accel/tcg/cputlb.c

/*
 *  Common CPU TLB handling
 *
 *  Copyright (c) 2003 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/main-loop.h"
#include "cpu.h"
#include "exec/exec-all.h"
#include "exec/memory.h"
#include "exec/address-spaces.h"
#include "exec/cpu_ldst.h"
#include "exec/cputlb.h"
#include "exec/memory-internal.h"
#include "exec/ram_addr.h"
#include "tcg/tcg.h"
#include "qemu/error-report.h"
#include "exec/log.h"
#include "exec/helper-proto.h"
#include "qemu/atomic.h"
#include "qemu/atomic128.h"

/* DEBUG defines, enable DEBUG_TLB_LOG to log to the CPU_LOG_MMU target */
/* #define DEBUG_TLB */
/* #define DEBUG_TLB_LOG */

#ifdef DEBUG_TLB
# define DEBUG_TLB_GATE 1
# ifdef DEBUG_TLB_LOG
#  define DEBUG_TLB_LOG_GATE 1
# else
#  define DEBUG_TLB_LOG_GATE 0
# endif
#else
# define DEBUG_TLB_GATE 0
# define DEBUG_TLB_LOG_GATE 0
#endif

#define tlb_debug(fmt, ...) do { \
    if (DEBUG_TLB_LOG_GATE) { \
        qemu_log_mask(CPU_LOG_MMU, "%s: " fmt, __func__, \
                      ## __VA_ARGS__); \
    } else if (DEBUG_TLB_GATE) { \
        fprintf(stderr, "%s: " fmt, __func__, ## __VA_ARGS__); \
    } \
} while (0)

#define assert_cpu_is_self(cpu) do {                              \
        if (DEBUG_TLB_GATE) {                                     \
            g_assert(!(cpu)->created || qemu_cpu_is_self(cpu));   \
        }                                                         \
    } while (0)

/* run_on_cpu_data.target_ptr should always be big enough for a
 * target_ulong even on 32 bit builds */
QEMU_BUILD_BUG_ON(sizeof(target_ulong) > sizeof(run_on_cpu_data));

/* We currently can't handle more than 16 bits in the MMUIDX bitmask.
 */
QEMU_BUILD_BUG_ON(NB_MMU_MODES > 16);
#define ALL_MMUIDX_BITS ((1 << NB_MMU_MODES) - 1)

static inline size_t sizeof_tlb(CPUArchState *env, uintptr_t mmu_idx)
{
    return env->tlb_mask[mmu_idx] + (1 << CPU_TLB_ENTRY_BITS);
}

static void tlb_window_reset(CPUTLBWindow *window, int64_t ns,
                             size_t max_entries)
{
    window->begin_ns = ns;
    window->max_entries = max_entries;
}

static void tlb_dyn_init(CPUArchState *env)
{
    int i;

    for (i = 0; i < NB_MMU_MODES; i++) {
        CPUTLBDesc *desc = &env->tlb_d[i];
        size_t n_entries = 1 << CPU_TLB_DYN_DEFAULT_BITS;

        tlb_window_reset(&desc->window, get_clock_realtime(), 0);
        desc->n_used_entries = 0;
        env->tlb_mask[i] = (n_entries - 1) << CPU_TLB_ENTRY_BITS;
        env->tlb_table[i] = g_new(CPUTLBEntry, n_entries);
        env->iotlb[i] = g_new(CPUIOTLBEntry, n_entries);
    }
}

/**
 * tlb_mmu_resize_locked() - perform TLB resize bookkeeping; resize if necessary
 * @env: CPU that owns the TLB
 * @mmu_idx: MMU index of the TLB
 *
 * Called with tlb_lock_held.
 *
 * We have two main constraints when resizing a TLB: (1) we only resize it
 * on a TLB flush (otherwise we'd have to take a perf hit by either rehashing
 * the array or unnecessarily flushing it), which means we do not control how
 * frequently the resizing can occur; (2) we don't have access to the guest's
 * future scheduling decisions, and therefore have to decide the magnitude of
 * the resize based on past observations.
 *
 * In general, a memory-hungry process can benefit greatly from an appropriately
 * sized TLB, since a guest TLB miss is very expensive. This doesn't mean that
 * we just have to make the TLB as large as possible; while an oversized TLB
 * results in minimal TLB miss rates, it also takes longer to be flushed
 * (flushes can be _very_ frequent), and the reduced locality can also hurt
 * performance.
 *
 * To achieve near-optimal performance for all kinds of workloads, we:
 *
 * 1. Aggressively increase the size of the TLB when the use rate of the
 * TLB being flushed is high, since it is likely that in the near future this
 * memory-hungry process will execute again, and its memory hungriness will
 * probably be similar.
 *
 * 2. Slowly reduce the size of the TLB as the use rate declines over a
 * reasonably large time window. The rationale is that if in such a time window
 * we have not observed a high TLB use rate, it is likely that we won't observe
 * it in the near future. In that case, once a time window expires we downsize
 * the TLB to match the maximum use rate observed in the window.
 *
 * 3. Try to keep the maximum use rate in a time window in the 30-70% range,
 * since in that range performance is likely near-optimal. Recall that the TLB
 * is direct mapped, so we want the use rate to be low (or at least not too
 * high), since otherwise we are likely to have a significant amount of
 * conflict misses.
 */
static void tlb_mmu_resize_locked(CPUArchState *env, int mmu_idx)
{
    CPUTLBDesc *desc = &env->tlb_d[mmu_idx];
    size_t old_size = tlb_n_entries(env, mmu_idx);
    size_t rate;
    size_t new_size = old_size;
    int64_t now = get_clock_realtime();
    int64_t window_len_ms = 100;
    int64_t window_len_ns = window_len_ms * 1000 * 1000;
    bool window_expired = now > desc->window.begin_ns + window_len_ns;

    if (desc->n_used_entries > desc->window.max_entries) {
        desc->window.max_entries = desc->n_used_entries;
    }
    rate = desc->window.max_entries * 100 / old_size;

    if (rate > 70) {
        new_size = MIN(old_size << 1, 1 << CPU_TLB_DYN_MAX_BITS);
    } else if (rate < 30 && window_expired) {
        size_t ceil = pow2ceil(desc->window.max_entries);
        size_t expected_rate = desc->window.max_entries * 100 / ceil;

        /*
         * Avoid undersizing when the max number of entries seen is just below
         * a pow2. For instance, if max_entries == 1025, the expected use rate
         * would be 1025/2048==50%. However, if max_entries == 1023, we'd get
         * 1023/1024==99.9% use rate, so we'd likely end up doubling the size
         * later. Thus, make sure that the expected use rate remains below 70%.
         * (and since we double the size, that means the lowest rate we'd
         * expect to get is 35%, which is still in the 30-70% range where
         * we consider that the size is appropriate.)
         */
        if (expected_rate > 70) {
            ceil *= 2;
        }
        new_size = MAX(ceil, 1 << CPU_TLB_DYN_MIN_BITS);
    }

    if (new_size == old_size) {
        if (window_expired) {
            tlb_window_reset(&desc->window, now, desc->n_used_entries);
        }
        return;
    }

    g_free(env->tlb_table[mmu_idx]);
    g_free(env->iotlb[mmu_idx]);

    tlb_window_reset(&desc->window, now, 0);
    /* desc->n_used_entries is cleared by the caller */
    env->tlb_mask[mmu_idx] = (new_size - 1) << CPU_TLB_ENTRY_BITS;
    env->tlb_table[mmu_idx] = g_try_new(CPUTLBEntry, new_size);
    env->iotlb[mmu_idx] = g_try_new(CPUIOTLBEntry, new_size);
    /*
     * If the allocations fail, try smaller sizes. We just freed some
     * memory, so going back to half of new_size has a good chance of working.
     * Increased memory pressure elsewhere in the system might cause the
     * allocations to fail though, so we progressively reduce the allocation
     * size, aborting if we cannot even allocate the smallest TLB we support.
     */
    while (env->tlb_table[mmu_idx] == NULL || env->iotlb[mmu_idx] == NULL) {
        if (new_size == (1 << CPU_TLB_DYN_MIN_BITS)) {
            error_report("%s: %s", __func__, strerror(errno));
            abort();
        }
        new_size = MAX(new_size >> 1, 1 << CPU_TLB_DYN_MIN_BITS);
        env->tlb_mask[mmu_idx] = (new_size - 1) << CPU_TLB_ENTRY_BITS;

        g_free(env->tlb_table[mmu_idx]);
        g_free(env->iotlb[mmu_idx]);
        env->tlb_table[mmu_idx] = g_try_new(CPUTLBEntry, new_size);
        env->iotlb[mmu_idx] = g_try_new(CPUIOTLBEntry, new_size);
    }
}

static inline void tlb_table_flush_by_mmuidx(CPUArchState *env, int mmu_idx)
{
    tlb_mmu_resize_locked(env, mmu_idx);
    memset(env->tlb_table[mmu_idx], -1, sizeof_tlb(env, mmu_idx));
    env->tlb_d[mmu_idx].n_used_entries = 0;
}

static inline void tlb_n_used_entries_inc(CPUArchState *env, uintptr_t mmu_idx)
{
    env->tlb_d[mmu_idx].n_used_entries++;
}

static inline void tlb_n_used_entries_dec(CPUArchState *env, uintptr_t mmu_idx)
{
    env->tlb_d[mmu_idx].n_used_entries--;
}

void tlb_init(CPUState *cpu)
{
    CPUArchState *env = cpu->env_ptr;

    qemu_spin_init(&env->tlb_c.lock);

    /* Ensure that cpu_reset performs a full flush.  */
    env->tlb_c.dirty = ALL_MMUIDX_BITS;

    tlb_dyn_init(env);
}

/* flush_all_helper: run fn across all cpus
 *
 * If the wait flag is set then the src cpu's helper will be queued as
 * "safe" work and the loop exited creating a synchronisation point
 * where all queued work will be finished before execution starts
 * again.
 */
static void flush_all_helper(CPUState *src, run_on_cpu_func fn,
                             run_on_cpu_data d)
{
    CPUState *cpu;

    CPU_FOREACH(cpu) {
        if (cpu != src) {
            async_run_on_cpu(cpu, fn, d);
        }
    }
}

void tlb_flush_counts(size_t *pfull, size_t *ppart, size_t *pelide)
{
    CPUState *cpu;
    size_t full = 0, part = 0, elide = 0;

    CPU_FOREACH(cpu) {
        CPUArchState *env = cpu->env_ptr;

        full += atomic_read(&env->tlb_c.full_flush_count);
        part += atomic_read(&env->tlb_c.part_flush_count);
        elide += atomic_read(&env->tlb_c.elide_flush_count);
    }
    *pfull = full;
    *ppart = part;
    *pelide = elide;
}

static void tlb_flush_one_mmuidx_locked(CPUArchState *env, int mmu_idx)
{
    tlb_table_flush_by_mmuidx(env, mmu_idx);
    memset(env->tlb_v_table[mmu_idx], -1, sizeof(env->tlb_v_table[0]));
    env->tlb_d[mmu_idx].large_page_addr = -1;
    env->tlb_d[mmu_idx].large_page_mask = -1;
    env->tlb_d[mmu_idx].vindex = 0;
}

static void tlb_flush_by_mmuidx_async_work(CPUState *cpu, run_on_cpu_data data)
{
    CPUArchState *env = cpu->env_ptr;
    uint16_t asked = data.host_int;
    uint16_t all_dirty, work, to_clean;

    assert_cpu_is_self(cpu);

    tlb_debug("mmu_idx:0x%04" PRIx16 "\n", asked);

    qemu_spin_lock(&env->tlb_c.lock);

    all_dirty = env->tlb_c.dirty;
    to_clean = asked & all_dirty;
    all_dirty &= ~to_clean;
    env->tlb_c.dirty = all_dirty;

    for (work = to_clean; work != 0; work &= work - 1) {
        int mmu_idx = ctz32(work);
        tlb_flush_one_mmuidx_locked(env, mmu_idx);
    }

    qemu_spin_unlock(&env->tlb_c.lock);

    cpu_tb_jmp_cache_clear(cpu);

    if (to_clean == ALL_MMUIDX_BITS) {
        atomic_set(&env->tlb_c.full_flush_count,
                   env->tlb_c.full_flush_count + 1);
    } else {
        atomic_set(&env->tlb_c.part_flush_count,
                   env->tlb_c.part_flush_count + ctpop16(to_clean));
        if (to_clean != asked) {
            atomic_set(&env->tlb_c.elide_flush_count,
                       env->tlb_c.elide_flush_count +
                       ctpop16(asked & ~to_clean));
        }
    }
}

void tlb_flush_by_mmuidx(CPUState *cpu, uint16_t idxmap)
{
    tlb_debug("mmu_idx: 0x%" PRIx16 "\n", idxmap);

    if (cpu->created && !qemu_cpu_is_self(cpu)) {
        async_run_on_cpu(cpu, tlb_flush_by_mmuidx_async_work,
                         RUN_ON_CPU_HOST_INT(idxmap));
    } else {
        tlb_flush_by_mmuidx_async_work(cpu, RUN_ON_CPU_HOST_INT(idxmap));
    }
}

void tlb_flush(CPUState *cpu)
{
    tlb_flush_by_mmuidx(cpu, ALL_MMUIDX_BITS);
}

void tlb_flush_by_mmuidx_all_cpus(CPUState *src_cpu, uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_by_mmuidx_async_work;

    tlb_debug("mmu_idx: 0x%"PRIx16"\n", idxmap);

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
    fn(src_cpu, RUN_ON_CPU_HOST_INT(idxmap));
}

void tlb_flush_all_cpus(CPUState *src_cpu)
{
    tlb_flush_by_mmuidx_all_cpus(src_cpu, ALL_MMUIDX_BITS);
}

void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *src_cpu, uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_by_mmuidx_async_work;

    tlb_debug("mmu_idx: 0x%"PRIx16"\n", idxmap);

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
    async_safe_run_on_cpu(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
}

void tlb_flush_all_cpus_synced(CPUState *src_cpu)
{
    tlb_flush_by_mmuidx_all_cpus_synced(src_cpu, ALL_MMUIDX_BITS);
}

static inline bool tlb_hit_page_anyprot(CPUTLBEntry *tlb_entry,
                                        target_ulong page)
{
    return tlb_hit_page(tlb_entry->addr_read, page) ||
           tlb_hit_page(tlb_addr_write(tlb_entry), page) ||
           tlb_hit_page(tlb_entry->addr_code, page);
}

/**
 * tlb_entry_is_empty - return true if the entry is not in use
 * @te: pointer to CPUTLBEntry
 */
static inline bool tlb_entry_is_empty(const CPUTLBEntry *te)
{
    return te->addr_read == -1 && te->addr_write == -1 && te->addr_code == -1;
}

/* Called with tlb_c.lock held */
static inline bool tlb_flush_entry_locked(CPUTLBEntry *tlb_entry,
                                          target_ulong page)
{
    if (tlb_hit_page_anyprot(tlb_entry, page)) {
        memset(tlb_entry, -1, sizeof(*tlb_entry));
        return true;
    }
    return false;
}

/* Called with tlb_c.lock held */
static inline void tlb_flush_vtlb_page_locked(CPUArchState *env, int mmu_idx,
                                              target_ulong page)
{
    int k;

    assert_cpu_is_self(ENV_GET_CPU(env));
    for (k = 0; k < CPU_VTLB_SIZE; k++) {
        if (tlb_flush_entry_locked(&env->tlb_v_table[mmu_idx][k], page)) {
            tlb_n_used_entries_dec(env, mmu_idx);
        }
    }
}

static void tlb_flush_page_locked(CPUArchState *env, int midx,
                                  target_ulong page)
{
    target_ulong lp_addr = env->tlb_d[midx].large_page_addr;
    target_ulong lp_mask = env->tlb_d[midx].large_page_mask;

    /* Check if we need to flush due to large pages.  */
    if ((page & lp_mask) == lp_addr) {
        tlb_debug("forcing full flush midx %d ("
                  TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
                  midx, lp_addr, lp_mask);
        tlb_flush_one_mmuidx_locked(env, midx);
    } else {
        if (tlb_flush_entry_locked(tlb_entry(env, midx, page), page)) {
            tlb_n_used_entries_dec(env, midx);
        }
        tlb_flush_vtlb_page_locked(env, midx, page);
    }
}

/* As we are going to hijack the bottom bits of the page address for a
 * mmuidx bit mask we need to fail to build if we can't do that
 */
QEMU_BUILD_BUG_ON(NB_MMU_MODES > TARGET_PAGE_BITS_MIN);

static void tlb_flush_page_by_mmuidx_async_work(CPUState *cpu,
                                                run_on_cpu_data data)
{
    CPUArchState *env = cpu->env_ptr;
    target_ulong addr_and_mmuidx = (target_ulong) data.target_ptr;
    target_ulong addr = addr_and_mmuidx & TARGET_PAGE_MASK;
    unsigned long mmu_idx_bitmap = addr_and_mmuidx & ALL_MMUIDX_BITS;
    int mmu_idx;

    assert_cpu_is_self(cpu);

    tlb_debug("page addr:" TARGET_FMT_lx " mmu_map:0x%lx\n",
              addr, mmu_idx_bitmap);

    qemu_spin_lock(&env->tlb_c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        if (test_bit(mmu_idx, &mmu_idx_bitmap)) {
            tlb_flush_page_locked(env, mmu_idx, addr);
        }
    }
    qemu_spin_unlock(&env->tlb_c.lock);

    tb_flush_jmp_cache(cpu, addr);
}

void tlb_flush_page_by_mmuidx(CPUState *cpu, target_ulong addr, uint16_t idxmap)
{
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%" PRIx16 "\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    if (!qemu_cpu_is_self(cpu)) {
        async_run_on_cpu(cpu, tlb_flush_page_by_mmuidx_async_work,
                         RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    } else {
        tlb_flush_page_by_mmuidx_async_work(
            cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    }
}

void tlb_flush_page(CPUState *cpu, target_ulong addr)
{
    tlb_flush_page_by_mmuidx(cpu, addr, ALL_MMUIDX_BITS);
}

void tlb_flush_page_by_mmuidx_all_cpus(CPUState *src_cpu, target_ulong addr,
                                       uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    fn(src_cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
}

void tlb_flush_page_all_cpus(CPUState *src, target_ulong addr)
{
    tlb_flush_page_by_mmuidx_all_cpus(src, addr, ALL_MMUIDX_BITS);
}

void tlb_flush_page_by_mmuidx_all_cpus_synced(CPUState *src_cpu,
                                              target_ulong addr,
                                              uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    async_safe_run_on_cpu(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
}

void tlb_flush_page_all_cpus_synced(CPUState *src, target_ulong addr)
{
    tlb_flush_page_by_mmuidx_all_cpus_synced(src, addr, ALL_MMUIDX_BITS);
}

/* update the TLBs so that writes to code in the virtual page 'addr'
   can be detected */
void tlb_protect_code(ram_addr_t ram_addr)
{
    cpu_physical_memory_test_and_clear_dirty(ram_addr, TARGET_PAGE_SIZE,
                                             DIRTY_MEMORY_CODE);
}

/* update the TLB so that writes in physical page 'phys_addr' are no longer
   tested for self modifying code */
void tlb_unprotect_code(ram_addr_t ram_addr)
{
    cpu_physical_memory_set_dirty_flag(ram_addr, DIRTY_MEMORY_CODE);
}


/*
 * Dirty write flag handling
 *
 * When the TCG code writes to a location it looks up the address in
 * the TLB and uses that data to compute the final address. If any of
 * the lower bits of the address are set then the slow path is forced.
 * There are a number of reasons to do this but for normal RAM the
 * most usual is detecting writes to code regions which may invalidate
 * generated code.
 *
 * Other vCPUs might be reading their TLBs during guest execution, so we update
 * te->addr_write with atomic_set. We don't need to worry about this for
 * oversized guests as MTTCG is disabled for them.
 *
 * Called with tlb_c.lock held.
 */
static void tlb_reset_dirty_range_locked(CPUTLBEntry *tlb_entry,
                                         uintptr_t start, uintptr_t length)
{
    uintptr_t addr = tlb_entry->addr_write;

    if ((addr & (TLB_INVALID_MASK | TLB_MMIO | TLB_NOTDIRTY)) == 0) {
        addr &= TARGET_PAGE_MASK;
        addr += tlb_entry->addend;
        if ((addr - start) < length) {
#if TCG_OVERSIZED_GUEST
            tlb_entry->addr_write |= TLB_NOTDIRTY;
#else
            atomic_set(&tlb_entry->addr_write,
                       tlb_entry->addr_write | TLB_NOTDIRTY);
#endif
        }
    }
}

/*
 * Called with tlb_c.lock held.
 * Called only from the vCPU context, i.e. the TLB's owner thread.
 */
static inline void copy_tlb_helper_locked(CPUTLBEntry *d, const CPUTLBEntry *s)
{
    *d = *s;
}

/* This is a cross vCPU call (i.e. another vCPU resetting the flags of
 * the target vCPU).
 * We must take tlb_c.lock to avoid racing with another vCPU update. The only
 * thing actually updated is the target TLB entry ->addr_write flags.
 */
void tlb_reset_dirty(CPUState *cpu, ram_addr_t start1, ram_addr_t length)
{
    CPUArchState *env;

    int mmu_idx;

    env = cpu->env_ptr;
    qemu_spin_lock(&env->tlb_c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        unsigned int i;
        unsigned int n = tlb_n_entries(env, mmu_idx);

        for (i = 0; i < n; i++) {
            tlb_reset_dirty_range_locked(&env->tlb_table[mmu_idx][i], start1,
                                         length);
        }

        for (i = 0; i < CPU_VTLB_SIZE; i++) {
            tlb_reset_dirty_range_locked(&env->tlb_v_table[mmu_idx][i], start1,
                                         length);
        }
    }
    qemu_spin_unlock(&env->tlb_c.lock);
}

/* Called with tlb_c.lock held */
static inline void tlb_set_dirty1_locked(CPUTLBEntry *tlb_entry,
                                         target_ulong vaddr)
{
    if (tlb_entry->addr_write == (vaddr | TLB_NOTDIRTY)) {
        tlb_entry->addr_write = vaddr;
    }
}

/* update the TLB corresponding to virtual page vaddr
   so that it is no longer dirty */
void tlb_set_dirty(CPUState *cpu, target_ulong vaddr)
{
    CPUArchState *env = cpu->env_ptr;
    int mmu_idx;

    assert_cpu_is_self(cpu);

    vaddr &= TARGET_PAGE_MASK;
    qemu_spin_lock(&env->tlb_c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        tlb_set_dirty1_locked(tlb_entry(env, mmu_idx, vaddr), vaddr);
    }

    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        int k;
        for (k = 0; k < CPU_VTLB_SIZE; k++) {
            tlb_set_dirty1_locked(&env->tlb_v_table[mmu_idx][k], vaddr);
        }
    }
    qemu_spin_unlock(&env->tlb_c.lock);
}

/* Our TLB does not support large pages, so remember the area covered by
   large pages and trigger a full TLB flush if these are invalidated.  */
static void tlb_add_large_page(CPUArchState *env, int mmu_idx,
                               target_ulong vaddr, target_ulong size)
{
    target_ulong lp_addr = env->tlb_d[mmu_idx].large_page_addr;
    target_ulong lp_mask = ~(size - 1);

    if (lp_addr == (target_ulong)-1) {
        /* No previous large page.  */
        lp_addr = vaddr;
    } else {
        /* Extend the existing region to include the new page.
           This is a compromise between unnecessary flushes and
           the cost of maintaining a full variable size TLB.  */
        lp_mask &= env->tlb_d[mmu_idx].large_page_mask;
        while (((lp_addr ^ vaddr) & lp_mask) != 0) {
            lp_mask <<= 1;
        }
    }
    env->tlb_d[mmu_idx].large_page_addr = lp_addr & lp_mask;
    env->tlb_d[mmu_idx].large_page_mask = lp_mask;
}

/* Add a new TLB entry. At most one entry for a given virtual address
 * is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
 * supplied size is only used by tlb_flush_page.
 *
 * Called from TCG-generated code, which is under an RCU read-side
 * critical section.
 */
void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
                             hwaddr paddr, MemTxAttrs attrs, int prot,
                             int mmu_idx, target_ulong size)
{
    CPUArchState *env = cpu->env_ptr;
    MemoryRegionSection *section;
    unsigned int index;
    target_ulong address;
    target_ulong code_address;
    uintptr_t addend;
    CPUTLBEntry *te, tn;
    hwaddr iotlb, xlat, sz, paddr_page;
    target_ulong vaddr_page;
    int asidx = cpu_asidx_from_attrs(cpu, attrs);

    assert_cpu_is_self(cpu);

    if (size <= TARGET_PAGE_SIZE) {
        sz = TARGET_PAGE_SIZE;
    } else {
        tlb_add_large_page(env, mmu_idx, vaddr, size);
        sz = size;
    }
    vaddr_page = vaddr & TARGET_PAGE_MASK;
    paddr_page = paddr & TARGET_PAGE_MASK;

    section = address_space_translate_for_iotlb(cpu, asidx, paddr_page,
                                                &xlat, &sz, attrs, &prot);
    assert(sz >= TARGET_PAGE_SIZE);

    tlb_debug("vaddr=" TARGET_FMT_lx " paddr=0x" TARGET_FMT_plx
              " prot=%x idx=%d\n",
              vaddr, paddr, prot, mmu_idx);

    address = vaddr_page;
    if (size < TARGET_PAGE_SIZE) {
        /*
         * Slow-path the TLB entries; we will repeat the MMU check and TLB
         * fill on every access.
         */
        address |= TLB_RECHECK;
    }
    if (!memory_region_is_ram(section->mr) &&
        !memory_region_is_romd(section->mr)) {
        /* IO memory case */
        address |= TLB_MMIO;
        addend = 0;
    } else {
        /* TLB_MMIO for rom/romd handled below */
        addend = (uintptr_t)memory_region_get_ram_ptr(section->mr) + xlat;
    }

    code_address = address;
    iotlb = memory_region_section_get_iotlb(cpu, section, vaddr_page,
                                            paddr_page, xlat, prot, &address);

    index = tlb_index(env, mmu_idx, vaddr_page);
    te = tlb_entry(env, mmu_idx, vaddr_page);

    /*
     * Hold the TLB lock for the rest of the function. We could acquire/release
     * the lock several times in the function, but it is faster to amortize the
     * acquisition cost by acquiring it just once. Note that this leads to
     * a longer critical section, but this is not a concern since the TLB lock
     * is unlikely to be contended.
     */
    qemu_spin_lock(&env->tlb_c.lock);

    /* Note that the tlb is no longer clean.  */
    env->tlb_c.dirty |= 1 << mmu_idx;

    /* Make sure there's no cached translation for the new page.  */
    tlb_flush_vtlb_page_locked(env, mmu_idx, vaddr_page);

    /*
     * Only evict the old entry to the victim tlb if it's for a
     * different page; otherwise just overwrite the stale data.
     */
    if (!tlb_hit_page_anyprot(te, vaddr_page) && !tlb_entry_is_empty(te)) {
        unsigned vidx = env->tlb_d[mmu_idx].vindex++ % CPU_VTLB_SIZE;
        CPUTLBEntry *tv = &env->tlb_v_table[mmu_idx][vidx];

        /* Evict the old entry into the victim tlb.  */
        copy_tlb_helper_locked(tv, te);
        env->iotlb_v[mmu_idx][vidx] = env->iotlb[mmu_idx][index];
        tlb_n_used_entries_dec(env, mmu_idx);
    }

    /* refill the tlb */
    /*
     * At this point iotlb contains a physical section number in the lower
     * TARGET_PAGE_BITS, and either
     *  + the ram_addr_t of the page base of the target RAM (if NOTDIRTY or ROM)
     *  + the offset within section->mr of the page base (otherwise)
     * We subtract the vaddr_page (which is page aligned and thus won't
     * disturb the low bits) to give an offset which can be added to the
     * (non-page-aligned) vaddr of the eventual memory access to get
     * the MemoryRegion offset for the access. Note that the vaddr we
     * subtract here is that of the page base, and not the same as the
     * vaddr we add back in io_readx()/io_writex()/get_page_addr_code().
     */
    env->iotlb[mmu_idx][index].addr = iotlb - vaddr_page;
    env->iotlb[mmu_idx][index].attrs = attrs;

    /* Now calculate the new entry */
    tn.addend = addend - vaddr_page;
    if (prot & PAGE_READ) {
        tn.addr_read = address;
    } else {
        tn.addr_read = -1;
    }

    if (prot & PAGE_EXEC) {
        tn.addr_code = code_address;
    } else {
        tn.addr_code = -1;
    }

    tn.addr_write = -1;
    if (prot & PAGE_WRITE) {
        if ((memory_region_is_ram(section->mr) && section->readonly)
            || memory_region_is_romd(section->mr)) {
            /* Write access calls the I/O callback.  */
            tn.addr_write = address | TLB_MMIO;
        } else if (memory_region_is_ram(section->mr)
                   && cpu_physical_memory_is_clean(
                       memory_region_get_ram_addr(section->mr) + xlat)) {
            tn.addr_write = address | TLB_NOTDIRTY;
        } else {
            tn.addr_write = address;
        }
        if (prot & PAGE_WRITE_INV) {
            tn.addr_write |= TLB_INVALID_MASK;
        }
    }

    copy_tlb_helper_locked(te, &tn);
    tlb_n_used_entries_inc(env, mmu_idx);
    qemu_spin_unlock(&env->tlb_c.lock);
}

/* Add a new TLB entry, but without specifying the memory
 * transaction attributes to be used.
 */
void tlb_set_page(CPUState *cpu, target_ulong vaddr,
                  hwaddr paddr, int prot,
                  int mmu_idx, target_ulong size)
{
    tlb_set_page_with_attrs(cpu, vaddr, paddr, MEMTXATTRS_UNSPECIFIED,
                            prot, mmu_idx, size);
}

static inline ram_addr_t qemu_ram_addr_from_host_nofail(void *ptr)
{
    ram_addr_t ram_addr;

    ram_addr = qemu_ram_addr_from_host(ptr);
    if (ram_addr == RAM_ADDR_INVALID) {
        error_report("Bad ram pointer %p", ptr);
        abort();
    }
    return ram_addr;
}

static uint64_t io_readx(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
                         int mmu_idx,
                         target_ulong addr, uintptr_t retaddr,
                         bool recheck, MMUAccessType access_type, int size)
{
    CPUState *cpu = ENV_GET_CPU(env);
    hwaddr mr_offset;
    MemoryRegionSection *section;
    MemoryRegion *mr;
    uint64_t val;
    bool locked = false;
    MemTxResult r;

    if (recheck) {
        /*
         * This is a TLB_RECHECK access, where the MMU protection
         * covers a smaller range than a target page, and we must
         * repeat the MMU check here. This tlb_fill() call might
         * longjump out if this access should cause a guest exception.
         */
        CPUTLBEntry *entry;
        target_ulong tlb_addr;

        tlb_fill(cpu, addr, size, access_type, mmu_idx, retaddr);

        entry = tlb_entry(env, mmu_idx, addr);
        tlb_addr = (access_type == MMU_DATA_LOAD ?
                    entry->addr_read : entry->addr_code);
        if (!(tlb_addr & ~(TARGET_PAGE_MASK | TLB_RECHECK))) {
            /* RAM access */
            uintptr_t haddr = addr + entry->addend;

            return ldn_p((void *)haddr, size);
        }
        /* Fall through for handling IO accesses */
    }

    section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
    mr = section->mr;
    mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
    cpu->mem_io_pc = retaddr;
    if (mr != &io_mem_rom && mr != &io_mem_notdirty && !cpu->can_do_io) {
        cpu_io_recompile(cpu, retaddr);
    }

    cpu->mem_io_vaddr = addr;
    cpu->mem_io_access_type = access_type;

    if (mr->global_locking && !qemu_mutex_iothread_locked()) {
        qemu_mutex_lock_iothread();
        locked = true;
    }
    r = memory_region_dispatch_read(mr, mr_offset,
                                    &val, size, iotlbentry->attrs);
    if (r != MEMTX_OK) {
        hwaddr physaddr = mr_offset +
            section->offset_within_address_space -
            section->offset_within_region;

        cpu_transaction_failed(cpu, physaddr, addr, size, access_type,
                               mmu_idx, iotlbentry->attrs, r, retaddr);
    }
    if (locked) {
        qemu_mutex_unlock_iothread();
    }

    return val;
}

static void io_writex(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
                      int mmu_idx,
                      uint64_t val, target_ulong addr,
                      uintptr_t retaddr, bool recheck, int size)
{
    CPUState *cpu = ENV_GET_CPU(env);
    hwaddr mr_offset;
    MemoryRegionSection *section;
    MemoryRegion *mr;
    bool locked = false;
    MemTxResult r;

    if (recheck) {
        /*
         * This is a TLB_RECHECK access, where the MMU protection
         * covers a smaller range than a target page, and we must
         * repeat the MMU check here. This tlb_fill() call might
         * longjump out if this access should cause a guest exception.
         */
        CPUTLBEntry *entry;
        target_ulong tlb_addr;

        tlb_fill(cpu, addr, size, MMU_DATA_STORE, mmu_idx, retaddr);

        entry = tlb_entry(env, mmu_idx, addr);
        tlb_addr = tlb_addr_write(entry);
        if (!(tlb_addr & ~(TARGET_PAGE_MASK | TLB_RECHECK))) {
            /* RAM access */
            uintptr_t haddr = addr + entry->addend;

            stn_p((void *)haddr, size, val);
            return;
        }
        /* Fall through for handling IO accesses */
    }

    section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
    mr = section->mr;
    mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
    if (mr != &io_mem_rom && mr != &io_mem_notdirty && !cpu->can_do_io) {
        cpu_io_recompile(cpu, retaddr);
    }
    cpu->mem_io_vaddr = addr;
    cpu->mem_io_pc = retaddr;

    if (mr->global_locking && !qemu_mutex_iothread_locked()) {
        qemu_mutex_lock_iothread();
        locked = true;
    }
    r = memory_region_dispatch_write(mr, mr_offset,
                                     val, size, iotlbentry->attrs);
    if (r != MEMTX_OK) {
        hwaddr physaddr = mr_offset +
            section->offset_within_address_space -
            section->offset_within_region;

        cpu_transaction_failed(cpu, physaddr, addr, size, MMU_DATA_STORE,
                               mmu_idx, iotlbentry->attrs, r, retaddr);
    }
    if (locked) {
        qemu_mutex_unlock_iothread();
    }
}

/* Return true if ADDR is present in the victim tlb, and has been copied
   back to the main tlb.  */
static bool victim_tlb_hit(CPUArchState *env, size_t mmu_idx, size_t index,
                           size_t elt_ofs, target_ulong page)
{
    size_t vidx;

    assert_cpu_is_self(ENV_GET_CPU(env));
    for (vidx = 0; vidx < CPU_VTLB_SIZE; ++vidx) {
        CPUTLBEntry *vtlb = &env->tlb_v_table[mmu_idx][vidx];
        target_ulong cmp;

        /* elt_ofs might correspond to .addr_write, so use atomic_read */
#if TCG_OVERSIZED_GUEST
        cmp = *(target_ulong *)((uintptr_t)vtlb + elt_ofs);
#else
        cmp = atomic_read((target_ulong *)((uintptr_t)vtlb + elt_ofs));
#endif

        if (cmp == page) {
            /* Found entry in victim tlb, swap tlb and iotlb.  */
            CPUTLBEntry tmptlb, *tlb = &env->tlb_table[mmu_idx][index];

            qemu_spin_lock(&env->tlb_c.lock);
            copy_tlb_helper_locked(&tmptlb, tlb);
            copy_tlb_helper_locked(tlb, vtlb);
            copy_tlb_helper_locked(vtlb, &tmptlb);
            qemu_spin_unlock(&env->tlb_c.lock);

            CPUIOTLBEntry tmpio, *io = &env->iotlb[mmu_idx][index];
            CPUIOTLBEntry *vio = &env->iotlb_v[mmu_idx][vidx];
            tmpio = *io; *io = *vio; *vio = tmpio;
            return true;
        }
    }
    return false;
}

/* Macro to call the above, with local variables from the use context.  */
#define VICTIM_TLB_HIT(TY, ADDR) \
  victim_tlb_hit(env, mmu_idx, index, offsetof(CPUTLBEntry, TY), \
                 (ADDR) & TARGET_PAGE_MASK)

/* NOTE: this function can trigger an exception */
/* NOTE2: the returned address is not exactly the physical address: it
 * is actually a ram_addr_t (in system mode; the user mode emulation
 * version of this function returns a guest virtual address).
 */
tb_page_addr_t get_page_addr_code(CPUArchState *env, target_ulong addr)
{
    uintptr_t mmu_idx = cpu_mmu_index(env, true);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    void *p;

    if (unlikely(!tlb_hit(entry->addr_code, addr))) {
        if (!VICTIM_TLB_HIT(addr_code, addr)) {
            tlb_fill(ENV_GET_CPU(env), addr, 0, MMU_INST_FETCH, mmu_idx, 0);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        assert(tlb_hit(entry->addr_code, addr));
    }

    if (unlikely(entry->addr_code & (TLB_RECHECK | TLB_MMIO))) {
        /*
         * Return -1 if we can't translate and execute from an entire
         * page of RAM here, which will cause us to execute by loading
         * and translating one insn at a time, without caching:
         *  - TLB_RECHECK: means the MMU protection covers a smaller range
         *    than a target page, so we must redo the MMU check every insn
         *  - TLB_MMIO: region is not backed by RAM
         */
        return -1;
    }

    p = (void *)((uintptr_t)addr + entry->addend);
    return qemu_ram_addr_from_host_nofail(p);
}

/* Probe for whether the specified guest write access is permitted.
 * If it is not permitted then an exception will be taken in the same
 * way as if this were a real write access (and we will not return).
 * Otherwise the function will return, and there will be a valid
 * entry in the TLB for this access.
 */
void probe_write(CPUArchState *env, target_ulong addr, int size, int mmu_idx,
                 uintptr_t retaddr)
{
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);

    if (!tlb_hit(tlb_addr_write(entry), addr)) {
        /* TLB entry is for a different page */
        if (!VICTIM_TLB_HIT(addr_write, addr)) {
            tlb_fill(ENV_GET_CPU(env), addr, size, MMU_DATA_STORE,
                     mmu_idx, retaddr);
        }
    }
}

/* Probe for a read-modify-write atomic operation.  Do not allow unaligned
 * operations, or io operations to proceed.  Return the host address.  */
static void *atomic_mmu_lookup(CPUArchState *env, target_ulong addr,
                               TCGMemOpIdx oi, uintptr_t retaddr,
                               NotDirtyInfo *ndi)
{
    size_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *tlbe = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = tlb_addr_write(tlbe);
    TCGMemOp mop = get_memop(oi);
    int a_bits = get_alignment_bits(mop);
    int s_bits = mop & MO_SIZE;
    void *hostaddr;

    /* Adjust the given return address.  */
    retaddr -= GETPC_ADJ;

    /* Enforce guest required alignment.  */
    if (unlikely(a_bits > 0 && (addr & ((1 << a_bits) - 1)))) {
        /* ??? Maybe indicate atomic op to cpu_unaligned_access */
        cpu_unaligned_access(ENV_GET_CPU(env), addr, MMU_DATA_STORE,
                             mmu_idx, retaddr);
    }

    /* Enforce qemu required alignment.  */
    if (unlikely(addr & ((1 << s_bits) - 1))) {
        /* We get here if guest alignment was not requested,
           or was not enforced by cpu_unaligned_access above.
           We might widen the access and emulate, but for now
           mark an exception and exit the cpu loop.  */
        goto stop_the_world;
    }

    /* Check TLB entry and enforce page permissions.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!VICTIM_TLB_HIT(addr_write, addr)) {
            tlb_fill(ENV_GET_CPU(env), addr, 1 << s_bits, MMU_DATA_STORE,
                     mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            tlbe = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_addr_write(tlbe) & ~TLB_INVALID_MASK;
    }

    /* Notice an IO access or a needs-MMU-lookup access */
    if (unlikely(tlb_addr & (TLB_MMIO | TLB_RECHECK))) {
        /* There's really nothing that can be done to
           support this apart from stop-the-world.  */
        goto stop_the_world;
    }

    /* Let the guest notice RMW on a write-only page.  */
    if (unlikely(tlbe->addr_read != (tlb_addr & ~TLB_NOTDIRTY))) {
        tlb_fill(ENV_GET_CPU(env), addr, 1 << s_bits, MMU_DATA_LOAD,
                 mmu_idx, retaddr);
        /* Since we don't support reads and writes to different addresses,
           and we do have the proper page loaded for write, this shouldn't
           ever return.  But just in case, handle via stop-the-world.  */
        goto stop_the_world;
    }

    hostaddr = (void *)((uintptr_t)addr + tlbe->addend);

    ndi->active = false;
    if (unlikely(tlb_addr & TLB_NOTDIRTY)) {
        ndi->active = true;
        memory_notdirty_write_prepare(ndi, ENV_GET_CPU(env), addr,
                                      qemu_ram_addr_from_host_nofail(hostaddr),
                                      1 << s_bits);
    }

    return hostaddr;

 stop_the_world:
    cpu_loop_exit_atomic(ENV_GET_CPU(env), retaddr);
}

#ifdef TARGET_WORDS_BIGENDIAN
#define NEED_BE_BSWAP 0
#define NEED_LE_BSWAP 1
#else
#define NEED_BE_BSWAP 1
#define NEED_LE_BSWAP 0
#endif

/*
 * Byte Swap Helper
 *
 * This should all dead code away depending on the build host and
 * access type.
 */

static inline uint64_t handle_bswap(uint64_t val, int size, bool big_endian)
{
    if ((big_endian && NEED_BE_BSWAP) || (!big_endian && NEED_LE_BSWAP)) {
        switch (size) {
        case 1: return val;
        case 2: return bswap16(val);
        case 4: return bswap32(val);
        case 8: return bswap64(val);
        default:
            g_assert_not_reached();
        }
    } else {
        return val;
    }
}

/*
 * Load Helpers
 *
 * We support two different access types. SOFTMMU_CODE_ACCESS is
 * specifically for reading instructions from system memory. It is
 * called by the translation loop and in some helpers where the code
 * is disassembled. It shouldn't be called directly by guest code.
 */

static uint64_t load_helper(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr,
                            size_t size, bool big_endian,
                            bool code_read)
{
    uintptr_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = code_read ? entry->addr_code : entry->addr_read;
    const size_t tlb_off = code_read ?
        offsetof(CPUTLBEntry, addr_code) : offsetof(CPUTLBEntry, addr_read);
    unsigned a_bits = get_alignment_bits(get_memop(oi));
    void *haddr;
    uint64_t res;

    /* Handle CPU specific unaligned behaviour */
    if (addr & ((1 << a_bits) - 1)) {
        cpu_unaligned_access(ENV_GET_CPU(env), addr,
                             code_read ? MMU_INST_FETCH : MMU_DATA_LOAD,
                             mmu_idx, retaddr);
    }

    /* If the TLB entry is for a different page, reload and try again.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!victim_tlb_hit(env, mmu_idx, index, tlb_off,
                            addr & TARGET_PAGE_MASK)) {
            tlb_fill(ENV_GET_CPU(env), addr, size,
                     code_read ? MMU_INST_FETCH : MMU_DATA_LOAD,
                     mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = code_read ? entry->addr_code : entry->addr_read;
    }

    /* Handle an IO access.  */
    if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
        CPUIOTLBEntry *iotlbentry = &env->iotlb[mmu_idx][index];
        uint64_t tmp;

        if ((addr & (size - 1)) != 0) {
            goto do_unaligned_access;
        }

        tmp = io_readx(env, iotlbentry, mmu_idx, addr, retaddr,
                       tlb_addr & TLB_RECHECK,
                       code_read ? MMU_INST_FETCH : MMU_DATA_LOAD, size);
        return handle_bswap(tmp, size, big_endian);
    }

    /* Handle slow unaligned access (it spans two pages or IO).  */
    if (size > 1
        && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
                    >= TARGET_PAGE_SIZE)) {
        target_ulong addr1, addr2;
        tcg_target_ulong r1, r2;
        unsigned shift;
    do_unaligned_access:
        addr1 = addr & ~(size - 1);
        addr2 = addr1 + size;
        r1 = load_helper(env, addr1, oi, retaddr, size, big_endian, code_read);
        r2 = load_helper(env, addr2, oi, retaddr, size, big_endian, code_read);
        shift = (addr & (size - 1)) * 8;

        if (big_endian) {
            /* Big-endian combine.  */
            res = (r1 << shift) | (r2 >> ((size * 8) - shift));
        } else {
            /* Little-endian combine.  */
            res = (r1 >> shift) | (r2 << ((size * 8) - shift));
        }
        return res & MAKE_64BIT_MASK(0, size * 8);
    }

    haddr = (void *)((uintptr_t)addr + entry->addend);

    switch (size) {
    case 1:
        res = ldub_p(haddr);
        break;
    case 2:
        if (big_endian) {
            res = lduw_be_p(haddr);
        } else {
            res = lduw_le_p(haddr);
        }
        break;
    case 4:
        if (big_endian) {
            res = (uint32_t)ldl_be_p(haddr);
        } else {
            res = (uint32_t)ldl_le_p(haddr);
        }
        break;
    case 8:
        if (big_endian) {
            res = ldq_be_p(haddr);
        } else {
            res = ldq_le_p(haddr);
        }
        break;
    default:
        g_assert_not_reached();
    }

    return res;
}

/*
 * For the benefit of TCG generated code, we want to avoid the
 * complication of ABI-specific return type promotion and always
 * return a value extended to the register size of the host. This is
 * tcg_target_long, except in the case of a 32-bit host and 64-bit
 * data, and for that we always have uint64_t.
 *
 * We don't bother with this widened value for SOFTMMU_CODE_ACCESS.
 */

tcg_target_ulong __attribute__((flatten))
helper_ret_ldub_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                    uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 1, false, false);
}

tcg_target_ulong __attribute__((flatten))
helper_le_lduw_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 2, false, false);
}

tcg_target_ulong __attribute__((flatten))
helper_be_lduw_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 2, true, false);
}

tcg_target_ulong __attribute__((flatten))
helper_le_ldul_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 4, false, false);
}

tcg_target_ulong __attribute__((flatten))
helper_be_ldul_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 4, true, false);
}

uint64_t __attribute__((flatten))
helper_le_ldq_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                  uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 8, false, false);
}

uint64_t __attribute__((flatten))
helper_be_ldq_mmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                  uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 8, true, false);
}

/*
 * Provide signed versions of the load routines as well.  We can of course
 * avoid this for 64-bit data, or for 32-bit data on 32-bit host.
 */


tcg_target_ulong helper_ret_ldsb_mmu(CPUArchState *env, target_ulong addr,
                                     TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int8_t)helper_ret_ldub_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_le_ldsw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int16_t)helper_le_lduw_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_be_ldsw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int16_t)helper_be_lduw_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_le_ldsl_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int32_t)helper_le_ldul_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_be_ldsl_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int32_t)helper_be_ldul_mmu(env, addr, oi, retaddr);
}

/*
 * Store Helpers
 */

static void store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
                         TCGMemOpIdx oi, uintptr_t retaddr, size_t size,
                         bool big_endian)
{
    uintptr_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = tlb_addr_write(entry);
    const size_t tlb_off = offsetof(CPUTLBEntry, addr_write);
    unsigned a_bits = get_alignment_bits(get_memop(oi));
    void *haddr;

    /* Handle CPU specific unaligned behaviour */
    if (addr & ((1 << a_bits) - 1)) {
        cpu_unaligned_access(ENV_GET_CPU(env), addr, MMU_DATA_STORE,
                             mmu_idx, retaddr);
    }

    /* If the TLB entry is for a different page, reload and try again.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!victim_tlb_hit(env, mmu_idx, index, tlb_off,
            addr & TARGET_PAGE_MASK)) {
            tlb_fill(ENV_GET_CPU(env), addr, size, MMU_DATA_STORE,
                     mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_addr_write(entry) & ~TLB_INVALID_MASK;
    }

    /* Handle an IO access.  */
    if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
        CPUIOTLBEntry *iotlbentry = &env->iotlb[mmu_idx][index];

        if ((addr & (size - 1)) != 0) {
            goto do_unaligned_access;
        }

        io_writex(env, iotlbentry, mmu_idx,
                  handle_bswap(val, size, big_endian),
                  addr, retaddr, tlb_addr & TLB_RECHECK, size);
        return;
    }

    /* Handle slow unaligned access (it spans two pages or IO).  */
    if (size > 1
        && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
                     >= TARGET_PAGE_SIZE)) {
        int i;
        uintptr_t index2;
        CPUTLBEntry *entry2;
        target_ulong page2, tlb_addr2;
    do_unaligned_access:
        /*
         * Ensure the second page is in the TLB.  Note that the first page
         * is already guaranteed to be filled, and that the second page
         * cannot evict the first.
         */
        page2 = (addr + size) & TARGET_PAGE_MASK;
        index2 = tlb_index(env, mmu_idx, page2);
        entry2 = tlb_entry(env, mmu_idx, page2);
        tlb_addr2 = tlb_addr_write(entry2);
        if (!tlb_hit_page(tlb_addr2, page2)
            && !victim_tlb_hit(env, mmu_idx, index2, tlb_off,
                               page2 & TARGET_PAGE_MASK)) {
            tlb_fill(ENV_GET_CPU(env), page2, size, MMU_DATA_STORE,
                     mmu_idx, retaddr);
        }

        /*
         * XXX: not efficient, but simple.
         * This loop must go in the forward direction to avoid issues
         * with self-modifying code in Windows 64-bit.
         */
        for (i = 0; i < size; ++i) {
            uint8_t val8;
            if (big_endian) {
                /* Big-endian extract.  */
                val8 = val >> (((size - 1) * 8) - (i * 8));
            } else {
                /* Little-endian extract.  */
                val8 = val >> (i * 8);
            }
            store_helper(env, addr + i, val8, oi, retaddr, 1, big_endian);
        }
        return;
    }

    haddr = (void *)((uintptr_t)addr + entry->addend);

    switch (size) {
    case 1:
        stb_p(haddr, val);
        break;
    case 2:
        if (big_endian) {
            stw_be_p(haddr, val);
        } else {
            stw_le_p(haddr, val);
        }
        break;
    case 4:
        if (big_endian) {
            stl_be_p(haddr, val);
        } else {
            stl_le_p(haddr, val);
        }
        break;
    case 8:
        if (big_endian) {
            stq_be_p(haddr, val);
        } else {
            stq_le_p(haddr, val);
        }
        break;
    default:
        g_assert_not_reached();
        break;
    }
}

void __attribute__((flatten))
helper_ret_stb_mmu(CPUArchState *env, target_ulong addr, uint8_t val,
                   TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 1, false);
}

void __attribute__((flatten))
helper_le_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 2, false);
}

void __attribute__((flatten))
helper_be_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 2, true);
}

void __attribute__((flatten))
helper_le_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 4, false);
}

void __attribute__((flatten))
helper_be_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 4, true);
}

void __attribute__((flatten))
helper_le_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 8, false);
}

void __attribute__((flatten))
helper_be_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, 8, true);
}

/* First set of helpers allows passing in of OI and RETADDR.  This makes
   them callable from other helpers.  */

#define EXTRA_ARGS     , TCGMemOpIdx oi, uintptr_t retaddr
#define ATOMIC_NAME(X) \
    HELPER(glue(glue(glue(atomic_ ## X, SUFFIX), END), _mmu))
#define ATOMIC_MMU_DECLS NotDirtyInfo ndi
#define ATOMIC_MMU_LOOKUP atomic_mmu_lookup(env, addr, oi, retaddr, &ndi)
#define ATOMIC_MMU_CLEANUP                              \
    do {                                                \
        if (unlikely(ndi.active)) {                     \
            memory_notdirty_write_complete(&ndi);       \
        }                                               \
    } while (0)

#define DATA_SIZE 1
#include "atomic_template.h"

#define DATA_SIZE 2
#include "atomic_template.h"

#define DATA_SIZE 4
#include "atomic_template.h"

#ifdef CONFIG_ATOMIC64
#define DATA_SIZE 8
#include "atomic_template.h"
#endif

#if HAVE_CMPXCHG128 || HAVE_ATOMIC128
#define DATA_SIZE 16
#include "atomic_template.h"
#endif

/* Second set of helpers are directly callable from TCG as helpers.  */

#undef EXTRA_ARGS
#undef ATOMIC_NAME
#undef ATOMIC_MMU_LOOKUP
#define EXTRA_ARGS         , TCGMemOpIdx oi
#define ATOMIC_NAME(X)     HELPER(glue(glue(atomic_ ## X, SUFFIX), END))
#define ATOMIC_MMU_LOOKUP  atomic_mmu_lookup(env, addr, oi, GETPC(), &ndi)

#define DATA_SIZE 1
#include "atomic_template.h"

#define DATA_SIZE 2
#include "atomic_template.h"

#define DATA_SIZE 4
#include "atomic_template.h"

#ifdef CONFIG_ATOMIC64
#define DATA_SIZE 8
#include "atomic_template.h"
#endif

/* Code access functions.  */

uint8_t __attribute__((flatten))
helper_ret_ldb_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                    uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 1, false, true);
}

uint16_t __attribute__((flatten))
helper_le_ldw_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 2, false, true);
}

uint16_t __attribute__((flatten))
helper_be_ldw_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 2, true, true);
}

uint32_t __attribute__((flatten))
helper_le_ldl_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 4, false, true);
}

uint32_t __attribute__((flatten))
helper_be_ldl_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 4, true, true);
}

uint64_t __attribute__((flatten))
helper_le_ldq_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 8, false, true);
}

uint64_t __attribute__((flatten))
helper_be_ldq_cmmu(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
                   uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, 8, true, true);
}