// pool_mid_tc.inc.h — Box: Mid Transfer Cache (per-thread inbox)
#ifndef POOL_MID_TC_INC_H
#define POOL_MID_TC_INC_H

typedef struct MidTC {
    atomic_uintptr_t inbox[POOL_NUM_CLASSES];
} MidTC;

#define MID_TC_BUCKETS 1024
typedef struct MidTCEntry { uint64_t tid; MidTC* tc; struct MidTCEntry* next; } MidTCEntry;
static pthread_mutex_t g_mid_tc_mu[MID_TC_BUCKETS];
static MidTCEntry* g_mid_tc_head[MID_TC_BUCKETS];
static __thread MidTC* t_mid_tc = NULL;
static int g_tc_enabled = 1;         // env: HAKMEM_TC_ENABLE (default 1)
static int g_tc_drain_unbounded = 1; // env: HAKMEM_TC_UNBOUNDED (default 1)
static int g_tc_drain_max = 0;       // env: HAKMEM_TC_DRAIN_MAX (0=unbounded)
static int g_tc_drain_trigger = 2;   // env: HAKMEM_TC_DRAIN_TRIGGER (ring->top < trigger)

static inline uint32_t mid_tc_hash(uint64_t tid) {
    tid ^= tid >> 33; tid *= 0xff51afd7ed558ccdULL; tid ^= tid >> 33; tid *= 0xc4ceb9fe1a85ec53ULL; tid ^= tid >> 33;
    return (uint32_t)(tid & (MID_TC_BUCKETS - 1));
}

// Thread-safe initialization using pthread_once
static pthread_once_t mid_tc_init_once_control = PTHREAD_ONCE_INIT;
static void mid_tc_init_impl(void) {
    for (int i = 0; i < MID_TC_BUCKETS; i++) {
        pthread_mutex_init(&g_mid_tc_mu[i], NULL);
        g_mid_tc_head[i] = NULL;
    }
}
static void mid_tc_init_once(void) {
    pthread_once(&mid_tc_init_once_control, mid_tc_init_impl);
}

static MidTC* mid_tc_get(void) {
    if (t_mid_tc) return t_mid_tc;
    mid_tc_init_once();
    MidTC* tc = (MidTC*)hkm_libc_calloc(1, sizeof(MidTC));  // P0 Fix: Use libc malloc
    if (!tc) return NULL;
    uint64_t tid = (uint64_t)(uintptr_t)pthread_self();
    uint32_t h = mid_tc_hash(tid);
    pthread_mutex_lock(&g_mid_tc_mu[h]);
    MidTCEntry* e = (MidTCEntry*)hkm_libc_malloc(sizeof(MidTCEntry));  // P0 Fix: Use libc malloc
    if (e) { e->tid = tid; e->tc = tc; e->next = g_mid_tc_head[h]; g_mid_tc_head[h] = e; }
    pthread_mutex_unlock(&g_mid_tc_mu[h]);
    t_mid_tc = tc;
    return tc;
}

static MidTC* mid_tc_lookup_by_tid(uint64_t tid) {
    mid_tc_init_once();
    uint32_t h = mid_tc_hash(tid);
    MidTCEntry* e = g_mid_tc_head[h];
    while (e) { if (e->tid == tid) return e->tc; e = e->next; }
    return NULL;
}

static inline void mid_tc_push(MidTC* tc, int class_idx, PoolBlock* b) {
    uintptr_t old_head;
    do {
        old_head = atomic_load_explicit(&tc->inbox[class_idx], memory_order_acquire);
        b->next = (PoolBlock*)old_head;
    } while (!atomic_compare_exchange_weak_explicit(&tc->inbox[class_idx], &old_head, (uintptr_t)b, memory_order_release, memory_order_relaxed));
}

static inline int mid_tc_drain_into_tls(int class_idx, PoolTLSRing* ring, PoolTLSBin* bin) {
    MidTC* tc = mid_tc_get();
    if (!tc) return 0;
    HKM_TIME_START(t_tc);
    uintptr_t head = atomic_exchange_explicit(&tc->inbox[class_idx], (uintptr_t)0, memory_order_acq_rel);
    if (!head) { HKM_TIME_END(HKM_CAT_TC_DRAIN, t_tc); return 0; }
    int moved = 0;
    int limit = (g_tc_drain_unbounded || g_tc_drain_max <= 0) ? INT32_MAX : g_tc_drain_max;
    PoolBlock* cur = (PoolBlock*)head;
    while (cur && moved < limit) {
        PoolBlock* nxt = cur->next;
        if (ring->top < POOL_L2_RING_CAP) {
            ring->items[ring->top++] = cur; moved++;
        } else {
            cur->next = bin->lo_head; bin->lo_head = cur; bin->lo_count++; moved++;
        }
        cur = nxt;
    }
    while (cur) { PoolBlock* nxt = cur->next; mid_tc_push(tc, class_idx, cur); cur = nxt; }
    HKM_TIME_END(HKM_CAT_TC_DRAIN, t_tc);
    return moved;
}

static inline int mid_tc_has_items(int class_idx) {
    MidTC* tc = t_mid_tc; // do not allocate on peek
    if (!tc) return 0;
    return atomic_load_explicit(&tc->inbox[class_idx], memory_order_relaxed) != 0;
}

#endif // POOL_MID_TC_INC_H