVM_DEV_IRQ_HOST_MSI : KVM_DEV_IRQ_HOST_INTX));
3652 }
3653
3654 int kvm_device_msi_assign(KVMState *s, uint32_t dev_id, int virq)
3655 {
3656     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSI |
3657                                               KVM_DEV_IRQ_GUEST_MSI, virq);
3658 }
3659
3660 int kvm_device_msi_deassign(KVMState *s, uint32_t dev_id)
3661 {
3662     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSI |
3663                                                 KVM_DEV_IRQ_HOST_MSI);
3664 }
3665
3666 bool kvm_device_msix_supported(KVMState *s)
3667 {
3668     /* The kernel lacks a corresponding KVM_CAP, so we probe by calling
3669      * KVM_ASSIGN_SET_MSIX_NR with an invalid parameter. */
3670     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, NULL) == -EFAULT;
3671 }
3672
3673 int kvm_device_msix_init_vectors(KVMState *s, uint32_t dev_id,
3674                                  uint32_t nr_vectors)
3675 {
3676     struct kvm_assigned_msix_nr msix_nr = {
3677         .assigned_dev_id = dev_id,
3678         .entry_nr = nr_vectors,
3679     };
3680
3681     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_NR, &msix_nr);
3682 }
3683
3684 int kvm_device_msix_set_vector(KVMState *s, uint32_t dev_id, uint32_t vector,
3685                                int virq)
3686 {
3687     struct kvm_assigned_msix_entry msix_entry = {
3688         .assigned_dev_id = dev_id,
3689         .gsi = virq,
3690         .entry = vector,
3691     };
3692
3693     return kvm_vm_ioctl(s, KVM_ASSIGN_SET_MSIX_ENTRY, &msix_entry);
3694 }
3695
3696 int kvm_device_msix_assign(KVMState *s, uint32_t dev_id)
3697 {
3698     return kvm_assign_irq_internal(s, dev_id, KVM_DEV_IRQ_HOST_MSIX |
3699                                               KVM_DEV_IRQ_GUEST_MSIX, 0);
3700 }
3701
3702 int kvm_device_msix_deassign(KVMState *s, uint32_t dev_id)
3703 {
3704     return kvm_deassign_irq_internal(s, dev_id, KVM_DEV_IRQ_GUEST_MSIX |
3705                                                 KVM_DEV_IRQ_HOST_MSIX);
3706 }
3707
3708 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
3709                              uint64_t address, uint32_t data, PCIDevice *dev)
3710 {
3711     X86IOMMUState *iommu = x86_iommu_get_default();
3712
3713     if (iommu) {
3714         int ret;
3715         MSIMessage src, dst;
3716         X86IOMMUClass *class = X86_IOMMU_GET_CLASS(iommu);
3717
3718         if (!class->int_remap) {
3719             return 0;
3720         }
3721
3722         src.address = route->u.msi.address_hi;
3723         src.address <<= VTD_MSI_ADDR_HI_SHIFT;
3724         src.address |= route->u.msi.address_lo;
3725         src.data = route->u.msi.data;
3726
3727         ret = class->int_remap(iommu, &src, &dst, dev ? \
3728                                pci_requester_id(dev) : \
3729                                X86_IOMMU_SID_INVALID);
3730         if (ret) {
3731             trace_kvm_x86_fixup_msi_error(route->gsi);
3732             return 1;
3733         }
3734
3735         route->u.msi.address_hi = dst.address >> VTD_MSI_ADDR_HI_SHIFT;
3736         route->u.msi.address_lo = dst.address & VTD_MSI_ADDR_LO_MASK;
3737         route->u.msi.data = dst.data;
3738     }
3739
3740     return 0;
3741 }
3742
3743 typedef struct MSIRouteEntry MSIRouteEntry;
3744
3745 struct MSIRouteEntry {
3746     PCIDevice *dev;             /* Device pointer */
3747     int vector;                 /* MSI/MSIX vector index */
3748     int virq;                   /* Virtual IRQ index */
3749     QLIST_ENTRY(MSIRouteEntry) list;
3750 };
3751
3752 /* List of used GSI routes */
3753 static QLIST_HEAD(, MSIRouteEntry) msi_route_list = \
3754     QLIST_HEAD_INITIALIZER(msi_route_list);
3755
3756 static void kvm_update_msi_routes_all(void *private, bool global,
3757                                       uint32_t index, uint32_t mask)
3758 {
3759     int cnt = 0;
3760     MSIRouteEntry *entry;
3761     MSIMessage msg;
3762     PCIDevice *dev;
3763
3764     /* TODO: explicit route update */
3765     QLIST_FOREACH(entry, &msi_route_list, list) {
3766         cnt++;
3767         dev = entry->dev;
3768         if (!msix_enabled(dev) && !msi_enabled(dev)) {
3769             continue;
3770         }
3771         msg = pci_get_msi_message(dev, entry->vector);
3772         kvm_irqchip_update_msi_route(kvm_state, entry->virq, msg, dev);
3773     }
3774     kvm_irqchip_commit_routes(kvm_state);
3775     trace_kvm_x86_update_msi_routes(cnt);
3776 }
3777
3778 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
3779                                 int vector, PCIDevice *dev)
3780 {
3781     static bool notify_list_inited = false;
3782     MSIRouteEntry *entry;
3783
3784     if (!dev) {
3785         /* These are (possibly) IOAPIC routes only used for split
3786          * kernel irqchip mode, while what we are housekeeping are
3787          * PCI devices only. */
3788         return 0;
3789     }
3790
3791     entry = g_new0(MSIRouteEntry, 1);
3792     entry->dev = dev;
3793     entry->vector = vector;
3794     entry->virq = route->gsi;
3795     QLIST_INSERT_HEAD(&msi_route_list, entry, list);
3796
3797     trace_kvm_x86_add_msi_route(route->gsi);
3798
3799     if (!notify_list_inited) {
3800         /* For the first time we do add route, add ourselves into
3801          * IOMMU's IEC notify list if needed. */
3802         X86IOMMUState *iommu = x86_iommu_get_default();
3803         if (iommu) {
3804             x86_iommu_iec_register_notifier(iommu,
3805                                             kvm_update_msi_routes_all,
3806                                             NULL);
3807         }
3808         notify_list_inited = true;
3809     }
3810     return 0;
3811 }
3812
3813 int kvm_arch_release_virq_post(int virq)
3814 {
3815     MSIRouteEntry *entry, *next;
3816     QLIST_FOREACH_SAFE(entry, &msi_route_list, list, next) {
3817         if (entry->virq == virq) {
3818             trace_kvm_x86_remove_msi_route(virq);
3819             QLIST_REMOVE(entry, list);
3820             g_free(entry);
3821             break;
3822         }
3823     }
3824     return 0;
3825 }
3826
3827 int kvm_arch_msi_data_to_gsi(uint32_t data)
3828 {
3829     abort();
3830 }