accel/tcg/cpu-exec: Use RCU_READ_LOCK_GUARD

[mirror_qemu.git] / accel / tcg / cpu-exec.c

/*
 *  emulator main execution loop
 *
 *  Copyright (c) 2003-2005 Fabrice Bellard
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
 */

#include "qemu/osdep.h"
#include "qemu/qemu-print.h"
#include "qapi/error.h"
#include "qapi/type-helpers.h"
#include "hw/core/tcg-cpu-ops.h"
#include "trace.h"
#include "disas/disas.h"
#include "exec/exec-all.h"
#include "tcg/tcg.h"
#include "qemu/atomic.h"
#include "qemu/rcu.h"
#include "exec/log.h"
#include "qemu/main-loop.h"
#if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
#include "hw/i386/apic.h"
#endif
#include "sysemu/cpus.h"
#include "exec/cpu-all.h"
#include "sysemu/cpu-timers.h"
#include "exec/replay-core.h"
#include "sysemu/tcg.h"
#include "exec/helper-proto-common.h"
#include "tb-jmp-cache.h"
#include "tb-hash.h"
#include "tb-context.h"
#include "internal-common.h"
#include "internal-target.h"

/* -icount align implementation. */

typedef struct SyncClocks {
    int64_t diff_clk;
    int64_t last_cpu_icount;
    int64_t realtime_clock;
} SyncClocks;

#if !defined(CONFIG_USER_ONLY)
/* Allow the guest to have a max 3ms advance.
 * The difference between the 2 clocks could therefore
 * oscillate around 0.
 */
#define VM_CLOCK_ADVANCE 3000000
#define THRESHOLD_REDUCE 1.5
#define MAX_DELAY_PRINT_RATE 2000000000LL
#define MAX_NB_PRINTS 100

int64_t max_delay;
int64_t max_advance;

static void align_clocks(SyncClocks *sc, CPUState *cpu)
{
    int64_t cpu_icount;

    if (!icount_align_option) {
        return;
    }

    cpu_icount = cpu->icount_extra + cpu->neg.icount_decr.u16.low;
    sc->diff_clk += icount_to_ns(sc->last_cpu_icount - cpu_icount);
    sc->last_cpu_icount = cpu_icount;

    if (sc->diff_clk > VM_CLOCK_ADVANCE) {
#ifndef _WIN32
        struct timespec sleep_delay, rem_delay;
        sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
        sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
        if (nanosleep(&sleep_delay, &rem_delay) < 0) {
            sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
        } else {
            sc->diff_clk = 0;
        }
#else
        Sleep(sc->diff_clk / SCALE_MS);
        sc->diff_clk = 0;
#endif
    }
}

static void print_delay(const SyncClocks *sc)
{
    static float threshold_delay;
    static int64_t last_realtime_clock;
    static int nb_prints;

    if (icount_align_option &&
        sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
        nb_prints < MAX_NB_PRINTS) {
        if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
            (-sc->diff_clk / (float)1000000000LL <
             (threshold_delay - THRESHOLD_REDUCE))) {
            threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
            qemu_printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
                        threshold_delay - 1,
                        threshold_delay);
            nb_prints++;
            last_realtime_clock = sc->realtime_clock;
        }
    }
}

static void init_delay_params(SyncClocks *sc, CPUState *cpu)
{
    if (!icount_align_option) {
        return;
    }
    sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
    sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
    sc->last_cpu_icount
        = cpu->icount_extra + cpu->neg.icount_decr.u16.low;
    if (sc->diff_clk < max_delay) {
        max_delay = sc->diff_clk;
    }
    if (sc->diff_clk > max_advance) {
        max_advance = sc->diff_clk;
    }

    /* Print every 2s max if the guest is late. We limit the number
       of printed messages to NB_PRINT_MAX(currently 100) */
    print_delay(sc);
}
#else
static void align_clocks(SyncClocks *sc, const CPUState *cpu)
{
}

static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
{
}
#endif /* CONFIG USER ONLY */

uint32_t curr_cflags(CPUState *cpu)
{
    uint32_t cflags = cpu->tcg_cflags;

    /*
     * Record gdb single-step.  We should be exiting the TB by raising
     * EXCP_DEBUG, but to simplify other tests, disable chaining too.
     *
     * For singlestep and -d nochain, suppress goto_tb so that
     * we can log -d cpu,exec after every TB.
     */
    if (unlikely(cpu->singlestep_enabled)) {
        cflags |= CF_NO_GOTO_TB | CF_NO_GOTO_PTR | CF_SINGLE_STEP | 1;
    } else if (qatomic_read(&one_insn_per_tb)) {
        cflags |= CF_NO_GOTO_TB | 1;
    } else if (qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
        cflags |= CF_NO_GOTO_TB;
    }

    return cflags;
}

struct tb_desc {
    vaddr pc;
    uint64_t cs_base;
    CPUArchState *env;
    tb_page_addr_t page_addr0;
    uint32_t flags;
    uint32_t cflags;
};

static bool tb_lookup_cmp(const void *p, const void *d)
{
    const TranslationBlock *tb = p;
    const struct tb_desc *desc = d;

    if ((tb_cflags(tb) & CF_PCREL || tb->pc == desc->pc) &&
        tb_page_addr0(tb) == desc->page_addr0 &&
        tb->cs_base == desc->cs_base &&
        tb->flags == desc->flags &&
        tb_cflags(tb) == desc->cflags) {
        /* check next page if needed */
        tb_page_addr_t tb_phys_page1 = tb_page_addr1(tb);
        if (tb_phys_page1 == -1) {
            return true;
        } else {
            tb_page_addr_t phys_page1;
            vaddr virt_page1;

            /*
             * We know that the first page matched, and an otherwise valid TB
             * encountered an incomplete instruction at the end of that page,
             * therefore we know that generating a new TB from the current PC
             * must also require reading from the next page -- even if the
             * second pages do not match, and therefore the resulting insn
             * is different for the new TB.  Therefore any exception raised
             * here by the faulting lookup is not premature.
             */
            virt_page1 = TARGET_PAGE_ALIGN(desc->pc);
            phys_page1 = get_page_addr_code(desc->env, virt_page1);
            if (tb_phys_page1 == phys_page1) {
                return true;
            }
        }
    }
    return false;
}

static TranslationBlock *tb_htable_lookup(CPUState *cpu, vaddr pc,
                                          uint64_t cs_base, uint32_t flags,
                                          uint32_t cflags)
{
    tb_page_addr_t phys_pc;
    struct tb_desc desc;
    uint32_t h;

    desc.env = cpu_env(cpu);
    desc.cs_base = cs_base;
    desc.flags = flags;
    desc.cflags = cflags;
    desc.pc = pc;
    phys_pc = get_page_addr_code(desc.env, pc);
    if (phys_pc == -1) {
        return NULL;
    }
    desc.page_addr0 = phys_pc;
    h = tb_hash_func(phys_pc, (cflags & CF_PCREL ? 0 : pc),
                     flags, cs_base, cflags);
    return qht_lookup_custom(&tb_ctx.htable, &desc, h, tb_lookup_cmp);
}

/* Might cause an exception, so have a longjmp destination ready */
static inline TranslationBlock *tb_lookup(CPUState *cpu, vaddr pc,
                                          uint64_t cs_base, uint32_t flags,
                                          uint32_t cflags)
{
    TranslationBlock *tb;
    CPUJumpCache *jc;
    uint32_t hash;

    /* we should never be trying to look up an INVALID tb */
    tcg_debug_assert(!(cflags & CF_INVALID));

    hash = tb_jmp_cache_hash_func(pc);
    jc = cpu->tb_jmp_cache;

    tb = qatomic_read(&jc->array[hash].tb);
    if (likely(tb &&
               jc->array[hash].pc == pc &&
               tb->cs_base == cs_base &&
               tb->flags == flags &&
               tb_cflags(tb) == cflags)) {
        goto hit;
    }

    tb = tb_htable_lookup(cpu, pc, cs_base, flags, cflags);
    if (tb == NULL) {
        return NULL;
    }

    jc->array[hash].pc = pc;
    qatomic_set(&jc->array[hash].tb, tb);

hit:
    /*
     * As long as tb is not NULL, the contents are consistent.  Therefore,
     * the virtual PC has to match for non-CF_PCREL translations.
     */
    assert((tb_cflags(tb) & CF_PCREL) || tb->pc == pc);
    return tb;
}

static void log_cpu_exec(vaddr pc, CPUState *cpu,
                         const TranslationBlock *tb)
{
    if (qemu_log_in_addr_range(pc)) {
        qemu_log_mask(CPU_LOG_EXEC,
                      "Trace %d: %p [%08" PRIx64
                      "/%016" VADDR_PRIx "/%08x/%08x] %s\n",
                      cpu->cpu_index, tb->tc.ptr, tb->cs_base, pc,
                      tb->flags, tb->cflags, lookup_symbol(pc));

        if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
            FILE *logfile = qemu_log_trylock();
            if (logfile) {
                int flags = 0;

                if (qemu_loglevel_mask(CPU_LOG_TB_FPU)) {
                    flags |= CPU_DUMP_FPU;
                }
#if defined(TARGET_I386)
                flags |= CPU_DUMP_CCOP;
#endif
                if (qemu_loglevel_mask(CPU_LOG_TB_VPU)) {
                    flags |= CPU_DUMP_VPU;
                }
                cpu_dump_state(cpu, logfile, flags);
                qemu_log_unlock(logfile);
            }
        }
    }
}

static bool check_for_breakpoints_slow(CPUState *cpu, vaddr pc,
                                       uint32_t *cflags)
{
    CPUBreakpoint *bp;
    bool match_page = false;

    /*
     * Singlestep overrides breakpoints.
     * This requirement is visible in the record-replay tests, where
     * we would fail to make forward progress in reverse-continue.
     *
     * TODO: gdb singlestep should only override gdb breakpoints,
     * so that one could (gdb) singlestep into the guest kernel's
     * architectural breakpoint handler.
     */
    if (cpu->singlestep_enabled) {
        return false;
    }

    QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
        /*
         * If we have an exact pc match, trigger the breakpoint.
         * Otherwise, note matches within the page.
         */
        if (pc == bp->pc) {
            bool match_bp = false;

            if (bp->flags & BP_GDB) {
                match_bp = true;
            } else if (bp->flags & BP_CPU) {
#ifdef CONFIG_USER_ONLY
                g_assert_not_reached();
#else
                CPUClass *cc = CPU_GET_CLASS(cpu);
                assert(cc->tcg_ops->debug_check_breakpoint);
                match_bp = cc->tcg_ops->debug_check_breakpoint(cpu);
#endif
            }

            if (match_bp) {
                cpu->exception_index = EXCP_DEBUG;
                return true;
            }
        } else if (((pc ^ bp->pc) & TARGET_PAGE_MASK) == 0) {
            match_page = true;
        }
    }

    /*
     * Within the same page as a breakpoint, single-step,
     * returning to helper_lookup_tb_ptr after each insn looking
     * for the actual breakpoint.
     *
     * TODO: Perhaps better to record all of the TBs associated
     * with a given virtual page that contains a breakpoint, and
     * then invalidate them when a new overlapping breakpoint is
     * set on the page.  Non-overlapping TBs would not be
     * invalidated, nor would any TB need to be invalidated as
     * breakpoints are removed.
     */
    if (match_page) {
        *cflags = (*cflags & ~CF_COUNT_MASK) | CF_NO_GOTO_TB | 1;
    }
    return false;
}

static inline bool check_for_breakpoints(CPUState *cpu, vaddr pc,
                                         uint32_t *cflags)
{
    return unlikely(!QTAILQ_EMPTY(&cpu->breakpoints)) &&
        check_for_breakpoints_slow(cpu, pc, cflags);
}

/**
 * helper_lookup_tb_ptr: quick check for next tb
 * @env: current cpu state
 *
 * Look for an existing TB matching the current cpu state.
 * If found, return the code pointer.  If not found, return
 * the tcg epilogue so that we return into cpu_tb_exec.
 */
const void *HELPER(lookup_tb_ptr)(CPUArchState *env)
{
    CPUState *cpu = env_cpu(env);
    TranslationBlock *tb;
    vaddr pc;
    uint64_t cs_base;
    uint32_t flags, cflags;

    cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);

    cflags = curr_cflags(cpu);
    if (check_for_breakpoints(cpu, pc, &cflags)) {
        cpu_loop_exit(cpu);
    }

    tb = tb_lookup(cpu, pc, cs_base, flags, cflags);
    if (tb == NULL) {
        return tcg_code_gen_epilogue;
    }

    if (qemu_loglevel_mask(CPU_LOG_TB_CPU | CPU_LOG_EXEC)) {
        log_cpu_exec(pc, cpu, tb);
    }

    return tb->tc.ptr;
}

/* Execute a TB, and fix up the CPU state afterwards if necessary */
/*
 * Disable CFI checks.
 * TCG creates binary blobs at runtime, with the transformed code.
 * A TB is a blob of binary code, created at runtime and called with an
 * indirect function call. Since such function did not exist at compile time,
 * the CFI runtime has no way to verify its signature and would fail.
 * TCG is not considered a security-sensitive part of QEMU so this does not
 * affect the impact of CFI in environment with high security requirements
 */
static inline TranslationBlock * QEMU_DISABLE_CFI
cpu_tb_exec(CPUState *cpu, TranslationBlock *itb, int *tb_exit)
{
    CPUArchState *env = cpu_env(cpu);
    uintptr_t ret;
    TranslationBlock *last_tb;
    const void *tb_ptr = itb->tc.ptr;

    if (qemu_loglevel_mask(CPU_LOG_TB_CPU | CPU_LOG_EXEC)) {
        log_cpu_exec(log_pc(cpu, itb), cpu, itb);
    }

    qemu_thread_jit_execute();
    ret = tcg_qemu_tb_exec(env, tb_ptr);
    cpu->neg.can_do_io = true;
    qemu_plugin_disable_mem_helpers(cpu);
    /*
     * TODO: Delay swapping back to the read-write region of the TB
     * until we actually need to modify the TB.  The read-only copy,
     * coming from the rx region, shares the same host TLB entry as
     * the code that executed the exit_tb opcode that arrived here.
     * If we insist on touching both the RX and the RW pages, we
     * double the host TLB pressure.
     */
    last_tb = tcg_splitwx_to_rw((void *)(ret & ~TB_EXIT_MASK));
    *tb_exit = ret & TB_EXIT_MASK;

    trace_exec_tb_exit(last_tb, *tb_exit);

    if (*tb_exit > TB_EXIT_IDX1) {
        /* We didn't start executing this TB (eg because the instruction
         * counter hit zero); we must restore the guest PC to the address
         * of the start of the TB.
         */
        CPUClass *cc = CPU_GET_CLASS(cpu);

        if (cc->tcg_ops->synchronize_from_tb) {
            cc->tcg_ops->synchronize_from_tb(cpu, last_tb);
        } else {
            tcg_debug_assert(!(tb_cflags(last_tb) & CF_PCREL));
            assert(cc->set_pc);
            cc->set_pc(cpu, last_tb->pc);
        }
        if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
            vaddr pc = log_pc(cpu, last_tb);
            if (qemu_log_in_addr_range(pc)) {
                qemu_log("Stopped execution of TB chain before %p [%016"
                         VADDR_PRIx "] %s\n",
                         last_tb->tc.ptr, pc, lookup_symbol(pc));
            }
        }
    }

    /*
     * If gdb single-step, and we haven't raised another exception,
     * raise a debug exception.  Single-step with another exception
     * is handled in cpu_handle_exception.
     */
    if (unlikely(cpu->singlestep_enabled) && cpu->exception_index == -1) {
        cpu->exception_index = EXCP_DEBUG;
        cpu_loop_exit(cpu);
    }

    return last_tb;
}


static void cpu_exec_enter(CPUState *cpu)
{
    CPUClass *cc = CPU_GET_CLASS(cpu);

    if (cc->tcg_ops->cpu_exec_enter) {
        cc->tcg_ops->cpu_exec_enter(cpu);
    }
}

static void cpu_exec_exit(CPUState *cpu)
{
    CPUClass *cc = CPU_GET_CLASS(cpu);

    if (cc->tcg_ops->cpu_exec_exit) {
        cc->tcg_ops->cpu_exec_exit(cpu);
    }
}

static void cpu_exec_longjmp_cleanup(CPUState *cpu)
{
    /* Non-buggy compilers preserve this; assert the correct value. */
    g_assert(cpu == current_cpu);

#ifdef CONFIG_USER_ONLY
    clear_helper_retaddr();
    if (have_mmap_lock()) {
        mmap_unlock();
    }
#else
    /*
     * For softmmu, a tlb_fill fault during translation will land here,
     * and we need to release any page locks held.  In system mode we
     * have one tcg_ctx per thread, so we know it was this cpu doing
     * the translation.
     *
     * Alternative 1: Install a cleanup to be called via an exception
     * handling safe longjmp.  It seems plausible that all our hosts
     * support such a thing.  We'd have to properly register unwind info
     * for the JIT for EH, rather that just for GDB.
     *
     * Alternative 2: Set and restore cpu->jmp_env in tb_gen_code to
     * capture the cpu_loop_exit longjmp, perform the cleanup, and
     * jump again to arrive here.
     */
    if (tcg_ctx->gen_tb) {
        tb_unlock_pages(tcg_ctx->gen_tb);
        tcg_ctx->gen_tb = NULL;
    }
#endif
    if (bql_locked()) {
        bql_unlock();
    }
    assert_no_pages_locked();
}

void cpu_exec_step_atomic(CPUState *cpu)
{
    CPUArchState *env = cpu_env(cpu);
    TranslationBlock *tb;
    vaddr pc;
    uint64_t cs_base;
    uint32_t flags, cflags;
    int tb_exit;

    if (sigsetjmp(cpu->jmp_env, 0) == 0) {
        start_exclusive();
        g_assert(cpu == current_cpu);
        g_assert(!cpu->running);
        cpu->running = true;

        cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);

        cflags = curr_cflags(cpu);
        /* Execute in a serial context. */
        cflags &= ~CF_PARALLEL;
        /* After 1 insn, return and release the exclusive lock. */
        cflags |= CF_NO_GOTO_TB | CF_NO_GOTO_PTR | 1;
        /*
         * No need to check_for_breakpoints here.
         * We only arrive in cpu_exec_step_atomic after beginning execution
         * of an insn that includes an atomic operation we can't handle.
         * Any breakpoint for this insn will have been recognized earlier.
         */

        tb = tb_lookup(cpu, pc, cs_base, flags, cflags);
        if (tb == NULL) {
            mmap_lock();
            tb = tb_gen_code(cpu, pc, cs_base, flags, cflags);
            mmap_unlock();
        }

        cpu_exec_enter(cpu);
        /* execute the generated code */
        trace_exec_tb(tb, pc);
        cpu_tb_exec(cpu, tb, &tb_exit);
        cpu_exec_exit(cpu);
    } else {
        cpu_exec_longjmp_cleanup(cpu);
    }

    /*
     * As we start the exclusive region before codegen we must still
     * be in the region if we longjump out of either the codegen or
     * the execution.
     */
    g_assert(cpu_in_exclusive_context(cpu));
    cpu->running = false;
    end_exclusive();
}

void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
{
    /*
     * Get the rx view of the structure, from which we find the
     * executable code address, and tb_target_set_jmp_target can
     * produce a pc-relative displacement to jmp_target_addr[n].
     */
    const TranslationBlock *c_tb = tcg_splitwx_to_rx(tb);
    uintptr_t offset = tb->jmp_insn_offset[n];
    uintptr_t jmp_rx = (uintptr_t)tb->tc.ptr + offset;
    uintptr_t jmp_rw = jmp_rx - tcg_splitwx_diff;

    tb->jmp_target_addr[n] = addr;
    tb_target_set_jmp_target(c_tb, n, jmp_rx, jmp_rw);
}

static inline void tb_add_jump(TranslationBlock *tb, int n,
                               TranslationBlock *tb_next)
{
    uintptr_t old;

    qemu_thread_jit_write();
    assert(n < ARRAY_SIZE(tb->jmp_list_next));
    qemu_spin_lock(&tb_next->jmp_lock);

    /* make sure the destination TB is valid */
    if (tb_next->cflags & CF_INVALID) {
        goto out_unlock_next;
    }
    /* Atomically claim the jump destination slot only if it was NULL */
    old = qatomic_cmpxchg(&tb->jmp_dest[n], (uintptr_t)NULL,
                          (uintptr_t)tb_next);
    if (old) {
        goto out_unlock_next;
    }

    /* patch the native jump address */
    tb_set_jmp_target(tb, n, (uintptr_t)tb_next->tc.ptr);

    /* add in TB jmp list */
    tb->jmp_list_next[n] = tb_next->jmp_list_head;
    tb_next->jmp_list_head = (uintptr_t)tb | n;

    qemu_spin_unlock(&tb_next->jmp_lock);

    qemu_log_mask(CPU_LOG_EXEC, "Linking TBs %p index %d -> %p\n",
                  tb->tc.ptr, n, tb_next->tc.ptr);
    return;

 out_unlock_next:
    qemu_spin_unlock(&tb_next->jmp_lock);
    return;
}

static inline bool cpu_handle_halt(CPUState *cpu)
{
#ifndef CONFIG_USER_ONLY
    if (cpu->halted) {
#if defined(TARGET_I386)
        if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
            X86CPU *x86_cpu = X86_CPU(cpu);
            bql_lock();
            apic_poll_irq(x86_cpu->apic_state);
            cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
            bql_unlock();
        }
#endif /* TARGET_I386 */
        if (!cpu_has_work(cpu)) {
            return true;
        }

        cpu->halted = 0;
    }
#endif /* !CONFIG_USER_ONLY */

    return false;
}

static inline void cpu_handle_debug_exception(CPUState *cpu)
{
    CPUClass *cc = CPU_GET_CLASS(cpu);
    CPUWatchpoint *wp;

    if (!cpu->watchpoint_hit) {
        QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
            wp->flags &= ~BP_WATCHPOINT_HIT;
        }
    }

    if (cc->tcg_ops->debug_excp_handler) {
        cc->tcg_ops->debug_excp_handler(cpu);
    }
}

static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
{
    if (cpu->exception_index < 0) {
#ifndef CONFIG_USER_ONLY
        if (replay_has_exception()
            && cpu->neg.icount_decr.u16.low + cpu->icount_extra == 0) {
            /* Execute just one insn to trigger exception pending in the log */
            cpu->cflags_next_tb = (curr_cflags(cpu) & ~CF_USE_ICOUNT)
                | CF_NOIRQ | 1;
        }
#endif
        return false;
    }
    if (cpu->exception_index >= EXCP_INTERRUPT) {
        /* exit request from the cpu execution loop */
        *ret = cpu->exception_index;
        if (*ret == EXCP_DEBUG) {
            cpu_handle_debug_exception(cpu);
        }
        cpu->exception_index = -1;
        return true;
    } else {
#if defined(CONFIG_USER_ONLY)
        /* if user mode only, we simulate a fake exception
           which will be handled outside the cpu execution
           loop */
#if defined(TARGET_I386)
        CPUClass *cc = CPU_GET_CLASS(cpu);
        cc->tcg_ops->fake_user_interrupt(cpu);
#endif /* TARGET_I386 */
        *ret = cpu->exception_index;
        cpu->exception_index = -1;
        return true;
#else
        if (replay_exception()) {
            CPUClass *cc = CPU_GET_CLASS(cpu);
            bql_lock();
            cc->tcg_ops->do_interrupt(cpu);
            bql_unlock();
            cpu->exception_index = -1;

            if (unlikely(cpu->singlestep_enabled)) {
                /*
                 * After processing the exception, ensure an EXCP_DEBUG is
                 * raised when single-stepping so that GDB doesn't miss the
                 * next instruction.
                 */
                *ret = EXCP_DEBUG;
                cpu_handle_debug_exception(cpu);
                return true;
            }
        } else if (!replay_has_interrupt()) {
            /* give a chance to iothread in replay mode */
            *ret = EXCP_INTERRUPT;
            return true;
        }
#endif
    }

    return false;
}

#ifndef CONFIG_USER_ONLY
/*
 * CPU_INTERRUPT_POLL is a virtual event which gets converted into a
 * "real" interrupt event later. It does not need to be recorded for
 * replay purposes.
 */
static inline bool need_replay_interrupt(int interrupt_request)
{
#if defined(TARGET_I386)
    return !(interrupt_request & CPU_INTERRUPT_POLL);
#else
    return true;
#endif
}
#endif /* !CONFIG_USER_ONLY */

static inline bool cpu_handle_interrupt(CPUState *cpu,
                                        TranslationBlock **last_tb)
{
    /*
     * If we have requested custom cflags with CF_NOIRQ we should
     * skip checking here. Any pending interrupts will get picked up
     * by the next TB we execute under normal cflags.
     */
    if (cpu->cflags_next_tb != -1 && cpu->cflags_next_tb & CF_NOIRQ) {
        return false;
    }

    /* Clear the interrupt flag now since we're processing
     * cpu->interrupt_request and cpu->exit_request.
     * Ensure zeroing happens before reading cpu->exit_request or
     * cpu->interrupt_request (see also smp_wmb in cpu_exit())
     */
    qatomic_set_mb(&cpu->neg.icount_decr.u16.high, 0);

    if (unlikely(qatomic_read(&cpu->interrupt_request))) {
        int interrupt_request;
        bql_lock();
        interrupt_request = cpu->interrupt_request;
        if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
            /* Mask out external interrupts for this step. */
            interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
        }
        if (interrupt_request & CPU_INTERRUPT_DEBUG) {
            cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
            cpu->exception_index = EXCP_DEBUG;
            bql_unlock();
            return true;
        }
#if !defined(CONFIG_USER_ONLY)
        if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
            /* Do nothing */
        } else if (interrupt_request & CPU_INTERRUPT_HALT) {
            replay_interrupt();
            cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
            cpu->halted = 1;
            cpu->exception_index = EXCP_HLT;
            bql_unlock();
            return true;
        }
#if defined(TARGET_I386)
        else if (interrupt_request & CPU_INTERRUPT_INIT) {
            X86CPU *x86_cpu = X86_CPU(cpu);
            CPUArchState *env = &x86_cpu->env;
            replay_interrupt();
            cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0, 0);
            do_cpu_init(x86_cpu);
            cpu->exception_index = EXCP_HALTED;
            bql_unlock();
            return true;
        }
#else
        else if (interrupt_request & CPU_INTERRUPT_RESET) {
            replay_interrupt();
            cpu_reset(cpu);
            bql_unlock();
            return true;
        }
#endif /* !TARGET_I386 */
        /* The target hook has 3 exit conditions:
           False when the interrupt isn't processed,
           True when it is, and we should restart on a new TB,
           and via longjmp via cpu_loop_exit.  */
        else {
            CPUClass *cc = CPU_GET_CLASS(cpu);

            if (cc->tcg_ops->cpu_exec_interrupt &&
                cc->tcg_ops->cpu_exec_interrupt(cpu, interrupt_request)) {
                if (need_replay_interrupt(interrupt_request)) {
                    replay_interrupt();
                }
                /*
                 * After processing the interrupt, ensure an EXCP_DEBUG is
                 * raised when single-stepping so that GDB doesn't miss the
                 * next instruction.
                 */
                if (unlikely(cpu->singlestep_enabled)) {
                    cpu->exception_index = EXCP_DEBUG;
                    bql_unlock();
                    return true;
                }
                cpu->exception_index = -1;
                *last_tb = NULL;
            }
            /* The target hook may have updated the 'cpu->interrupt_request';
             * reload the 'interrupt_request' value */
            interrupt_request = cpu->interrupt_request;
        }
#endif /* !CONFIG_USER_ONLY */
        if (interrupt_request & CPU_INTERRUPT_EXITTB) {
            cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
            /* ensure that no TB jump will be modified as
               the program flow was changed */
            *last_tb = NULL;
        }

        /* If we exit via cpu_loop_exit/longjmp it is reset in cpu_exec */
        bql_unlock();
    }

    /* Finally, check if we need to exit to the main loop.  */
    if (unlikely(qatomic_read(&cpu->exit_request))
        || (icount_enabled()
            && (cpu->cflags_next_tb == -1 || cpu->cflags_next_tb & CF_USE_ICOUNT)
            && cpu->neg.icount_decr.u16.low + cpu->icount_extra == 0)) {
        qatomic_set(&cpu->exit_request, 0);
        if (cpu->exception_index == -1) {
            cpu->exception_index = EXCP_INTERRUPT;
        }
        return true;
    }

    return false;
}

static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
                                    vaddr pc, TranslationBlock **last_tb,
                                    int *tb_exit)
{
    int32_t insns_left;

    trace_exec_tb(tb, pc);
    tb = cpu_tb_exec(cpu, tb, tb_exit);
    if (*tb_exit != TB_EXIT_REQUESTED) {
        *last_tb = tb;
        return;
    }

    *last_tb = NULL;
    insns_left = qatomic_read(&cpu->neg.icount_decr.u32);
    if (insns_left < 0) {
        /* Something asked us to stop executing chained TBs; just
         * continue round the main loop. Whatever requested the exit
         * will also have set something else (eg exit_request or
         * interrupt_request) which will be handled by
         * cpu_handle_interrupt.  cpu_handle_interrupt will also
         * clear cpu->icount_decr.u16.high.
         */
        return;
    }

    /* Instruction counter expired.  */
    assert(icount_enabled());
#ifndef CONFIG_USER_ONLY
    /* Ensure global icount has gone forward */
    icount_update(cpu);
    /* Refill decrementer and continue execution.  */
    insns_left = MIN(0xffff, cpu->icount_budget);
    cpu->neg.icount_decr.u16.low = insns_left;
    cpu->icount_extra = cpu->icount_budget - insns_left;

    /*
     * If the next tb has more instructions than we have left to
     * execute we need to ensure we find/generate a TB with exactly
     * insns_left instructions in it.
     */
    if (insns_left > 0 && insns_left < tb->icount)  {
        assert(insns_left <= CF_COUNT_MASK);
        assert(cpu->icount_extra == 0);
        cpu->cflags_next_tb = (tb->cflags & ~CF_COUNT_MASK) | insns_left;
    }
#endif
}

/* main execution loop */

static int __attribute__((noinline))
cpu_exec_loop(CPUState *cpu, SyncClocks *sc)
{
    int ret;

    /* if an exception is pending, we execute it here */
    while (!cpu_handle_exception(cpu, &ret)) {
        TranslationBlock *last_tb = NULL;
        int tb_exit = 0;

        while (!cpu_handle_interrupt(cpu, &last_tb)) {
            TranslationBlock *tb;
            vaddr pc;
            uint64_t cs_base;
            uint32_t flags, cflags;

            cpu_get_tb_cpu_state(cpu_env(cpu), &pc, &cs_base, &flags);

            /*
             * When requested, use an exact setting for cflags for the next
             * execution.  This is used for icount, precise smc, and stop-
             * after-access watchpoints.  Since this request should never
             * have CF_INVALID set, -1 is a convenient invalid value that
             * does not require tcg headers for cpu_common_reset.
             */
            cflags = cpu->cflags_next_tb;
            if (cflags == -1) {
                cflags = curr_cflags(cpu);
            } else {
                cpu->cflags_next_tb = -1;
            }

            if (check_for_breakpoints(cpu, pc, &cflags)) {
                break;
            }

            tb = tb_lookup(cpu, pc, cs_base, flags, cflags);
            if (tb == NULL) {
                CPUJumpCache *jc;
                uint32_t h;

                mmap_lock();
                tb = tb_gen_code(cpu, pc, cs_base, flags, cflags);
                mmap_unlock();

                /*
                 * We add the TB in the virtual pc hash table
                 * for the fast lookup
                 */
                h = tb_jmp_cache_hash_func(pc);
                jc = cpu->tb_jmp_cache;
                jc->array[h].pc = pc;
                qatomic_set(&jc->array[h].tb, tb);
            }

#ifndef CONFIG_USER_ONLY
            /*
             * We don't take care of direct jumps when address mapping
             * changes in system emulation.  So it's not safe to make a
             * direct jump to a TB spanning two pages because the mapping
             * for the second page can change.
             */
            if (tb_page_addr1(tb) != -1) {
                last_tb = NULL;
            }
#endif
            /* See if we can patch the calling TB. */
            if (last_tb) {
                tb_add_jump(last_tb, tb_exit, tb);
            }

            cpu_loop_exec_tb(cpu, tb, pc, &last_tb, &tb_exit);

            /* Try to align the host and virtual clocks
               if the guest is in advance */
            align_clocks(sc, cpu);
        }
    }
    return ret;
}

static int cpu_exec_setjmp(CPUState *cpu, SyncClocks *sc)
{
    /* Prepare setjmp context for exception handling. */
    if (unlikely(sigsetjmp(cpu->jmp_env, 0) != 0)) {
        cpu_exec_longjmp_cleanup(cpu);
    }

    return cpu_exec_loop(cpu, sc);
}

int cpu_exec(CPUState *cpu)
{
    int ret;
    SyncClocks sc = { 0 };

    /* replay_interrupt may need current_cpu */
    current_cpu = cpu;

    if (cpu_handle_halt(cpu)) {
        return EXCP_HALTED;
    }

    RCU_READ_LOCK_GUARD();
    cpu_exec_enter(cpu);

    /*
     * Calculate difference between guest clock and host clock.
     * This delay includes the delay of the last cycle, so
     * what we have to do is sleep until it is 0. As for the
     * advance/delay we gain here, we try to fix it next time.
     */
    init_delay_params(&sc, cpu);

    ret = cpu_exec_setjmp(cpu, &sc);

    cpu_exec_exit(cpu);
    return ret;
}

bool tcg_exec_realizefn(CPUState *cpu, Error **errp)
{
    static bool tcg_target_initialized;
    CPUClass *cc = CPU_GET_CLASS(cpu);

    if (!tcg_target_initialized) {
        cc->tcg_ops->initialize();
        tcg_target_initialized = true;
    }

    cpu->tb_jmp_cache = g_new0(CPUJumpCache, 1);
    tlb_init(cpu);
#ifndef CONFIG_USER_ONLY
    tcg_iommu_init_notifier_list(cpu);
#endif /* !CONFIG_USER_ONLY */
    /* qemu_plugin_vcpu_init_hook delayed until cpu_index assigned. */

    return true;
}

/* undo the initializations in reverse order */
void tcg_exec_unrealizefn(CPUState *cpu)
{
#ifndef CONFIG_USER_ONLY
    tcg_iommu_free_notifier_list(cpu);
#endif /* !CONFIG_USER_ONLY */

    tlb_destroy(cpu);
    g_free_rcu(cpu->tb_jmp_cache, rcu);
}