// smallobject_mid_v35.c
// Phase v11a-3: MID v3.5 HotBox implementation
//
// Design:
//   - TLS-cached page for fast allocation
//   - Refill via ColdIface when page exhausted
//   - Retire via ColdIface when page is full
//   - Stats/Learner integration for observability

#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include "box/smallobject_mid_v35_box.h"
#include "box/smallobject_segment_mid_v3_box.h"
#include "box/smallobject_cold_iface_mid_v3_box.h"
#include "tiny_region_id.h"  // For tiny_region_id_write_header

// SmallPageMeta is defined in smallobject_segment_mid_v3_box.h
// (included via smallobject_cold_iface_mid_v3_box.h)

// ============================================================================
// TLS Context (per-thread fast path state)
// ============================================================================

typedef struct {
    void *page[8];           // Current page per class
    uint32_t offset[8];      // Allocation offset (slot index)
    uint32_t capacity[8];    // Slots per page per class
    SmallPageMeta_MID_v3 *meta[8];  // Page metadata for retire check
} SmallMidV35TlsCtx;

static __thread SmallMidV35TlsCtx tls_mid_v35_ctx = {0};

// ============================================================================
// Slot Configuration (C5/C6/C7)
// ============================================================================

// Slot sizes for C5, C6, C7
static const size_t g_slot_sizes[8] = {
    0,     // C0: not used
    0,     // C1: not used
    0,     // C2: not used
    0,     // C3: not used
    0,     // C4: not used (ULTRA handles this)
    384,   // C5: 257-384 bytes → 384 byte slots
    512,   // C6: 385-512 bytes → 512 byte slots
    1024,  // C7: 513-1024 bytes → 1024 byte slots (ULTRA handles this)
};

// Slots per 64KB page
static const uint32_t g_slots_per_page[8] = {
    0, 0, 0, 0, 0,
    170,  // C5: 65536 / 384 = 170
    128,  // C6: 65536 / 512 = 128
    64,   // C7: 65536 / 1024 = 64
};

// ============================================================================
// Init
// ============================================================================

void small_mid_v35_init(void) {
    // Initialize any global state if needed
    // For v11a-3: nothing to do (TLS is zero-initialized)
}

// ============================================================================
// Alloc
// ============================================================================

void* small_mid_v35_alloc(uint32_t class_idx, size_t size) {
    (void)size;  // Unused for now
    if (class_idx < 5 || class_idx > 7) return NULL;  // Only C5-C7

    SmallMidV35TlsCtx *ctx = &tls_mid_v35_ctx;

    // Fast path: allocate from TLS cached page
    if (ctx->page[class_idx] && ctx->offset[class_idx] < ctx->capacity[class_idx]) {
        size_t slot_size = g_slot_sizes[class_idx];
        void *base = (char*)ctx->page[class_idx] + ctx->offset[class_idx] * slot_size;
        ctx->offset[class_idx]++;

        // Update page metadata
        if (ctx->meta[class_idx]) {
            ctx->meta[class_idx]->alloc_count++;
        }

        // Write header (1-byte Tiny header with class_idx)
        // Note: Assumes HAKMEM_TINY_HEADER_CLASSIDX is enabled
        tiny_region_id_write_header(base, class_idx);

        // Return USER pointer (BASE + 1 byte header)
        return (char*)base + 1;
    }

    // Slow path: need new page via ColdIface
    SmallPageMeta_MID_v3 *page = small_cold_mid_v3_refill_page(class_idx);
    if (!page) {
        // Fallback to legacy or return NULL
        return NULL;
    }

    // Update TLS cache
    ctx->page[class_idx] = page->ptr;
    ctx->offset[class_idx] = 1;  // First slot already allocated
    ctx->capacity[class_idx] = g_slots_per_page[class_idx];
    ctx->meta[class_idx] = page;

    // Record first allocation in page metadata
    page->alloc_count = 1;

    // Write header for first slot
    tiny_region_id_write_header(page->ptr, class_idx);

    // Return first slot (USER pointer)
    return (char*)page->ptr + 1;
}

// ============================================================================
// Free (Simplified: just count, no freelist yet)
// ============================================================================

void small_mid_v35_free(void *ptr, uint32_t class_idx) {
    if (!ptr || class_idx < 5 || class_idx > 7) return;

    // For v11a-3: simplified free (just increment free_count)
    // In future phases: implement freelist for reuse

    // Calculate BASE from USER pointer
    void *base = (char*)ptr - 1;

    // Find page metadata via simple calculation
    // Note: Assumes 64KB pages aligned to 64KB boundary
    size_t page_size = 64 * 1024;  // 64KB
    void *page_base = (void*)((uintptr_t)base & ~(page_size - 1));

    // Check if this is the current TLS page
    SmallMidV35TlsCtx *ctx = &tls_mid_v35_ctx;
    SmallPageMeta_MID_v3 *meta = ctx->meta[class_idx];

    if (meta && meta->ptr == page_base) {
        // Free to current TLS page
        meta->free_count++;

        // Check if page is fully empty
        if (meta->free_count >= meta->capacity) {
            // Retire page via ColdIface
            small_cold_mid_v3_retire_page(meta);

            // Clear TLS cache for this class
            ctx->page[class_idx] = NULL;
            ctx->offset[class_idx] = 0;
            ctx->meta[class_idx] = NULL;
        }
    } else {
        // Different page: need RegionIdBox lookup
        // For v11a-3: simple fallback - just count the free
        // Real implementation needs proper page lookup
        // TODO: Implement cross-page free via RegionIdBox in v11b
    }
}