Boxify superslab registry, add bench profile, and document C7 hotpath experiments

CURRENT_TASK.md

@@ -31,6 +31,7 @@
   - FROZEN デフォルト（legacy プロファイル）：Page Box は C5〜C7 のみ ON、Warm は C0〜C7 すべて ON（C0〜C4 cap=4、C5〜C7 cap=8）。
   - ENV `HAKMEM_TINY_POLICY_PROFILE=legacy|c5_7_only|tinyplus_all` で切替可能（未指定は legacy）。
   - Stats は OBSERVE 用に積むだけ、Learner は空実装のまま。
+- mimalloc/system との最新ベンチ (Release, prefault デフォルト, policy=legacy, mode=2) を README_PERF に追記。C7-only 48.8M vs mimalloc 95.3M / system 73.9M、129–1024B 50.0M vs 128.4M / 97.7M、full 50.9M vs 123.6M / 83.5M、Tiny-only 8–128B 93.2M vs 123.7M / 66.3M。
 - TLS Bind Box の導入:
   - `core/box/ss_tls_bind_box.h` に `ss_tls_bind_one()` を追加し、「Superslab + slab_idx → TLS」のバインド処理（`superslab_init_slab` / `meta->class_idx` 設定 / `tiny_tls_bind_slab`）を 1 箇所に集約。
   - `superslab_refill()`（Shared Pool 経路）および Warm Pool 実験経路から、この Box を経由して TLS に接続するよう統一。
@@ -51,44 +52,68 @@
   - `core/box/tiny_class_policy_box.{h,c}` にクラス別ポリシー構造体 `TinyClassPolicy` と `tiny_policy_get(class_idx)` を追加。
     - FROZEN デフォルト: Page Box = C5–C7, Warm = 全クラス（C0–C4 cap=4 / C5–C7 cap=8）。
     - `HAKMEM_TINY_POLICY_PROFILE=legacy|c5_7_only|tinyplus_all` でプロファイル切替可能（未知値は legacy にフォールバック）。
-  - `core/box/tiny_class_stats_box.{h,c}` に OBSERVE 用の軽量カウンタ（UC miss / Warm hit / Shared Pool lock など）を追加。
+- `core/box/tiny_class_stats_box.{h,c}` に OBSERVE 用の軽量カウンタ（UC miss / Warm hit / Shared Pool lock など）を追加。
   - `core/box/tiny_policy_learner_box.{h,c}` に Learner の骨組みを追加（現状は FROZEN/OBSERVE モード向けの雛形）。
   - `core/front/tiny_unified_cache.c` / Page Box / Warm Pool 経路を `tiny_policy_get(class_idx)` ベースでゲートし、Hot path からは Policy Box を読む形に統一。
+- `bench_random_mixed` に RSS ダンプ（`getrusage(RUSAGE_SELF).ru_maxrss`）を追加し、各 allocator で ops/s と合わせて常駐メモリを記録できるようにした。  
+- 新規比較表 `PERF_COMPARISON_ALLOCATORS.md` を追加。C7-only / 129–1024B / 16–1024B で HAKMEM(full/larson_guard) は ~50M ops/s / ~29MB RSS、system は ~95–78M ops/s / ~1.6MB RSS、mimalloc は ~74–126M ops/s / ~1.8MB RSS。  
+- SS stats (HAKMEM_SS_STATS_DUMP=1, full profile, 16–1024B ws=256/1M): live Superslab は C2=1, C7=1（empty_events: C7=1）、RSS は ~29MB。予算を 2 に絞っても同じ配置で RSS 変化なし → RSS は Superslab 枚数より TLS/Warm/Page stack 等の常駐分が支配的。  
 
 ### 性能の現状（Random Mixed, HEAD）
-- 注記 (2025-12-05, policy legacy プロファイル試験値):
+- 条件: Release, `HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2 HAKMEM_WARM_C7_MAX=8 HAKMEM_WARM_C7_PREFETCH=4`, ws=256
-  - Release: `HAKMEM_TINY_PROFILE=full HAKMEM_TINY_POLICY_PROFILE=legacy ./bench_random_mixed_hakmem 1000000 256 42` → 約 4.9M ops/s（導入前 27M との乖離あり、要フォロー）。
+  - **C7-only (size=1024, iters=200K, ws=32)**  
-  - Release C7-only: `HAKMEM_BENCH_C7_ONLY=1 ... HAKMEM_TINY_POLICY_PROFILE=legacy` → 約 2.7M ops/s（空スラブガード導入前の遅さに戻っており要再調査）。
+    - policy=legacy: 47.3M / 47.3M / 43.9M ops/s（平均 ≈ **46M**）。C7 uc_miss=6660 / warm_hit=3329 / shared_lock=5 / tls_carve_success=3329。  
-- 条件: `bench_random_mixed_hakmem 1000000 256 42`（1T, ws=256, RELEASE, 16–1024B）
+    - policy=auto（Learner: score=lock*4+miss）: 45.6M / 44.6M / 39.7M ops/s（平均 ≈ **43–45M**）、統計は legacy と同一（C7 固定 ON）。  
-  - HAKMEM: 約 27.6M ops/s（C7 Warm/TLS 修復後）
+    - guard 比較: full **42.4M ops/s** vs larson_guard **40.7M ops/s**（-4%程度で安全側ガードを維持）。  
-  - system malloc: 約 90–100M ops/s
+  - **129–1024B (iters=1M, ws=256)**  
-  - mimalloc: 約 120–130M ops/s
+    - legacy: **51.5M ops/s**。C5 uc_miss=1/warm_hit=0/shared_lock=1、C6 uc_miss=1/warm_hit=0/shared_lock=2、C7 uc_miss=17196/warm_hit=8597/shared_lock=5/tls_carve_success=8597。  
-- 条件: `bench_random_mixed_hakmem 1000000 256 42` +
+    - auto: **51.9M ops/s**（Learner=lock 重視でも C7 のみ ON、統計ほぼ同じ）。  
-  `HAKMEM_BENCH_MIN_SIZE=8 HAKMEM_BENCH_MAX_SIZE=128`（Tiny-only, 8–128B）
+    - guard 比較: full **49.0M ops/s** vs larson_guard **48.4M ops/s**（-1.2%）。  
-  - HAKMEM Tiny Front: 約 80–90M ops/s（mimalloc と同オーダー）
+  - **full random_mixed 16–1024B (iters=1M, ws=256)**  
-- 条件: `bench_random_mixed_hakmem 1000000 256 42` +
+    - legacy: **51.0M ops/s**。C7 uc_miss=16702/warm_hit=8350/shared_lock=5/tls_carve_success=8350（C5/C6 は uc_miss=1〜2）。  
-  `HAKMEM_BENCH_MIN_SIZE=129 HAKMEM_BENCH_MAX_SIZE=1024`（Tiny C5–C7 のみ）
+    - auto: **50.0M ops/s**（C7 固定 ON のまま、他クラスはほぼ動かず）。  
-  - HAKMEM: 約 28.0M ops/s（Warm/TLS ガード適用後）
+- 補足:
-- 条件: C7 専用 micro-bench（Debug, `HAKMEM_BENCH_C7_ONLY=1 HAKMEM_TINY_PROFILE=full HAKMEM_WARM_C7_MAX=8 HAKMEM_WARM_C7_PREFETCH=4` ほか）
+  - WarmPool-STATS と TinyClassStats を統合。`HAKMEM_WARM_POOL_STATS=1` で C7-only 実行時に hits=3329 / misses=1 / prefilled=1 を確認（TinyClassStats の warm_hit=3329 と一致）。
-  - mode 0（Legacy Warm）: 約 2.0M ops/s、C7 Warm ヒット 0・Shared Pool ロック多数（`slab_carve_from_ss` が 0 を頻発）
+  - `TinyClassPolicy` に `tls_carve_enabled` を追加し、デフォルトで C5–C7 を ON。`TinyClassStats` に tls_carve_attempt/success を追加済み。
-  - mode 1（Bind-only）: 約 20M ops/s（iters=200K, ws=32）、Warm hit ≈100%・Shared Pool ロック 5 回まで減少
+  - Learner のスコアを `score = shared_lock * 4 + uc_miss` に変更済み（auto プロファイル専用）。現状のワークロードでは C7 が圧倒的に優勢で、C5/C6 はまだほぼ選ばれない。
-  - mode 2（Bind+TLS carve 実験）: mode 1 と同等〜わずかに上（UC ミスは増えるが `uc_miss_tls` に集中し、avg_refill は短縮）
+
-- 条件: C7 専用 micro-bench（Release, `HAKMEM_BENCH_C7_ONLY=1 HAKMEM_TINY_PROFILE=full HAKMEM_WARM_C7_MAX=8 HAKMEM_WARM_C7_PREFETCH=4`）
+### サイズ→クラス対応（HAKMEM_TINY_HEADER_CLASSIDX=1 のため size+1 で判定）
-  - HAKMEM: 約 18.8M ops/s（空スラブ強制ガード + リセット導入後、Debug と同オーダーまで回復）
+- `hak_tiny_size_to_class(size)` は `needed=size+1` で `g_size_to_class_lut_2k` を引くため、512B 要求は 513B として class 7 判定になる（現状の挙動は仕様どおり）。
-- 結論:
+- 代表サイズのマップ（データサイズ→class_idx / 総バイト数）  
-  - Tiny front 自体（8–128B）は十分速く、mimalloc と同オーダーまで出ている。
+  - 8B → C1（16B stride）  
-  - C5–C7 経路は「満杯 C7 slab を Warm に再供給していた」問題を空スラブ限定ガード＋Release/Debug 共通リセットで解消し、
+  - 16B → C2（32B）  
-    C7-only Release も ~18.8M ops/s に回復。Random Mixed Release も 27M クラスまで改善。
+  - 32B → C3（64B）  
+  - 64B → C4（128B）  
+  - 128B → C5（256B）  
+  - 256B → C6（512B）  
+  - 512B → C7（2048B stride / 32 blocks per slab）  
+  - 1024B → C7（同上）  
+- 512B 固定ベンチで C7 経路が動くのはこのヘッダ加算による設計上の結果。現時点では「C7 支配」を前提に C5/C6 は拡張枠として観測を続ける。
+
+### C5/C6 専用ワークロードの速報（Release, ws=512, iters=1,000,000, size fixed）
+- 条件: `HAKMEM_BENCH_MIN_SIZE=256 HAKMEM_BENCH_MAX_SIZE=256 (実質 C6)`、`HAKMEM_TINY_PROFILE=full`、`HAKMEM_WARM_TLS_BIND_C7=2`、`HAKMEM_TINY_STATS_DUMP=1`
+  - policy=legacy: Throughput ≈ **89.9M ops/s**。C6: uc_miss=5, warm_hit=1, shared_lock=2, tls_carve_attempt=1, tls_carve_success=1。
+  - policy=auto: Throughput ≈ **87.5M ops/s**。C6 の統計はほぼ同じ（uc_miss=5, warm_hit=1, tls_carve_attempt/success=1）。C5 ほぼ負荷なし。
+- 補足: C5/C6 はワーキングセットを広げても Warm/TLS carve のヒットは少数（キャッシュヒット優位なため）。専用負荷を増やす場合はさらに ws を広げて観測予定。
+- Larson ベンチ（Release, 10 runs, `./test_larson.sh`）  
+  - profile=full: 1.15〜1.26M ops/s  
+  - profile=larson_guard: 1.10〜1.27M ops/s（≈-3〜0%でほぼ同等）。`HAKMEM_SS_STATS_DUMP=1` で Superslab live が 1 前後に収まり、SEGV/OOM なし。サンプルログは `docs/analysis/SUPERSLAB_STATS_SNAPSHOT.md` に記録。
+
+### 新しいログ/ENV スイッチ
+- `HAKMEM_TINY_POLICY_LOG=0/1`: Policy 初期化/auto update のログ抑制（デフォルト ON）。
+- `HAKMEM_TINY_WARM_LOG=0/1`: C7 prefill 関連ログ（PREFILL_META/skip 等）の抑制（デフォルト ON）。
+- `HAKMEM_TINY_PAGEBOX_LOG=0/1`: Page Box の登録ログ抑制（Debug のみ、デフォルト ON）。
+- 長時間ラン時は上記を 0 にしてノイズを抑える運用を推奨。短時間デバッグ時のみ 1 にする。
 
 ### 次にやること（広い条件での安定化確認）
 1. `HAKMEM_BENCH_MIN_SIZE=129 HAKMEM_BENCH_MAX_SIZE=1024` や通常の `bench_random_mixed_hakmem 1000000 256 42` で
-   空スラブ限定ガードが副作用なく動くかを継続確認（現状 Release で 27–28M ops/s を確認済み）。
+   空スラブ限定ガードが副作用なく動くかを継続確認（現状 Release で 29–30M ops/s を確認済み）。
 2. ドキュメント更新:
    - Release だけ C7 Warm が死んでいた根本原因 = 満杯 C7 slab を Shared Pool がリセットせず再供給していた。
-   - Acquire の空スラブ強制ガード＋Release/Debug 共通リセットで C7-only Release が ~18.8M ops/s まで回復した。
+   - Acquire の空スラブ強制ガード＋Stage3(LRU) 再利用時の Superslab 全スロットリセット＋Warm/TLS carve 有効化で、
+     C7-only Release が ~20–25M ops/s クラスに回復し、Random Mixed 16–1024B Release も ~29–30M ops/s クラスまで改善した。
 3. 次フェーズ案:
-   - C5/C6 でも同様の Warm/TLS 最適化・空スラブガードを適用するか、
+   - Superslab ガード（Stats/Reset/Stage3/Budget/larson_guard）まで完了。以降は mimalloc/system との比較最適化や、必要に応じた C5/C6 Tiny-Plus 拡張を検討。
-   - Random Mixed 全体のボトルネック（Shared Pool ロック/Wrapper/mid-size path など）を洗うかを選択。
 
 ### 次フェーズ（Tiny 全クラス向け Page Box / Warm / Policy 汎用化の検討）
 - 方向性:
@@ -109,7 +134,67 @@
   - `TinyClassPolicyBox`/`TinyClassStatsBox`/`TinyPolicyLearnerBox` を追加し、デフォルトで C5〜C7 に Page Box + Warm を許可（Warm cap=8）。  
   - unified_cache_refill の Page/Warm 経路は `tiny_policy_get()` の返り値でゲートし、Warm push は per-class cap を尊重。  
   - Page Box 初期化もデフォルトで C5〜C7 を有効化。OBSERVE 用の軽量 stats increment を UC miss / Warm hit に接続済み。
+- 次ステップの設計メモ:
+  - TinyPageBoxContext を class 汎用構造に広げ、C5/C6 も「TLS Bind で page 登録 → UC refill で page 内 freelist からバッチ供給」を C7 と共有できるようにする（実装は未着手、設計メモのみ）。
 
 ### メモ
 - ページフォルト問題は Prefault Box + ウォームアップで一定水準まで解消済みで、現在の主ボトルネックはユーザー空間の箱（Unified Cache / free / Pool）側に移っている。
 - 以降の最適化は「箱を削る」ではなく、「HOT 層で踏む箱を減らし、Tiny 的なシンプル経路と Tiny-Plus 経路（Page Box + Warm）をクラス別ポリシーでどう使い分けるか」にフォーカスする。
+
+## 今後のフォーカス（C7 支配を前提に一旦整理）
+- 設計明記: 257–512→C6, 513–2048→C7（size+1 判定）。実負荷は C7 が受ける設計として確定。C5/C6 は拡張枠・観測対象。
+- 優先度: C5-only ≈91M ops/s、512B 固定も C7 経路で ≈47M ops/s → C5/C6 最適化は auto/実験用に留め、本命は C7 Tiny-Plus＋Policy。
+- プロファイル運用: legacy=本番、auto=C7固定＋上位2クラス観測用のまま据え置き。学習拡張は新ワークロードで C5/C6 がホットになった際に検討。
+- 次の大きな箱候補: (1) mimalloc/system とのフルベンチ整理（論文/README 更新）、(2) hakorune 側 PHI/JoinIR の開発にリソースを戻す。
+
+## 巨大 BSS グローバルの棚卸しと今後
+
+- `nm -S --size-sort bench_random_mixed_hakmem` と SS_STATS のサンプルから、RSS を支配しているのは Tiny 層ではなく巨大 BSS 配列であることを確認。
+  - 代表例: `g_super_reg` ≈24MB, `g_shared_pool` ≈2.3MB, `g_super_reg_by_class` ≈1MB, `g_rem_side` ≈1MB など。
+  - SS_STATS（ws=64, iters=10k）では live Superslab は C2=1, C7=1 程度で、巨大レジストリの大半は未使用キャパシティになっている。
+  - Tiny 用メモリ会計 Box（`tiny_mem_stats_box`）では UC/Warm/Page/TLS/Policy-Stats 合計でも ≈40KB 程度と判明し、RSS≈29MB の主因ではないことを確認。
+- docs/analysis/LARGE_GLOBALS_OVERVIEW.md に各大型シンボルのサイズ/役割と SS_STATS とのギャップを一覧化済み。
+
+次フェーズ候補:
+- Superslab Registry / Shared Pool / Remote Queue を Box 化し、プロファイル別に「必要なだけ動的確保」できる SuperRegBox / SharedPoolBox / RemoteSideBox への移行を検討。
+- `HAKMEM_PROFILE` や ENV から「bench 向け縮小設定」と「本番向けフル設定」を切り替えられるようにし、RSS を抑えつつ Box 構造は維持する。
+
+進捗（巨大BSS Box化フェーズ）
+- docs/analysis/LARGE_GLOBALS_OVERVIEW.md に大型シンボルの定義元・役割・縮小目安を追記（SuperReg/SharedPool/Remote など）。
+- 設計スタブを追加:
+  - `core/box/super_reg_box.h` … レジストリ容量をプロファイルで切替するための API メモ。
+  - `core/box/shared_pool_box.h` … Shared Pool の容量/ガードをプロファイルに紐づけるための API メモ。
+  - `core/box/remote_side_box.h` … Remote Queue テーブルをプロファイルで縮小するための API メモ。
+- `HAKMEM_PROFILE=bench` を追加し、SuperReg/SharedPool/Remote の「論理有効スロット」を 1/8〜1/16 に制限するラッパを実装（配列は現状サイズのまま）。`bench_random_mixed_hakmem` は full/bench ともビルド・完走済み。C7-only/129–1024B/16–1024B で ops/s は ±数% 以内、RSS は ~32.6MB でほぼ不変（論理制限のみのため）。  
+- SuperReg/Remote を Box 内で動的確保に置き換え、`HAKMEM_PROFILE=bench` では実容量も縮小（SuperReg: 1/8〜1/16、Remote: log2 を 12〜）。C7-only 200k/ws32 では full=29.6MB → bench=7.2MB (ops ≈44.4M 同レンジ) まで RSS を削減できた。
+- bench 実容量版での広いワークロード検証: 129–1024B ws=256/1M は full=48.9M ops/s & 29.6MB → bench=49.2M & 7.2MB。16–1024B ws=256/1M は full=48.3M & 29.7MB → bench=48.8M & 7.2MB。SS_STATS（bench）でも live Superslab は C2=1, C7=1 に収まり、Tiny 層メモリは ~41KB のまま。
+- 次ステップ: SharedPool 側も必要なら動的化/縮小を検討しつつ、RSS をさらに攻めるか、CPU パス最適化に戻るか判断。***
+
+### フェーズ整理と次の方針
+- SharedPool は現状サイズを維持し、`HAKMEM_PROFILE=full` を本番、`HAKMEM_PROFILE=bench` を対 mimalloc/system の軽量プロファイルとして運用（bench は SuperReg/Remote 縮小済み、RSS≈7.2MB）。
+- 巨大BSS Box化フェーズは「bench で RSS≈7.2MB / ops≈同等」まで完了。今後は perf（CPUサイクル）最適化にフォーカス。
+
+ホットパス perf フェーズの TODO（案）
+1. tiny_alloc_fast / tiny_free_fast_v2 の再プロファイル：残存分岐・間接呼び出し・重い箱を特定。
+2. Unified Cache ヒットパスを最短化：ヒット時を 1–2 load + 軽分岐に近づける（必要なら C7 専用インライン版検討）。
+3. free パス Gatekeeper/Box の再配線：C7 ホットケースだけ分岐極小のストレートラインにする。
+
+目標: シングルスレッド小オブジェクトで ~50M ops/s → 70M〜80M 帯を狙う（mimalloc との差を半減イメージ）。***
+
+補足（CPU ホットパス観測メモ）
+- `HAKMEM_PROFILE=bench HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2` で perf を試行したが、`perf_event_paranoid` 制約で `cycles` が取れず page-fault サンプルのみ（`__memset_avx2_unaligned_erms` が warmup を支配）。`perf.data` は即削除済み。集計結果と次の測定案は `docs/analysis/CPU_HOTPATH_OVERVIEW.md` に記載。
+- C7 alloc/free flattening と UC ヒット簡略化の設計メモを追加：`docs/analysis/C7_HOTPATH_FLATTENING.md`, `docs/analysis/C7_FREE_HOTPATH.md`。実装はこれから。
+- C7 ホットパス用フックを追加（`core/box/tiny_c7_hotpath_box.h` + `HAKMEM_TINY_C7_HOT`）。今は既存 Hot/Cold Box をクラス固定で呼ぶ薄いラッパなので挙動は同一。  
+  - bench プロファイルの perf stat (3 run 平均):  
+    - 16–1024B: cycles≈109.5M, inst≈233.5M (IPC≈2.13, br-miss≈2.90%)  
+    - 16–1024B + `HAKMEM_TINY_C7_HOT=1`: cycles≈111.8M, inst≈242.1M (IPC≈2.16, br-miss≈2.75%)
+
+### C7 ホットパス平坦化（第1段階）の結果メモ
+- `HAKMEM_PROFILE=bench HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2`、129–1024B ws=256/1M（Release）で:
+  - `HAKMEM_TINY_C7_HOT=0`: ≈49.7M ops/s
+  - `HAKMEM_TINY_C7_HOT=1`: ≈46.7M ops/s（分岐ミスは僅かに改善するがスループットはノイズ〜微減）
+- 16–1024B ws=256/1M では:
+  - hot=0: ops≈47.4M, IPC≈2.13, br-miss≈2.90%
+  - hot=1: ops≈47.4–47.6M, IPC≈2.16, br-miss≈2.75%
+- 現状の C7 ホットパス実装は「ヒット専用 UC ＋ TLS→UC→cold 直線化」の初期版で、大幅な伸びはまだ無い。回帰はなく、分岐ミス率はわずかに改善。今後さらに UC ヒット専用関数の最短化や free 側の直線化を詰める余地あり。
+- 方針: `HAKMEM_TINY_C7_HOT` は実験用スイッチとして残し、デフォルト OFF。perf フェーズは bench プロファイルで ≈50M ops/s / RSS ≈7MB を維持できる現行経路を基準に一旦完了とする。***

Makefile

@@ -219,12 +219,12 @@ LDFLAGS += $(EXTRA_LDFLAGS)
 
 # Targets
 TARGET = test_hakmem
-OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/free_publish_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/pagefault_telemetry_box.o core/box/tiny_sizeclass_hist_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/page_arena.o core/front/tiny_unified_cache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o test_hakmem.o
+OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/free_publish_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/pagefault_telemetry_box.o core/box/tiny_sizeclass_hist_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/ss_budget_box.o core/box/tiny_mem_stats_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/box/super_reg_box.o core/box/shared_pool_box.o core/box/remote_side_box.o core/page_arena.o core/front/tiny_unified_cache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o test_hakmem.o
 OBJS = $(OBJS_BASE)
 
 # Shared library
 SHARED_LIB = libhakmem.so
-SHARED_OBJS = hakmem_shared.o hakmem_config_shared.o hakmem_tiny_config_shared.o hakmem_ucb1_shared.o hakmem_bigcache_shared.o hakmem_pool_shared.o hakmem_l25_pool_shared.o hakmem_site_rules_shared.o hakmem_tiny_shared.o core/box/ss_allocation_box_shared.o superslab_stats_shared.o superslab_cache_shared.o superslab_ace_shared.o superslab_slab_shared.o superslab_backend_shared.o core/superslab_head_stub_shared.o hakmem_smallmid_shared.o core/box/superslab_expansion_box_shared.o core/box/integrity_box_shared.o core/box/mailbox_box_shared.o core/box/front_gate_box_shared.o core/box/front_gate_classifier_shared.o core/box/free_publish_box_shared.o core/box/capacity_box_shared.o core/box/carve_push_box_shared.o core/box/prewarm_box_shared.o core/box/ss_hot_prewarm_box_shared.o core/box/front_metrics_box_shared.o core/box/bench_fast_box_shared.o core/box/ss_addr_map_box_shared.o core/box/slab_recycling_box_shared.o core/box/pagefault_telemetry_box_shared.o core/box/tiny_sizeclass_hist_box_shared.o core/box/tiny_env_box_shared.o core/box/tiny_route_box_shared.o core/box/tiny_page_box_shared.o core/box/tiny_class_policy_box_shared.o core/box/tiny_class_stats_box_shared.o core/box/tiny_policy_learner_box_shared.o core/box/wrapper_env_box_shared.o core/page_arena_shared.o core/front/tiny_unified_cache_shared.o core/tiny_alloc_fast_push_shared.o core/link_stubs_shared.o core/tiny_failfast_shared.o tiny_sticky_shared.o tiny_remote_shared.o tiny_publish_shared.o tiny_debug_ring_shared.o hakmem_tiny_magazine_shared.o hakmem_tiny_stats_shared.o hakmem_tiny_sfc_shared.o hakmem_tiny_query_shared.o hakmem_tiny_rss_shared.o hakmem_tiny_registry_shared.o hakmem_tiny_remote_target_shared.o hakmem_tiny_bg_spill_shared.o tiny_adaptive_sizing_shared.o hakmem_super_registry_shared.o hakmem_shared_pool_shared.o hakmem_shared_pool_acquire_shared.o hakmem_shared_pool_release_shared.o hakmem_elo_shared.o hakmem_batch_shared.o hakmem_p2_shared.o hakmem_sizeclass_dist_shared.o hakmem_evo_shared.o hakmem_debug_shared.o hakmem_sys_shared.o hakmem_whale_shared.o hakmem_policy_shared.o hakmem_ace_shared.o hakmem_ace_stats_shared.o hakmem_ace_controller_shared.o hakmem_ace_metrics_shared.o hakmem_ace_ucb1_shared.o hakmem_prof_shared.o hakmem_learner_shared.o hakmem_size_hist_shared.o hakmem_learn_log_shared.o hakmem_syscall_shared.o tiny_fastcache_shared.o
+SHARED_OBJS = hakmem_shared.o hakmem_config_shared.o hakmem_tiny_config_shared.o hakmem_ucb1_shared.o hakmem_bigcache_shared.o hakmem_pool_shared.o hakmem_l25_pool_shared.o hakmem_site_rules_shared.o hakmem_tiny_shared.o core/box/ss_allocation_box_shared.o superslab_stats_shared.o superslab_cache_shared.o superslab_ace_shared.o superslab_slab_shared.o superslab_backend_shared.o core/superslab_head_stub_shared.o hakmem_smallmid_shared.o core/box/superslab_expansion_box_shared.o core/box/integrity_box_shared.o core/box/mailbox_box_shared.o core/box/front_gate_box_shared.o core/box/front_gate_classifier_shared.o core/box/free_publish_box_shared.o core/box/capacity_box_shared.o core/box/carve_push_box_shared.o core/box/prewarm_box_shared.o core/box/ss_hot_prewarm_box_shared.o core/box/front_metrics_box_shared.o core/box/bench_fast_box_shared.o core/box/ss_addr_map_box_shared.o core/box/slab_recycling_box_shared.o core/box/pagefault_telemetry_box_shared.o core/box/tiny_sizeclass_hist_box_shared.o core/box/tiny_env_box_shared.o core/box/tiny_route_box_shared.o core/box/tiny_page_box_shared.o core/box/tiny_class_policy_box_shared.o core/box/tiny_class_stats_box_shared.o core/box/tiny_policy_learner_box_shared.o core/box/ss_budget_box_shared.o core/box/tiny_mem_stats_box_shared.o core/box/wrapper_env_box_shared.o core/page_arena_shared.o core/front/tiny_unified_cache_shared.o core/tiny_alloc_fast_push_shared.o core/link_stubs_shared.o core/tiny_failfast_shared.o tiny_sticky_shared.o tiny_remote_shared.o tiny_publish_shared.o tiny_debug_ring_shared.o hakmem_tiny_magazine_shared.o hakmem_tiny_stats_shared.o hakmem_tiny_sfc_shared.o hakmem_tiny_query_shared.o hakmem_tiny_rss_shared.o hakmem_tiny_registry_shared.o hakmem_tiny_remote_target_shared.o hakmem_tiny_bg_spill_shared.o tiny_adaptive_sizing_shared.o hakmem_super_registry_shared.o hakmem_shared_pool_shared.o hakmem_shared_pool_acquire_shared.o hakmem_shared_pool_release_shared.o hakmem_elo_shared.o hakmem_batch_shared.o hakmem_p2_shared.o hakmem_sizeclass_dist_shared.o hakmem_evo_shared.o hakmem_debug_shared.o hakmem_sys_shared.o hakmem_whale_shared.o hakmem_policy_shared.o hakmem_ace_shared.o hakmem_ace_stats_shared.o hakmem_ace_controller_shared.o hakmem_ace_metrics_shared.o hakmem_ace_ucb1_shared.o hakmem_prof_shared.o hakmem_learner_shared.o hakmem_size_hist_shared.o hakmem_learn_log_shared.o hakmem_syscall_shared.o tiny_fastcache_shared.o core/box/super_reg_box_shared.o core/box/shared_pool_box_shared.o core/box/remote_side_box_shared.o
 
 # Pool TLS Phase 1 (enable with POOL_TLS_PHASE1=1)
 ifeq ($(POOL_TLS_PHASE1),1)
@@ -251,7 +251,7 @@ endif
 # Benchmark targets
 BENCH_HAKMEM = bench_allocators_hakmem
 BENCH_SYSTEM = bench_allocators_system
-BENCH_HAKMEM_OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/free_publish_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/pagefault_telemetry_box.o core/box/tiny_sizeclass_hist_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/page_arena.o core/front/tiny_unified_cache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o bench_allocators_hakmem.o
+BENCH_HAKMEM_OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/free_publish_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/pagefault_telemetry_box.o core/box/tiny_sizeclass_hist_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/ss_budget_box.o core/box/tiny_mem_stats_box.o core/box/c7_meta_used_counter_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/box/super_reg_box.o core/box/shared_pool_box.o core/box/remote_side_box.o core/page_arena.o core/front/tiny_unified_cache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o bench_allocators_hakmem.o
 BENCH_HAKMEM_OBJS = $(BENCH_HAKMEM_OBJS_BASE)
 ifeq ($(POOL_TLS_PHASE1),1)
 BENCH_HAKMEM_OBJS += pool_tls.o pool_refill.o pool_tls_arena.o pool_tls_registry.o pool_tls_remote.o
@@ -428,7 +428,7 @@ test-box-refactor: box-refactor
 	./larson_hakmem 10 8 128 1024 1 12345 4
 
 # Phase 4: Tiny Pool benchmarks (properly linked with hakmem)
-TINY_BENCH_OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/free_publish_box.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/tiny_sizeclass_hist_box.o core/box/pagefault_telemetry_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/page_arena.o core/front/tiny_unified_cache.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o
+TINY_BENCH_OBJS_BASE = hakmem.o hakmem_config.o hakmem_tiny_config.o hakmem_ucb1.o hakmem_bigcache.o hakmem_pool.o hakmem_l25_pool.o hakmem_site_rules.o hakmem_tiny.o core/box/ss_allocation_box.o superslab_stats.o superslab_cache.o superslab_ace.o superslab_slab.o superslab_backend.o core/superslab_head_stub.o hakmem_smallmid.o core/box/superslab_expansion_box.o core/box/integrity_box.o core/box/mailbox_box.o core/box/front_gate_box.o core/box/front_gate_classifier.o core/box/free_publish_box.o core/box/capacity_box.o core/box/carve_push_box.o core/box/prewarm_box.o core/box/ss_hot_prewarm_box.o core/box/front_metrics_box.o core/box/bench_fast_box.o core/box/ss_addr_map_box.o core/box/slab_recycling_box.o core/box/pagefault_telemetry_box.o core/box/tiny_sizeclass_hist_box.o core/box/tiny_env_box.o core/box/tiny_route_box.o core/box/tiny_page_box.o core/box/tiny_class_policy_box.o core/box/tiny_class_stats_box.o core/box/tiny_policy_learner_box.o core/box/ss_budget_box.o core/box/tiny_mem_stats_box.o core/box/c7_meta_used_counter_box.o core/box/wrapper_env_box.o core/box/ptr_trace_box.o core/box/link_missing_stubs.o core/box/super_reg_box.o core/box/shared_pool_box.o core/box/remote_side_box.o core/page_arena.o core/front/tiny_unified_cache.o tiny_sticky.o tiny_remote.o tiny_publish.o tiny_debug_ring.o hakmem_tiny_magazine.o hakmem_tiny_stats.o hakmem_tiny_sfc.o hakmem_tiny_query.o hakmem_tiny_rss.o hakmem_tiny_registry.o hakmem_tiny_remote_target.o hakmem_tiny_bg_spill.o tiny_adaptive_sizing.o hakmem_super_registry.o hakmem_shared_pool.o hakmem_shared_pool_acquire.o hakmem_shared_pool_release.o hakmem_elo.o hakmem_batch.o hakmem_p2.o hakmem_sizeclass_dist.o hakmem_evo.o hakmem_debug.o hakmem_sys.o hakmem_whale.o hakmem_policy.o hakmem_ace.o hakmem_ace_stats.o hakmem_prof.o hakmem_learner.o hakmem_size_hist.o hakmem_learn_log.o hakmem_syscall.o hakmem_ace_metrics.o hakmem_ace_ucb1.o hakmem_ace_controller.o tiny_fastcache.o core/tiny_alloc_fast_push.o core/link_stubs.o core/tiny_failfast.o
 TINY_BENCH_OBJS = $(TINY_BENCH_OBJS_BASE)
 ifeq ($(POOL_TLS_PHASE1),1)
 TINY_BENCH_OBJS += pool_tls.o pool_refill.o core/pool_tls_arena.o pool_tls_registry.o pool_tls_remote.o

PERF_COMPARISON_ALLOCATORS.md

@@ -0,0 +1,27 @@
+# Allocator Throughput / RSS Comparison (Release)
+
+環境: 1 thread, `HAKMEM_WARM_TLS_BIND_C7=2`, RSS は `ru_maxrss` (KB) を MB 換算。  
+Hakmem は `full` と `larson_guard` プロファイルを計測。その他は system / mimalloc の素の挙動。
+
+| workload             | allocator              | ops/s            | max RSS (MB) |
+|----------------------|------------------------|------------------|--------------|
+| C7-only (1024B, ws32, 200k) | hakmem-full         | 44,381,807       | 29.6 |
+|                      | hakmem-bench          | 44,439,813       | **7.2** |
+|                      | hakmem-larson_guard    | 48,455,082       | 28.9 |
+|                      | mimalloc               | 74,433,394       | 1.8 |
+|                      | system                 | 78,514,783       | 1.6 |
+| 129–1024B (ws256, 1M) | hakmem-full           | 48,895,987       | 29.6 |
+|                      | hakmem-bench          | 49,226,419       | **7.2** |
+|                      | hakmem-larson_guard    | 52,327,019       | 28.8 |
+|                      | mimalloc               | 106,310,868      | 1.9 |
+|                      | system                 | 95,633,188       | 1.6 |
+| 16–1024B (ws256, 1M) | hakmem-full            | 48,276,749       | 29.7 |
+|                      | hakmem-bench          | 48,759,807       | **7.2** |
+|                      | hakmem-larson_guard    | 50,494,992       | 28.9 |
+|                      | mimalloc               | 126,403,649      | 1.9 |
+|                      | system                 | 95,361,993       | 1.6 |
+
+所感 (現時点):
+- スループットは system/mimalloc が優勢。Hakmem (full/guard) は C7 特化ワークロードで 44–48M ops/s 帯。
+- bench プロファイルを「実配列縮小」版に切り替えたことで、C7-only/129–1024B/16–1024B いずれも RSS は ~29MB → ~7MB まで低減（ops/s は同レンジ）。
+- RSS は system/mimalloc が圧倒的に小さい (1.6–1.9MB)。Hakmem は full/guard で ~29MB、bench 版は 7MB 前後まで圧縮できた。***

README_PERF_ANALYSIS.md

@@ -1,8 +1,35 @@
 # HAKMEM Allocator Performance Analysis Results
 
-**最新メモ (2025-12-05)**: C7 Warm/TLS Bind は本番経路を Bind-only (mode=1) に統一。Debug では `HAKMEM_WARM_TLS_BIND_C7=0/1/2` で切替可能だが、Release は常に mode=1 固定。C7-only ワークロードでは mode=1 が legacy (mode=0) 比で ~4–10x 速く、mode=2 は TLS carve 実験として残置。  
+**最新メモ (2025-12-06, Release)**  
-**追記 (2025-12-05, Release 修復)**: Release だけ C7 Warm が死んでいた原因は「満杯 C7 slab が Shared Pool に居残り、空スラブが Warm に渡っていなかった」こと。Acquire で C7 は空スラブ限定、Release でメタをリセットするガードを導入し、C7-only Release で ~18.8M ops/s、Random Mixed Release で ~27–28M ops/s まで回復。  
+- 新規比較表: `PERF_COMPARISON_ALLOCATORS.md` に HAKMEM (full/larson_guard) / mimalloc / system の ops/s と RSS を掲載。C7-only/129–1024/full いずれも HAKMEM は ~50M ops/s / ~29MB RSS、system/mimalloc は 75–126M ops/s / 1.6–1.9MB RSS で優位。  
-**追記 (2025-12-05, Policy Box)**: `TinyClassPolicyBox` を導入し、`HAKMEM_TINY_POLICY_PROFILE=legacy|c5_7_only|tinyplus_all` で Page/Warm ポリシーを切替可能にした。現状 legacy（PageBox= C5–C7, Warm= 全クラス cap 4/8）でランダム混在 Release は ~4.9M ops/s と低下しており、Warm 道の有効化状態を追加調査中。
+- Random Mixed 129–1024B, ws=256, iters=1M, `HAKMEM_WARM_TLS_BIND_C7=2`:  
+  - policy=legacy ≈ **51.5M ops/s**。TinyClassStats: C7 uc_miss=17196 / warm_hit=8597 / shared_lock=5 / tls_carve_attempt=8597 / success=8597（C5/C6 は uc_miss=1〜2）。  
+  - policy=auto（score=shared_lock*4+uc_miss）≈ **51.9M ops/s**（C7 固定 ON、C5/C6 はほぼ動かず）。  
+  - policy=c5_7_only ≈ **50.1M ops/s**。  
+- C7-only (size=1024, ws=32, iters=200K):  
+  - legacy: 平均 ≈ **46M ops/s**（Warm hit 3329 / tls_carve_success 3329 / shared_lock=5）。  
+  - auto: 平均 ≈ **44M ops/s**（統計ほぼ同じ、C7 固定 ON）。  
+- C7 guard vs full（Superslab 予算＋空スラブ限定）:  
+  - C7-only: full **42.4M ops/s** vs larson_guard **40.7M ops/s**（-4%）。  
+  - 129–1024B: full **49.0M ops/s** vs larson_guard **48.4M ops/s**（-1.2%）。  
+- C5/C6 固定サイズ (size=256≒C6, ws=512, iters=1M, stats dump ON):  
+  - policy=legacy ≈ **89.9M ops/s**（C6 uc_miss=5 / warm_hit=1 / tls_carve_success=1）。  
+  - policy=auto ≈ **87.5M ops/s**（統計ほぼ同じ、C5 はほぼゼロ）。  
+- WarmPool-STATS を TinyClassStats と統合。`HAKMEM_WARM_POOL_STATS=1` で C7-only 実行時に hits=3329 / misses=1 / prefilled=1 を確認（warm_hit と一致）。  
+- ログ抑制 ENV: `HAKMEM_TINY_POLICY_LOG` / `HAKMEM_TINY_WARM_LOG` / `HAKMEM_TINY_PAGEBOX_LOG` を 0 にすると長時間ランのノイズが減る（短時間の C7 デバッグ時だけ 1 にすると便利）。  
+- C7-only (mode=2) は Release/Debug ともに ~20M ops/s 帯（ログを多めに出すと 40M 付近まで振れる）。  
+- サイズ→クラス: `hak_tiny_size_to_class(size+1)` により 257–512B→C6、513–2048B→C7。512B も C7 が受ける設計で、実負荷の多くが C7 に集中する（C5/C6 は拡張枠）。  
+- mimalloc/system 比較（Release, `HAKMEM_TINY_PROFILE=full HAKMEM_TINY_POLICY_PROFILE=legacy HAKMEM_WARM_TLS_BIND_C7=2`, prefault=10% デフォルト, ログOFF）  
+  | workload (cycles/ws/size帯) | HAKMEM | mimalloc | system | 備考 |  
+  | --- | --- | --- | --- | --- |  
+  | C7-only (200K / 32 / 1024) | **48.8M ops/s** | 95.3M | 73.9M | mode=2, Warm+TLS carve |  
+  | Tiny-mixed 129–1024B (1M / 256) | **50.0M** | 128.4M | 97.7M | 513–2048B を C7 が受ける設計 |  
+  | full 16–1024B (1M / 256) | **50.9M** | 123.6M | 83.5M | デフォルト帯 |  
+  | Tiny-only 8–128B (200K / 400) | **93.2M** | 123.7M | 66.3M | Warm/TLS はほぼ踏まれず |  
+  現状ベスト: mimalloc が全帯域で最速。HAKMEM は C7 専用ワークロードで 50M 付近、Tiny-only では system より高速だが mimalloc には未到達。  
+**前回メモ (2025-12-05)**: C7 Warm/TLS Bind を Bind-only (mode=1) を本番経路とし、Release でも mode=2 を実験で有効化可能。C7-only で mode=1 が legacy (mode=0) 比で ~4–10x。  
+**Release 修復メモ (2025-12-05)**: 満杯 C7 slab が Shared Pool に残留していたため Warm が死んでいた。Acquire/Stage3 で空スラブ限定＆リセットを入れて C7-only Release ~23.7M ops/s → 20M+ 帯まで回復。  
+**Policy/OBSERVE/LEARN (2025-12-05)**: `TinyClassPolicyBox` 追加。`HAKMEM_TINY_POLICY_PROFILE=legacy|c5_7_only|tinyplus_all|auto` で Page/Warm を切替。OBSERVE では C7 がホットスポットで、`auto` プロファイルは C7 固定ON + score 上位2クラス（C5/C6 など）を自動で Tiny-Plus に昇格させる。
 
 **分析実施日**: 2025-11-28  
 **分析対象**: HAKMEM allocator (commit 0ce20bb83)  

bench_random_mixed.c

@@ -15,10 +15,7 @@
 #include <string.h>
 #include <strings.h>
 #include <stdatomic.h>
-#define C7_META_COUNTER_DEFINE
+#include <sys/resource.h>
-#include "core/box/c7_meta_used_counter_box.h"
-#undef C7_META_COUNTER_DEFINE
-#include "core/box/warm_pool_rel_counters_box.h"
 
 #ifdef USE_HAKMEM
 #include "hakmem.h"
@@ -26,6 +23,9 @@
 #include "core/box/c7_meta_used_counter_box.h"
 #include "core/box/tiny_class_stats_box.h"
 #include "core/box/tiny_class_policy_box.h"
+#include "core/box/ss_stats_box.h"
+#include "core/box/warm_pool_rel_counters_box.h"
+#include "core/box/tiny_mem_stats_box.h"
 
 // Box BenchMeta: Benchmark metadata management (bypass hakmem wrapper)
 // Phase 15: Separate BenchMeta (slots array) from CoreAlloc (user workload)
@@ -61,10 +61,30 @@ static inline int bench_is_c7_only_mode(void) {
   return bench_mode_c7_only;
 }
 
+// C5/C6 専用ベンチモード (ENV: HAKMEM_BENCH_C5_ONLY / HAKMEM_BENCH_C6_ONLY)
+static int bench_mode_c5_only = -1;
+static int bench_mode_c6_only = -1;
+static inline int bench_is_c5_only_mode(void) {
+  if (bench_mode_c5_only == -1) {
+    const char* e = getenv("HAKMEM_BENCH_C5_ONLY");
+    bench_mode_c5_only = (e && *e && *e != '0') ? 1 : 0;
+  }
+  return bench_mode_c5_only;
+}
+static inline int bench_is_c6_only_mode(void) {
+  if (bench_mode_c6_only == -1) {
+    const char* e = getenv("HAKMEM_BENCH_C6_ONLY");
+    bench_mode_c6_only = (e && *e && *e != '0') ? 1 : 0;
+  }
+  return bench_mode_c6_only;
+}
+
 int main(int argc, char** argv){
   int cycles = (argc>1)? atoi(argv[1]) : 10000000; // total ops (10M for steady-state measurement)
   int ws     = (argc>2)? atoi(argv[2]) : 8192;     // working-set slots
   uint32_t seed = (argc>3)? (uint32_t)strtoul(argv[3],NULL,10) : 1234567u;
+  struct rusage ru0 = {0}, ru1 = {0};
+  getrusage(RUSAGE_SELF, &ru0);
 
   // サイズレンジ（Tiny-only / Non-Tiny-only の比較用）
   // 既定: 16..1040 bytes（元の挙動と同等）
@@ -97,8 +117,14 @@ int main(int argc, char** argv){
   if (min_size < 1) min_size = 1;
   if (max_size < min_size) max_size = min_size;
 
-  // C7 専用モード: サイズを C7 帯に固定（現行 C7 ブロックサイズ ≈ 1024B）
+  // C5/C6/C7 専用モード: サイズを各クラス帯に固定
-  if (bench_is_c7_only_mode()) {
+  if (bench_is_c5_only_mode()) {
+    min_size = 256;
+    max_size = 256;
+  } else if (bench_is_c6_only_mode()) {
+    min_size = 512;
+    max_size = 512;
+  } else if (bench_is_c7_only_mode()) {
     min_size = 1024;
     max_size = 1024;
   }
@@ -238,10 +264,13 @@ int main(int argc, char** argv){
   for (int i=0;i<ws;i++){ if (slots[i]) { free(slots[i]); slots[i]=NULL; } }
   fprintf(stderr, "[TEST] Drain phase completed.\n");
   uint64_t end = now_ns();
+  getrusage(RUSAGE_SELF, &ru1);
   double sec = (double)(end-start)/1e9;
   double tput = (double)cycles / (sec>0.0?sec:1e-9);
   // Include params in output to avoid confusion about test conditions
   printf("Throughput = %9.0f ops/s [iter=%d ws=%d] time=%.3fs\n", tput, cycles, ws, sec);
+  long rss_kb = ru1.ru_maxrss;
+  fprintf(stderr, "[RSS] max_kb=%ld\n", rss_kb);
   (void)allocs; (void)frees;
 
   // Box BenchMeta: Use __libc_free to bypass hakmem wrapper
@@ -270,6 +299,14 @@ int main(int argc, char** argv){
     tiny_class_stats_dump_global(stderr, "[CLASS_STATS_GLOBAL]");
   }
 
+  const char* tiny_mem_dump_env = getenv("HAKMEM_TINY_MEM_DUMP");
+  if (tiny_mem_dump_env && *tiny_mem_dump_env && *tiny_mem_dump_env != '0') {
+    tiny_mem_stats_dump();
+  }
+
+  // Superslab/slab counters (ENV: HAKMEM_SS_STATS_DUMP=1)
+  ss_stats_dump_if_requested();
+
   // Warm Pool Stats (ENV-gated: HAKMEM_WARM_POOL_STATS=1)
   extern void tiny_warm_pool_print_stats_public(void);
   tiny_warm_pool_print_stats_public();

core/box/c7_hotpath_env_box.h

@@ -0,0 +1,15 @@
+// c7_hotpath_env_box.h - ENV gate for C7 hotpath
+// Purpose: isolate the ENV handling so hotpath code can assume gate済み。
+#pragma once
+
+#include <stdlib.h>
+
+// ENV gate: HAKMEM_TINY_C7_HOT=1 で有効化（デフォルト OFF）
+static inline int tiny_c7_hot_enabled(void) {
+    static int g_enable = -1;
+    if (__builtin_expect(g_enable == -1, 0)) {
+        const char* e = getenv("HAKMEM_TINY_C7_HOT");
+        g_enable = (e && *e && *e != '0') ? 1 : 0;
+    }
+    return g_enable;
+}

core/box/c7_meta_used_counter_box.c

@@ -0,0 +1,8 @@
+// c7_meta_used_counter_box.c
+// Definitions for C7 meta->used increment counters (Release/Debug共通)
+#include "c7_meta_used_counter_box.h"
+
+_Atomic uint64_t g_c7_meta_used_inc_total = 0;
+_Atomic uint64_t g_c7_meta_used_inc_backend = 0;
+_Atomic uint64_t g_c7_meta_used_inc_tls = 0;
+_Atomic uint64_t g_c7_meta_used_inc_front = 0;

core/box/carve_push_box.d

@@ -17,8 +17,9 @@ core/box/carve_push_box.o: core/box/carve_push_box.c \
  core/box/../tiny_region_id.h core/box/../tiny_box_geometry.h \
  core/box/../ptr_track.h core/box/../hakmem_super_registry.h \
  core/box/../box/ss_addr_map_box.h \
- core/box/../box/../hakmem_build_flags.h core/box/../tiny_debug_api.h \
+ core/box/../box/../hakmem_build_flags.h core/box/../box/super_reg_box.h \
- core/box/carve_push_box.h core/box/capacity_box.h core/box/tls_sll_box.h \
+ core/box/../tiny_debug_api.h core/box/carve_push_box.h \
+ core/box/capacity_box.h core/box/tls_sll_box.h \
  core/box/../hakmem_internal.h core/box/../hakmem.h \
  core/box/../hakmem_config.h core/box/../hakmem_features.h \
  core/box/../hakmem_sys.h core/box/../hakmem_whale.h \
@@ -70,6 +71,7 @@ core/box/../ptr_track.h:
 core/box/../hakmem_super_registry.h:
 core/box/../box/ss_addr_map_box.h:
 core/box/../box/../hakmem_build_flags.h:
+core/box/../box/super_reg_box.h:
 core/box/../tiny_debug_api.h:
 core/box/carve_push_box.h:
 core/box/capacity_box.h:

core/box/front_gate_box.d

@@ -11,20 +11,21 @@ core/box/front_gate_box.o: core/box/front_gate_box.c \
  core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
  core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
  core/tiny_debug_ring.h core/tiny_remote.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/tiny_debug_api.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/box/tiny_layout_box.h core/box/../hakmem_tiny_config.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
- core/box/tiny_header_box.h core/box/tiny_layout_box.h \
+ core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
- core/box/../tiny_region_id.h core/box/tls_sll_box.h \
+ core/box/tiny_layout_box.h core/box/../tiny_region_id.h \
- core/box/../hakmem_internal.h core/box/../hakmem.h \
+ core/box/tls_sll_box.h core/box/../hakmem_internal.h \
- core/box/../hakmem_build_flags.h core/box/../hakmem_config.h \
+ core/box/../hakmem.h core/box/../hakmem_build_flags.h \
- core/box/../hakmem_features.h core/box/../hakmem_sys.h \
+ core/box/../hakmem_config.h core/box/../hakmem_features.h \
- core/box/../hakmem_whale.h core/box/../box/ptr_type_box.h \
+ core/box/../hakmem_sys.h core/box/../hakmem_whale.h \
- core/box/../hakmem_debug_master.h core/box/../tiny_remote.h \
+ core/box/../box/ptr_type_box.h core/box/../hakmem_debug_master.h \
- core/box/../hakmem_tiny_integrity.h core/box/../hakmem_tiny.h \
+ core/box/../tiny_remote.h core/box/../hakmem_tiny_integrity.h \
- core/box/../ptr_track.h core/box/../ptr_trace.h \
+ core/box/../hakmem_tiny.h core/box/../ptr_track.h \
- core/box/../hakmem_trace_master.h core/box/../hakmem_stats_master.h \
+ core/box/../ptr_trace.h core/box/../hakmem_trace_master.h \
- core/box/../tiny_debug_ring.h core/box/ss_addr_map_box.h \
+ core/box/../hakmem_stats_master.h core/box/../tiny_debug_ring.h \
- core/box/../superslab/superslab_inline.h core/box/tiny_ptr_bridge_box.h \
+ core/box/ss_addr_map_box.h core/box/../superslab/superslab_inline.h \
+ core/box/tiny_ptr_bridge_box.h \
  core/box/../hakmem_tiny_superslab_internal.h \
  core/box/../hakmem_tiny_superslab.h core/box/../box/ss_hot_cold_box.h \
  core/box/../box/../superslab/superslab_types.h \
@@ -63,6 +64,7 @@ core/tiny_debug_ring.h:
 core/tiny_remote.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/tiny_debug_api.h:
 core/box/tiny_layout_box.h:
 core/box/../hakmem_tiny_config.h:

core/box/front_gate_classifier.d

@@ -11,8 +11,9 @@ core/box/front_gate_classifier.o: core/box/front_gate_classifier.c \
  core/box/../superslab/../tiny_box_geometry.h \
  core/box/../tiny_debug_ring.h core/box/../tiny_remote.h \
  core/box/../box/ss_addr_map_box.h \
- core/box/../box/../hakmem_build_flags.h core/box/../hakmem_tiny.h \
+ core/box/../box/../hakmem_build_flags.h core/box/../box/super_reg_box.h \
- core/box/../hakmem_trace.h core/box/../hakmem_tiny_mini_mag.h \
+ core/box/../hakmem_tiny.h core/box/../hakmem_trace.h \
+ core/box/../hakmem_tiny_mini_mag.h \
  core/box/../box/hak_lane_classify.inc.h core/box/../box/ptr_type_box.h \
  core/box/../tiny_debug_api.h core/box/../hakmem_tiny_superslab.h \
  core/box/../superslab/superslab_inline.h \
@@ -38,6 +39,7 @@ core/box/../tiny_debug_ring.h:
 core/box/../tiny_remote.h:
 core/box/../box/ss_addr_map_box.h:
 core/box/../box/../hakmem_build_flags.h:
+core/box/../box/super_reg_box.h:
 core/box/../hakmem_tiny.h:
 core/box/../hakmem_trace.h:
 core/box/../hakmem_tiny_mini_mag.h:

core/box/remote_side_box.c

@@ -0,0 +1,88 @@
+#include "remote_side_box.h"
+
+#include <stdatomic.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifndef REM_SIDE_LOG2
+#define REM_SIDE_LOG2 20
+#endif
+
+static _Atomic uint32_t g_remote_log2 = REM_SIDE_LOG2;
+static _Atomic uint32_t g_remote_size = (1u << REM_SIDE_LOG2);
+static _Atomic uint32_t g_remote_mask = (1u << REM_SIDE_LOG2) - 1;
+static _Atomic int g_remote_profile_inited = 0;
+static rem_side_entry* g_remote_slots = NULL;
+static _Atomic int g_remote_allocated = 0;
+
+static void remote_side_apply_profile(const char* profile) {
+    if (g_remote_profile_inited) {
+        return;
+    }
+    const char* env_profile = profile ? profile : getenv("HAKMEM_PROFILE");
+    int is_bench = (env_profile && strcmp(env_profile, "bench") == 0);
+
+    uint32_t log2 = REM_SIDE_LOG2;
+    if (is_bench && REM_SIDE_LOG2 > 4) {
+        // bench 用: ハッシュ幅だけ 1/8〜1/16 程度に論理縮小
+        log2 = REM_SIDE_LOG2 - 3;  // 1/8
+        if (log2 < 12) {
+            log2 = 12;  // 4096 entries までは確保
+        }
+    }
+
+    uint32_t size = (1u << log2);
+    uint32_t mask = size - 1;
+
+    atomic_store_explicit(&g_remote_log2, log2, memory_order_relaxed);
+    atomic_store_explicit(&g_remote_size, size, memory_order_relaxed);
+    atomic_store_explicit(&g_remote_mask, mask, memory_order_relaxed);
+    atomic_store_explicit(&g_remote_profile_inited, 1, memory_order_release);
+}
+
+void remote_side_init(RemoteSideBox* box, const char* profile) {
+    (void)box;
+    remote_side_apply_profile(profile);
+
+    if (atomic_load_explicit(&g_remote_allocated, memory_order_acquire)) {
+        return;
+    }
+
+    uint32_t size = remote_side_effective_size();
+    g_remote_slots = (rem_side_entry*)calloc(size, sizeof(rem_side_entry));
+    if (!g_remote_slots) {
+        fprintf(stderr, "[REMOTE_SIDE] failed to allocate %zu bytes\n",
+                (size_t)size * sizeof(rem_side_entry));
+        abort();
+    }
+    atomic_store_explicit(&g_remote_allocated, 1, memory_order_release);
+}
+
+uint32_t remote_side_effective_log2(void) {
+    if (!atomic_load_explicit(&g_remote_profile_inited, memory_order_acquire)) {
+        remote_side_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_remote_log2, memory_order_relaxed);
+}
+
+uint32_t remote_side_effective_size(void) {
+    if (!atomic_load_explicit(&g_remote_profile_inited, memory_order_acquire)) {
+        remote_side_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_remote_size, memory_order_relaxed);
+}
+
+uint32_t remote_side_effective_mask(void) {
+    if (!atomic_load_explicit(&g_remote_profile_inited, memory_order_acquire)) {
+        remote_side_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_remote_mask, memory_order_relaxed);
+}
+
+rem_side_entry* remote_side_table(void) {
+    if (!atomic_load_explicit(&g_remote_allocated, memory_order_acquire)) {
+        remote_side_init(NULL, NULL);
+    }
+    return g_remote_slots;
+}

core/box/remote_side_box.h

@@ -0,0 +1,21 @@
+#pragma once
+// RemoteSideBox: tiny_remote の REM_SIDE をプロファイルで論理的に絞るための薄いラッパ
+
+#include <stdint.h>
+#include <stdatomic.h>
+
+typedef struct rem_side_entry {
+    _Atomic(uintptr_t) key;  // node pointer
+    _Atomic(uintptr_t) val;  // next pointer
+} rem_side_entry;
+
+typedef struct RemoteSideBox RemoteSideBox;
+
+// profile が NULL のときは HAKMEM_PROFILE を見る。
+void remote_side_init(RemoteSideBox* box, const char* profile);
+
+// 有効サイズ/マスク（配列自体は REM_SIDE_SIZE のまま）
+uint32_t remote_side_effective_size(void);
+uint32_t remote_side_effective_mask(void);
+uint32_t remote_side_effective_log2(void);
+rem_side_entry* remote_side_table(void);

core/box/shared_pool_box.c

@@ -0,0 +1,50 @@
+#include "shared_pool_box.h"
+
+#include <stdatomic.h>
+#include <stdlib.h>
+#include <string.h>
+
+// 既存の g_shared_pool 配列上に「論理的な上限」だけを被せる。
+static _Atomic uint32_t g_sp_total_limit = 0;       // 0 = 無制限（現行のまま）
+static _Atomic uint32_t g_sp_class_limit = 0;       // 0 = 無制限
+static _Atomic int g_sp_profile_inited = 0;
+
+static void shared_pool_apply_profile(const char* profile) {
+    if (g_sp_profile_inited) {
+        return;
+    }
+    const char* env_profile = profile ? profile : getenv("HAKMEM_PROFILE");
+    int is_bench = (env_profile && strcmp(env_profile, "bench") == 0);
+
+    uint32_t total_limit = 0;
+    uint32_t class_limit = 0;
+    if (is_bench) {
+        // bench 用: ひとまず控えめな論理上限だけ入れる
+        total_limit = 65536;  // 元の 1M よりかなり少ない
+        class_limit = 2048;   // クラスあたりの active slot 上限の目安
+    }
+
+    atomic_store_explicit(&g_sp_total_limit, total_limit, memory_order_relaxed);
+    atomic_store_explicit(&g_sp_class_limit, class_limit, memory_order_relaxed);
+    atomic_store_explicit(&g_sp_profile_inited, 1, memory_order_release);
+}
+
+void shared_pool_box_init(SharedPoolBox* box, const char* profile) {
+    (void)box;
+    shared_pool_apply_profile(profile);
+}
+
+uint32_t shared_pool_effective_total_slots(void) {
+    if (!atomic_load_explicit(&g_sp_profile_inited, memory_order_acquire)) {
+        shared_pool_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_sp_total_limit, memory_order_relaxed);
+}
+
+uint32_t shared_pool_effective_class_slots(int class_idx) {
+    (void)class_idx;
+    if (!atomic_load_explicit(&g_sp_profile_inited, memory_order_acquire)) {
+        shared_pool_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_sp_class_limit, memory_order_relaxed);
+}

core/box/shared_pool_box.h

@@ -0,0 +1,18 @@
+#pragma once
+// SharedPoolBox: 既存の g_shared_pool の上に「論理上限」を被せる軽量ラッパ。
+// 目的:
+//   - HAKMEM_PROFILE=bench などのときに Shared Pool の増殖を論理的に抑える。
+//   - 配列サイズ自体は現状のまま（BSS をまだ縮めない）。
+
+#include <stdint.h>
+
+typedef struct SharedPoolBox SharedPoolBox;
+
+// profile が NULL のときは HAKMEM_PROFILE を読む。
+void shared_pool_box_init(SharedPoolBox* box, const char* profile);
+
+// これ以上増やさない総枠。full では元の制限なし、bench では小さめ。
+uint32_t shared_pool_effective_total_slots(void);
+
+// クラス別の論理上限（active slots がこの値を超えたら新規追加を抑制）
+uint32_t shared_pool_effective_class_slots(int class_idx);

core/box/ss_ace_box.c

@@ -175,8 +175,12 @@ static void ace_observe_and_decide(int k) {
     int ss_count = 0;
     uint32_t total_live = 0;
 
-    for (int i = 0; i < SUPER_REG_SIZE; i++) {
+    SuperRegEntry* reg = super_reg_entries();
-        SuperRegEntry* e = &g_super_reg[i];
+    int reg_cap = super_reg_effective_size();
+    if (!reg || reg_cap <= 0) return;
+
+    for (int i = 0; i < reg_cap; i++) {
+        SuperRegEntry* e = &reg[i];
 
         // Atomic read (thread-safe)
         uintptr_t base = atomic_load_explicit(

core/box/ss_allocation_box.c

@@ -284,6 +284,10 @@ SuperSlab* superslab_allocate(uint8_t size_class) {
         }
     } while (0);
 
+    if (!from_cache) {
+        ss_stats_on_ss_alloc_class(size_class);
+    }
+
     return ss;
 }
 

core/box/ss_budget_box.c

@@ -0,0 +1,122 @@
+// ss_budget_box.c - Superslab Budget Box
+// Box Theory: Budget/limit guard for Superslab growth.
+// - ENV:
+//     HAKMEM_SS_BUDGET_GLOBAL : global cap (0 = unlimited, default varies)
+//     HAKMEM_SS_BUDGET_C0..C7 : per-class cap override (0 = unlimited)
+//     HAKMEM_SS_BUDGET_C7     : shorthand most often used
+// - Profile hint:
+//     HAKMEM_TINY_PROFILE=larson_guard → stricter defaults.
+
+#include "ss_budget_box.h"
+
+#include <stdatomic.h>
+#include <stdlib.h>
+#include <strings.h>
+#include <stdio.h>
+
+#include "ss_stats_box.h"
+
+static _Atomic int g_budget_init = 0;
+static int g_ss_budget_global = 0;
+static int g_ss_budget_per_class[8] = {0};
+
+static int ss_budget_parse_env(const char* name, int fallback) {
+    const char* e = getenv(name);
+    if (e && *e) {
+        int v = atoi(e);
+        if (v < 0) v = 0;
+        return v;
+    }
+    return fallback;
+}
+
+static void ss_budget_init_once(void) {
+    if (atomic_load_explicit(&g_budget_init, memory_order_acquire)) {
+        return;
+    }
+
+    // Profile hint: larson_guard uses tighter defaults to cap RSS.
+    const char* profile = getenv("HAKMEM_TINY_PROFILE");
+    int is_larson_guard = (profile && strcasecmp(profile, "larson_guard") == 0);
+
+    // Defaults: unlimited unless larson_guard
+    int default_global = is_larson_guard ? 512 : 0;
+    g_ss_budget_global = ss_budget_parse_env("HAKMEM_SS_BUDGET_GLOBAL", default_global);
+
+    for (int i = 0; i < 8; i++) {
+        int def = 0;
+        if (is_larson_guard) {
+            // Larson guard: modest per-class caps, C7 is a bit looser.
+            def = (i == 7) ? 192 : 96;
+        }
+        g_ss_budget_per_class[i] = def;
+    }
+
+    // Per-class overrides: HAKMEM_SS_BUDGET_C7 or HAKMEM_SS_BUDGET_C{idx}
+    for (int i = 0; i < 8; i++) {
+        char buf[32];
+        snprintf(buf, sizeof(buf), "HAKMEM_SS_BUDGET_C%d", i);
+        int override = ss_budget_parse_env(buf, g_ss_budget_per_class[i]);
+        g_ss_budget_per_class[i] = override;
+    }
+    // Support the legacy shorthand HAKMEM_SS_BUDGET_C7
+    g_ss_budget_per_class[7] =
+        ss_budget_parse_env("HAKMEM_SS_BUDGET_C7", g_ss_budget_per_class[7]);
+
+    atomic_store_explicit(&g_budget_init, 1, memory_order_release);
+}
+
+static inline uint64_t ss_budget_global_live_sum(void) {
+    uint64_t sum = 0;
+    for (int i = 0; i < 8; i++) {
+        sum += atomic_load_explicit(&g_ss_live_by_class[i], memory_order_relaxed);
+    }
+    return sum;
+}
+
+bool ss_budget_on_alloc(int class_idx) {
+    ss_budget_init_once();
+
+    if (class_idx < 0 || class_idx >= 8) {
+        return true;  // outside Tiny; do not gate here
+    }
+
+    uint64_t live_cls = atomic_load_explicit(&g_ss_live_by_class[class_idx],
+                                             memory_order_relaxed);
+    int class_cap = g_ss_budget_per_class[class_idx];
+    if (class_cap > 0 && live_cls >= (uint64_t)class_cap) {
+        static _Atomic uint32_t log_once = 0;
+        if (atomic_fetch_add_explicit(&log_once, 1, memory_order_relaxed) < 4) {
+            fprintf(stderr,
+                    "[SS_BUDGET_DENY] class=%d live=%llu cap=%d\n",
+                    class_idx,
+                    (unsigned long long)live_cls,
+                    class_cap);
+        }
+        return false;
+    }
+
+    int global_cap = g_ss_budget_global;
+    if (global_cap > 0) {
+        uint64_t live_total = ss_budget_global_live_sum();
+        if (live_total >= (uint64_t)global_cap) {
+            static _Atomic uint32_t g_log_once = 0;
+            if (atomic_fetch_add_explicit(&g_log_once, 1, memory_order_relaxed) < 4) {
+                fprintf(stderr,
+                        "[SS_BUDGET_DENY_GLOBAL] live_total=%llu cap=%d class=%d\n",
+                        (unsigned long long)live_total,
+                        global_cap,
+                        class_idx);
+            }
+            return false;
+        }
+    }
+
+    return true;
+}
+
+void ss_budget_on_free(int class_idx) {
+    (void)class_idx;
+    ss_budget_init_once();
+    // We currently rely on ss_stats_on_ss_free_class() to update live counters.
+}

core/box/ss_budget_box.h

@@ -0,0 +1,19 @@
+// ss_budget_box.h - Superslab Budget Box
+// Box Theory: centralize budget/limit checks for Superslab allocations.
+// Responsibilities:
+//   - Read budget ENV once (global + per-class override)
+//   - Provide cheap checks before allocating new Superslabs
+//   - Allow symmetric free hook for future accounting
+
+#ifndef HAKMEM_SS_BUDGET_BOX_H
+#define HAKMEM_SS_BUDGET_BOX_H
+
+#include <stdbool.h>
+
+// Return false when allocation should be denied due to budget exhaustion.
+bool ss_budget_on_alloc(int class_idx);
+
+// Hook for future bookkeeping; currently a no-op placeholder.
+void ss_budget_on_free(int class_idx);
+
+#endif // HAKMEM_SS_BUDGET_BOX_H

core/box/ss_slab_reset_box.h

@@ -13,12 +13,15 @@ static inline void ss_slab_reset_meta_for_tiny(SuperSlab* ss,
     if (!ss) return;
     if (slab_idx < 0 || slab_idx >= ss_slabs_capacity(ss)) return;
 
+    // class_idx < 0 means "unassigned" (255). Otherwise keep the requested class.
+    uint8_t target_class = (class_idx < 0) ? 255u : (uint8_t)class_idx;
+
     TinySlabMeta* meta = &ss->slabs[slab_idx];
     meta->used = 0;
     meta->carved = 0;
     meta->freelist = NULL;
-    meta->class_idx = (uint8_t)class_idx;
+    meta->class_idx = target_class;
-    ss->class_map[slab_idx] = (uint8_t)class_idx;
+    ss->class_map[slab_idx] = target_class;
 
     // Reset remote queue state to avoid stale pending frees on reuse.
     atomic_store_explicit(&ss->remote_heads[slab_idx], 0, memory_order_relaxed);

core/box/ss_stats_box.c

@@ -1,8 +1,10 @@
 // ss_stats_box.c - SuperSlab Statistics Box Implementation
 #include "ss_stats_box.h"
+#include <stdbool.h>
 #include "../superslab/superslab_inline.h"
 #include <pthread.h>
 #include <stdio.h>
+#include <stdlib.h>
 
 // ============================================================================
 // Global Statistics State
@@ -30,6 +32,11 @@ _Atomic uint64_t g_free_ss_enter = 0;          // hak_tiny_free_superslab() entr
 _Atomic uint64_t g_free_local_box_calls = 0;   // same-thread freelist pushes
 _Atomic uint64_t g_free_remote_box_calls = 0;  // cross-thread remote pushes
 
+// Superslab/slab observability (Tiny-only; relaxed updates)
+_Atomic uint64_t g_ss_live_by_class[8]   = {0};
+_Atomic uint64_t g_ss_empty_events[8]    = {0};
+_Atomic uint64_t g_slab_live_events[8]   = {0};
+
 // ============================================================================
 // Statistics Update Implementation
 // ============================================================================
@@ -56,6 +63,36 @@ void ss_stats_cache_store(void) {
     pthread_mutex_unlock(&g_superslab_lock);
 }
 
+void ss_stats_on_ss_alloc_class(int class_idx) {
+    if (class_idx >= 0 && class_idx < 8) {
+        atomic_fetch_add_explicit(&g_ss_live_by_class[class_idx], 1, memory_order_relaxed);
+    }
+}
+
+void ss_stats_on_ss_free_class(int class_idx) {
+    if (class_idx >= 0 && class_idx < 8) {
+        // Saturating-style decrement to avoid underflow from mismatched hooks
+        uint64_t prev = atomic_load_explicit(&g_ss_live_by_class[class_idx], memory_order_relaxed);
+        if (prev > 0) {
+            atomic_fetch_sub_explicit(&g_ss_live_by_class[class_idx], 1, memory_order_relaxed);
+        }
+    }
+}
+
+void ss_stats_on_ss_scan(int class_idx, int slab_live, int is_empty) {
+    if (class_idx < 0 || class_idx >= 8) {
+        return;
+    }
+    if (slab_live > 0) {
+        atomic_fetch_add_explicit(&g_slab_live_events[class_idx],
+                                  (uint64_t)slab_live,
+                                  memory_order_relaxed);
+    }
+    if (is_empty) {
+        atomic_fetch_add_explicit(&g_ss_empty_events[class_idx], 1, memory_order_relaxed);
+    }
+}
+
 // ============================================================================
 // Statistics Reporting Implementation
 // ============================================================================
@@ -92,3 +129,23 @@ void superslab_print_global_stats(void) {
     printf("Total bytes allocated: %lu MB\n", g_bytes_allocated / (1024 * 1024));
     pthread_mutex_unlock(&g_superslab_lock);
 }
+
+void ss_stats_dump_if_requested(void) {
+    const char* env = getenv("HAKMEM_SS_STATS_DUMP");
+    if (!env || !*env || *env == '0') {
+        return;
+    }
+    fprintf(stderr, "[SS_STATS] class live empty_events slab_live_events\n");
+    for (int c = 0; c < 8; c++) {
+        uint64_t live = atomic_load_explicit(&g_ss_live_by_class[c], memory_order_relaxed);
+        uint64_t empty = atomic_load_explicit(&g_ss_empty_events[c], memory_order_relaxed);
+        uint64_t slab_live = atomic_load_explicit(&g_slab_live_events[c], memory_order_relaxed);
+        if (live || empty || slab_live) {
+            fprintf(stderr, "  C%d: live=%llu empty=%llu slab_live=%llu\n",
+                    c,
+                    (unsigned long long)live,
+                    (unsigned long long)empty,
+                    (unsigned long long)slab_live);
+        }
+    }
+}

core/box/ss_stats_box.h

@@ -43,6 +43,16 @@ extern _Atomic uint64_t g_free_ss_enter;
 extern _Atomic uint64_t g_free_local_box_calls;
 extern _Atomic uint64_t g_free_remote_box_calls;
 
+// ============================================================================
+// Superslab / Slab live-state observability (Tiny classes 0..7)
+// ============================================================================
+// NOTE: These are “event-style” counters updated at key transitions
+// (alloc/free/reset) to keep overhead minimal. They are intended for
+// regression detection and coarse budgeting rather than exact gauges.
+extern _Atomic uint64_t g_ss_live_by_class[8];   // +1 on alloc, -1 on free (best-effort)
+extern _Atomic uint64_t g_ss_empty_events[8];    // Observations of fully-empty Superslabs
+extern _Atomic uint64_t g_slab_live_events[8];   // Observations of live slabs during scans
+
 // ============================================================================
 // Statistics Update API
 // ============================================================================
@@ -59,6 +69,11 @@ void ss_stats_cache_reuse(void);
 // Thread-safe: mutex protected
 void ss_stats_cache_store(void);
 
+// Event-style observability helpers (Tiny classes only, relaxed atomics)
+void ss_stats_on_ss_alloc_class(int class_idx);
+void ss_stats_on_ss_free_class(int class_idx);
+void ss_stats_on_ss_scan(int class_idx, int slab_live, int is_empty);
+
 // ============================================================================
 // Statistics Reporting API
 // ============================================================================
@@ -69,4 +84,7 @@ void superslab_print_stats(SuperSlab* ss);
 // Print global SuperSlab statistics
 void superslab_print_global_stats(void);
 
+// ENV: HAKMEM_SS_STATS_DUMP=1 → dump coarse Superslab/slab counters once
+void ss_stats_dump_if_requested(void);
+
 #endif // HAKMEM_SS_STATS_BOX_H

core/box/ss_tls_bind_box.h

@@ -119,7 +119,7 @@ static inline int ss_tls_bind_one(int class_idx,
     tls->slab_base = tiny_slab_base_for(ss, slab_idx);
 
     // Notify Tiny Page Box (if enabled for this class)
-    tiny_page_box_on_new_slab(tls);
+    tiny_page_box_on_new_slab(class_idx, tls);
 
     // Sanity check: TLS must now describe this slab for this class.
     // On failure, revert TLS to safe state and return 0.

core/box/super_reg_box.c

@@ -0,0 +1,143 @@
+#include "super_reg_box.h"
+
+#include <stdatomic.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "hakmem_super_registry.h"
+
+// プロファイル別の実容量・論理上限
+static _Atomic int g_super_reg_effective_size = SUPER_REG_SIZE;
+static _Atomic int g_super_reg_effective_mask = SUPER_REG_MASK;
+static _Atomic int g_super_reg_effective_per_class = SUPER_REG_PER_CLASS;
+static _Atomic int g_super_reg_profile_inited = 0;
+
+// 動的に確保する実配列
+static SuperRegEntry* g_super_reg_entries = NULL;
+static SuperSlab** g_super_reg_by_class_slots = NULL;
+static int g_super_reg_by_class_stride = SUPER_REG_PER_CLASS;
+static _Atomic int g_super_reg_allocated = 0;
+
+static inline int super_reg_clamp_power_of_two(int requested, int fallback) {
+    // SUPER_REG_SIZE は 2 のべき乗なので、requested もそれ未満のべき乗に丸める。
+    if (requested <= 0 || requested > SUPER_REG_SIZE) {
+        return fallback;
+    }
+    // 丸め: 最上位ビットだけを残す（2 のべき乗に丸め下げ）
+    int v = requested;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    v = v - (v >> 1);
+    // 有効値は最低でも 1024 にしておく
+    if (v < 1024) {
+        v = 1024;
+    }
+    return v;
+}
+
+static void super_reg_apply_profile(const char* profile) {
+    if (g_super_reg_profile_inited) {
+        return;
+    }
+
+    const char* env_profile = profile ? profile : getenv("HAKMEM_PROFILE");
+    const int is_bench = (env_profile && strcmp(env_profile, "bench") == 0);
+
+    int eff_size = SUPER_REG_SIZE;
+    int eff_per_class = SUPER_REG_PER_CLASS;
+
+    if (is_bench) {
+        // 論理上の利用範囲だけ縮める（配列は従来サイズのまま）
+        eff_size = SUPER_REG_SIZE >> 3;          // 1/8 に論理制限
+        eff_per_class = SUPER_REG_PER_CLASS >> 4;  // 1/16
+    }
+
+    eff_size = super_reg_clamp_power_of_two(eff_size, SUPER_REG_SIZE);
+    eff_per_class = eff_per_class > 0 ? eff_per_class : SUPER_REG_PER_CLASS;
+
+    atomic_store_explicit(&g_super_reg_effective_size, eff_size, memory_order_relaxed);
+    atomic_store_explicit(&g_super_reg_effective_mask, eff_size - 1, memory_order_relaxed);
+    atomic_store_explicit(&g_super_reg_effective_per_class,
+                          eff_per_class,
+                          memory_order_relaxed);
+    atomic_store_explicit(&g_super_reg_profile_inited, 1, memory_order_release);
+}
+
+void super_reg_init(SuperRegBox* box, const char* profile) {
+    (void)box;
+    super_reg_apply_profile(profile);
+
+    if (atomic_load_explicit(&g_super_reg_allocated, memory_order_acquire)) {
+        return;
+    }
+
+    int eff_size = super_reg_effective_size();
+    int per_class = super_reg_effective_per_class();
+
+    // Allocate registry table
+    size_t reg_bytes = (size_t)eff_size * sizeof(SuperRegEntry);
+    g_super_reg_entries = (SuperRegEntry*)calloc(eff_size, sizeof(SuperRegEntry));
+    if (!g_super_reg_entries) {
+        fprintf(stderr, "[SUPER_REG] failed to allocate %zu bytes for registry\n", reg_bytes);
+        abort();
+    }
+
+    // Allocate per-class table (contiguous 1D block)
+    size_t per_class_bytes = (size_t)TINY_NUM_CLASSES * (size_t)per_class * sizeof(SuperSlab*);
+    g_super_reg_by_class_slots = (SuperSlab**)calloc(TINY_NUM_CLASSES * (size_t)per_class,
+                                                     sizeof(SuperSlab*));
+    if (!g_super_reg_by_class_slots) {
+        fprintf(stderr, "[SUPER_REG] failed to allocate %zu bytes for per-class registry\n",
+                per_class_bytes);
+        abort();
+    }
+    g_super_reg_by_class_stride = per_class;
+
+    atomic_store_explicit(&g_super_reg_allocated, 1, memory_order_release);
+}
+
+int super_reg_effective_size(void) {
+    if (!atomic_load_explicit(&g_super_reg_profile_inited, memory_order_acquire)) {
+        super_reg_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_super_reg_effective_size, memory_order_relaxed);
+}
+
+int super_reg_effective_mask(void) {
+    if (!atomic_load_explicit(&g_super_reg_profile_inited, memory_order_acquire)) {
+        super_reg_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_super_reg_effective_mask, memory_order_relaxed);
+}
+
+int super_reg_effective_per_class(void) {
+    if (!atomic_load_explicit(&g_super_reg_profile_inited, memory_order_acquire)) {
+        super_reg_apply_profile(NULL);
+    }
+    return atomic_load_explicit(&g_super_reg_effective_per_class, memory_order_relaxed);
+}
+
+SuperRegEntry* super_reg_entries(void) {
+    if (!atomic_load_explicit(&g_super_reg_allocated, memory_order_acquire)) {
+        super_reg_init(NULL, NULL);
+    }
+    return g_super_reg_entries;
+}
+
+SuperSlab** super_reg_by_class_slots(void) {
+    if (!atomic_load_explicit(&g_super_reg_allocated, memory_order_acquire)) {
+        super_reg_init(NULL, NULL);
+    }
+    return g_super_reg_by_class_slots;
+}
+
+int super_reg_by_class_stride(void) {
+    if (!atomic_load_explicit(&g_super_reg_allocated, memory_order_acquire)) {
+        super_reg_init(NULL, NULL);
+    }
+    return g_super_reg_by_class_stride;
+}

core/box/super_reg_box.h

@@ -0,0 +1,77 @@
+#pragma once
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#ifndef TINY_NUM_CLASSES
+#define TINY_NUM_CLASSES 8
+#endif
+// SuperRegBox (設計メモ / API スタブ)
+// -------------------------------------
+// 役割:
+//   - g_super_reg / g_super_reg_by_class への直接依存を断ち、レジストリ容量を
+//     プロファイル（full/prod/bench/larson_guard 等）で切り替えられるようにする箱。
+//   - Box 内部だけで容量決定・確保・破棄を閉じ、外側は薄い API を呼ぶだけにする。
+//
+// プロファイル方針（案）:
+//   - full/prod : 現行の SUPER_REG_SIZE (=1,048,576) と SUPER_REG_PER_CLASS (=16,384) を維持
+//   - bench     : SUPER_REG_SIZE を 1/16〜1/8 程度 (例: 65,536)、per-class は 1,024 などに縮小
+//   - guard     : bench 同等かさらに小さくして fail-fast（ENOMEM）を優先
+//
+// スレッド安全性:
+//   - 既存のロック/atomic 公算を流用しつつ、構造体にまとめて「初期化済みか」を判定。
+//
+// 想定 API（実装は今後）:
+
+typedef struct SuperSlab SuperSlab;
+typedef struct SuperRegBox SuperRegBox;
+struct SuperRegEntry;
+
+// プロファイル/ENV に応じて容量を決定し、内部配列を確保。
+// profile が NULL のときは HAKMEM_PROFILE (bench / full など) を読む。
+void super_reg_init(SuperRegBox* box, const char* profile);
+
+// 現在有効なスロット数/マスク
+int super_reg_effective_size(void);
+int super_reg_effective_mask(void);
+int super_reg_effective_per_class(void);
+
+// レジストリ実体へのアクセス（Box 内部で動的確保）
+struct SuperRegEntry* super_reg_entries(void);
+SuperSlab** super_reg_by_class_slots(void);
+int super_reg_by_class_stride(void);
+static inline SuperSlab* super_reg_by_class_at(int class_idx, int idx) {
+    SuperSlab** slots = super_reg_by_class_slots();
+    int stride = super_reg_by_class_stride();
+    if (!slots || stride <= 0 || class_idx < 0 || idx < 0 ||
+        class_idx >= TINY_NUM_CLASSES || idx >= stride) {
+        return NULL;
+    }
+    return slots[class_idx * stride + idx];
+}
+static inline void super_reg_by_class_set(int class_idx, int idx, SuperSlab* ss) {
+    SuperSlab** slots = super_reg_by_class_slots();
+    int stride = super_reg_by_class_stride();
+    if (!slots || stride <= 0 || class_idx < 0 || idx < 0 ||
+        class_idx >= TINY_NUM_CLASSES || idx >= stride) {
+        return;
+    }
+    slots[class_idx * stride + idx] = ss;
+}
+
+// Superslab 登録/解除（既存の hak_super_register/unregister 相当を箱内に閉じ込める）
+bool super_reg_register(SuperRegBox* box, SuperSlab* ss, uint32_t class_idx);
+void super_reg_unregister(SuperRegBox* box, SuperSlab* ss, uint32_t class_idx);
+
+// アドレス検索／クラス別イテレーション（必要最小限の薄い API）
+SuperSlab* super_reg_find_by_addr(SuperRegBox* box, void* ptr);
+SuperSlab* super_reg_iter_for_class(SuperRegBox* box, uint32_t class_idx, void** cursor);
+
+// 将来のメモリ削減策（コメントのみ）
+//  - g_super_reg/g_super_reg_by_class を「malloc/mmap でプロファイル毎に確保」するようにし、
+//    BSS から切り離す。
+//  - bench プロファイルでは固定長を大幅に縮め、足りなければ ENOMEM を返して fail-fast。
+//  - prod では現行サイズを維持しつつ、Box 境界でのみアクセスさせる。***
+
+// 前方宣言（実装は既存の superslab に依存）
+// typedef struct SuperSlab SuperSlab;  // 上で宣言済み

core/box/tiny_c7_hotpath_box.h

@@ -0,0 +1,63 @@
+// C7 専用の実験的ホットパス。HAKMEM_TINY_C7_HOT=1 でのみ有効化し、
+// デフォルト（未設定/0）のときは従来経路に完全フォールバックする。
+// 本番デフォルトで ON にしない前提の A/B 用スイッチ。
+#pragma once
+
+#include "../hakmem_build_flags.h"
+#include "c7_hotpath_env_box.h"
+#include "tiny_c7_uc_hit_box.h"
+#include "tiny_c7_warm_spill_box.h"
+#include "tiny_c7_stats_sample_box.h"
+#include "tiny_front_hot_box.h"
+#include "tiny_front_cold_box.h"
+#include "front_gate_box.h"
+#include "tls_sll_box.h"
+#include "ptr_conversion_box.h"
+
+// C7 alloc ホットパス。
+// 順序:
+//   1) TLS/SFC (front_gate_try_pop) を先に覗く
+//   2) Unified Cache のヒット専用パス tiny_uc_pop_c7_hit_only()
+//   3) それでもダメなら通常の cold refill（refill/統計は cold 側に任せる）
+static inline void* tiny_c7_alloc_hot(size_t size) {
+    (void)size;  // size は class_idx=7 前提なので未使用
+    void* user = NULL;
+
+    // 1) SFC/TLS SLL 直叩き（ユーザーポインタが返る）
+    if (front_gate_try_pop(/*class_idx=*/7, &user)) {
+        return user;
+    }
+
+    // 2) Unified Cache ヒット
+    user = tiny_uc_pop_c7_hit_only();
+    if (__builtin_expect(user != NULL, 1)) {
+        return user;
+    }
+
+    // 3) Cold refill へフォールバック
+    return tiny_cold_refill_and_alloc(7);
+}
+
+// C7 free ホットパス。BASE を受け取り TLS→UC の順に試す。
+static inline int tiny_c7_free_hot(void* base) {
+    // 1) TLS SLL へ直接 push（BASE のまま渡す）
+    extern int g_tls_sll_enable;
+    if (__builtin_expect(g_tls_sll_enable, 1)) {
+        if (tls_sll_push(7, HAK_BASE_FROM_RAW(base), UINT32_MAX)) {
+            return 1;
+        }
+    }
+
+    // 2) Unified Cache へ push（ヒット専用の軽量版）
+    if (tiny_uc_push_c7_hot(base)) {
+        return 1;
+    }
+
+    // 3) Warm spill（将来用のフック）
+    if (tiny_c7_warm_spill_one(base)) {
+        return 1;
+    }
+
+    // 4) 最後に cold free パスへフォールバック
+    return tiny_cold_drain_and_free(7, base);
+}

core/box/tiny_c7_stats_sample_box.h

@@ -0,0 +1,9 @@
+// tiny_c7_stats_sample_box.h - Lightweight sampling helper for C7 stats
+// 現状は簡易 1/16 サンプリング。hot path から #if を排除するための小箱。
+#pragma once
+
+static inline int tiny_c7_stats_sample(void) {
+    static __thread unsigned counter = 0;
+    counter++;
+    return (counter & 0xF) == 0;  // 約 1/16
+}

core/box/tiny_c7_uc_hit_box.h

@@ -0,0 +1,58 @@
+// tiny_c7_uc_hit_box.h - C7 専用 Unified Cache hit-only helpers
+// 契約: ヒット時のみ処理。ミス時は NULL/0 を返し、refill・統計は行わない。
+#pragma once
+
+#include "../front/tiny_unified_cache.h"
+#include "tiny_layout_box.h"
+
+// C7 UC ヒット専用 pop
+static inline void* tiny_uc_pop_c7_hit_only(void) {
+    TinyUnifiedCache* cache = &g_unified_cache[7];
+
+#if !HAKMEM_TINY_FRONT_PGO
+    if (__builtin_expect(cache->slots == NULL, 0)) {
+        unified_cache_init();
+        if (cache->slots == NULL) {
+            return NULL;
+        }
+    }
+#endif
+
+    if (__builtin_expect(cache->head == cache->tail, 0)) {
+        return NULL;
+    }
+
+    void* base = cache->slots[cache->head];
+    cache->head = (cache->head + 1) & cache->mask;
+
+#if HAKMEM_TINY_HEADER_CLASSIDX
+    tiny_region_id_write_header(base, 7);
+    size_t user_offset = tiny_user_offset(7);
+    return (void*)((char*)base + user_offset);
+#else
+    return base;
+#endif
+}
+
+// C7 UC ヒット専用 push
+static inline int tiny_uc_push_c7_hot(void* base) {
+    TinyUnifiedCache* cache = &g_unified_cache[7];
+
+#if !HAKMEM_TINY_FRONT_PGO
+    if (__builtin_expect(cache->slots == NULL, 0)) {
+        unified_cache_init();
+        if (cache->slots == NULL) {
+            return 0;
+        }
+    }
+#endif
+
+    uint16_t next_tail = (cache->tail + 1) & cache->mask;
+    if (__builtin_expect(next_tail == cache->head, 0)) {
+        return 0;  // full
+    }
+
+    cache->slots[cache->tail] = base;
+    cache->tail = next_tail;
+    return 1;
+}

core/box/tiny_c7_warm_spill_box.h

@@ -0,0 +1,9 @@
+// tiny_c7_warm_spill_box.h - C7 Warm spill hook (placeholder)
+// Purpose: allow swapping spill実装 without touchingホットパス。
+#pragma once
+
+// いまは no-op。将来 Warm spill を挿すときに差し替える。
+static inline int tiny_c7_warm_spill_one(void* base) {
+    (void)base;
+    return 0;
+}

core/box/tiny_class_policy_box.c

@@ -6,17 +6,20 @@
 #include <string.h>
 #include <strings.h>
 #include "tiny_policy_learner_box.h"
+#include "tiny_mem_stats_box.h"
 
 TinyClassPolicy g_tiny_class_policy[TINY_NUM_CLASSES];
 static _Atomic int g_tiny_class_policy_init_done = 0;
 static _Atomic int g_tiny_class_policy_logged = 0;
 static _Atomic int g_tiny_class_policy_profile_auto = 0;
+static _Atomic int g_tiny_class_policy_mem_recorded = 0;
 
 static inline TinyClassPolicy tiny_class_policy_default_entry(void) {
     TinyClassPolicy p = {0};
     p.page_box_enabled = 0;
     p.warm_enabled = 0;
     p.warm_cap = 0;
+    p.tls_carve_enabled = 0;
     return p;
 }
 
@@ -30,6 +33,7 @@ static void tiny_class_policy_set_legacy(void) {
     for (int i = 0; i < TINY_NUM_CLASSES; i++) {
         g_tiny_class_policy[i].warm_enabled = 1;
         g_tiny_class_policy[i].warm_cap = (i < 5) ? 4 : 8;
+        g_tiny_class_policy[i].tls_carve_enabled = (i >= 5) ? 1 : 0;
     }
     for (int i = 5; i < TINY_NUM_CLASSES; i++) {
         g_tiny_class_policy[i].page_box_enabled = 1;
@@ -45,6 +49,7 @@ static void tiny_class_policy_set_c5_7_only(void) {
         g_tiny_class_policy[i].page_box_enabled = 1;
         g_tiny_class_policy[i].warm_enabled = 1;
         g_tiny_class_policy[i].warm_cap = 8;
+        g_tiny_class_policy[i].tls_carve_enabled = 1;
     }
 }
 
@@ -53,6 +58,18 @@ static void tiny_class_policy_set_tinyplus_all(void) {
     tiny_class_policy_set_legacy();
 }
 
+static void tiny_class_policy_set_larson_guard(void) {
+    // Start from legacy, then tighten warm caps to reduce RSS for larson-style loads.
+    tiny_class_policy_set_legacy();
+    for (int i = 0; i < TINY_NUM_CLASSES; i++) {
+        if (i < 5) {
+            g_tiny_class_policy[i].warm_cap = 2;
+        } else {
+            g_tiny_class_policy[i].warm_cap = 4;
+        }
+    }
+}
+
 static void tiny_class_policy_set_auto(void) {
     // auto プロファイルは legacy をベースにして、後段の learner に委譲
     tiny_class_policy_set_legacy();
@@ -72,6 +89,10 @@ static const char* tiny_class_policy_set_profile(const char* profile) {
         tiny_class_policy_set_tinyplus_all();
         atomic_store_explicit(&g_tiny_class_policy_profile_auto, 0, memory_order_release);
         return "tinyplus_all";
+    } else if (strcasecmp(profile, "larson_guard") == 0) {
+        tiny_class_policy_set_larson_guard();
+        atomic_store_explicit(&g_tiny_class_policy_profile_auto, 0, memory_order_release);
+        return "larson_guard";
     } else if (strcasecmp(profile, "auto") == 0) {
         tiny_class_policy_set_auto();
         return "auto";
@@ -84,16 +105,20 @@ static const char* tiny_class_policy_set_profile(const char* profile) {
 }
 
 void tiny_class_policy_dump(const char* tag) {
+    if (!tiny_policy_log_enabled()) {
+        return;
+    }
     const char* header = tag ? tag : "[POLICY_DUMP]";
     fprintf(stderr, "%s\n", header);
     for (int cls = 0; cls < TINY_NUM_CLASSES; cls++) {
         TinyClassPolicy* p = &g_tiny_class_policy[cls];
         fprintf(stderr,
-                "  C%d: page=%u warm=%u cap=%u\n",
+                "  C%d: page=%u warm=%u cap=%u tls_carve=%u\n",
                 cls,
                 p->page_box_enabled,
                 p->warm_enabled,
-                p->warm_cap);
+                p->warm_cap,
+                p->tls_carve_enabled);
     }
 }
 
@@ -105,8 +130,13 @@ void tiny_class_policy_init_once(void) {
     const char* profile = getenv("HAKMEM_TINY_POLICY_PROFILE");
     const char* active_profile = tiny_class_policy_set_profile(profile);
 
+    if (atomic_exchange_explicit(&g_tiny_class_policy_mem_recorded, 1, memory_order_acq_rel) == 0) {
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_policy));
+    }
+
     // 1-shot ダンプでポリシーの内容を可視化（デバッグ用）
-    if (atomic_exchange_explicit(&g_tiny_class_policy_logged, 1, memory_order_acq_rel) == 0) {
+    if (tiny_policy_log_enabled() &&
+        atomic_exchange_explicit(&g_tiny_class_policy_logged, 1, memory_order_acq_rel) == 0) {
         fprintf(stderr, "[POLICY_INIT] profile=%s\n", active_profile);
         tiny_class_policy_dump(NULL);
     }
@@ -121,3 +151,8 @@ void tiny_class_policy_refresh_auto(void) {
     }
     tiny_policy_learner_tick();
 }
+
+int tiny_class_policy_is_auto(void) {
+    tiny_class_policy_init_once();
+    return atomic_load_explicit(&g_tiny_class_policy_profile_auto, memory_order_acquire);
+}

core/box/tiny_class_policy_box.h

@@ -15,23 +15,37 @@
 
 #include <stdatomic.h>
 #include <stdint.h>
+#include <stdlib.h>
 #include "../hakmem_tiny_config.h"
 
 typedef struct TinyClassPolicy {
     uint8_t page_box_enabled;  // Enable Tiny Page Box for this class
     uint8_t warm_enabled;      // Enable Warm Pool for this class
     uint8_t warm_cap;          // Max warm SuperSlabs to keep (per-thread)
-    uint8_t reserved;
+    uint8_t tls_carve_enabled; // Enable Warm→TLS carve experiment for this class
 } TinyClassPolicy;
 
 extern TinyClassPolicy g_tiny_class_policy[TINY_NUM_CLASSES];
 
+// ENV-gated policy logging (default ON; disable with HAKMEM_TINY_POLICY_LOG=0)
+static inline int tiny_policy_log_enabled(void) {
+    static int g_policy_log = -1;
+    if (__builtin_expect(g_policy_log == -1, 0)) {
+        const char* e = getenv("HAKMEM_TINY_POLICY_LOG");
+        g_policy_log = (e && *e && *e != '0') ? 1 : 0;
+    }
+    return g_policy_log;
+}
+
 // Initialize policy table once (idempotent).
 void tiny_class_policy_init_once(void);
 
 // Refresh auto profile based on learner output (no-op for non-auto profiles)
 void tiny_class_policy_refresh_auto(void);
 
+// True when active profile is "auto" (learner-managed)
+int tiny_class_policy_is_auto(void);
+
 // Debug helper: dump current policy (tag optional)
 void tiny_class_policy_dump(const char* tag);
 

core/box/tiny_class_stats_box.c

@@ -1,6 +1,7 @@
 // tiny_class_stats_box.c - Thread-local stats storage for Tiny classes
 
 #include "tiny_class_stats_box.h"
+#include "tiny_mem_stats_box.h"
 #include <stdio.h>
 #include <string.h>
 
@@ -8,6 +9,20 @@ __thread TinyClassStatsThread g_tiny_class_stats = {0};
 _Atomic uint64_t g_tiny_class_stats_uc_miss_global[TINY_NUM_CLASSES] = {0};
 _Atomic uint64_t g_tiny_class_stats_warm_hit_global[TINY_NUM_CLASSES] = {0};
 _Atomic uint64_t g_tiny_class_stats_shared_lock_global[TINY_NUM_CLASSES] = {0};
+_Atomic uint64_t g_tiny_class_stats_tls_carve_attempt_global[TINY_NUM_CLASSES] = {0};
+_Atomic uint64_t g_tiny_class_stats_tls_carve_success_global[TINY_NUM_CLASSES] = {0};
+static _Atomic int g_tiny_class_stats_mem_recorded = 0;
+
+static void tiny_class_stats_record_mem_once(void) {
+    if (atomic_exchange_explicit(&g_tiny_class_stats_mem_recorded, 1, memory_order_acq_rel) == 0) {
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats));
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats_uc_miss_global));
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats_warm_hit_global));
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats_shared_lock_global));
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats_tls_carve_attempt_global));
+        tiny_mem_stats_add_policy_stats((ssize_t)sizeof(g_tiny_class_stats_tls_carve_success_global));
+    }
+}
 
 void tiny_class_stats_reset_thread(void) {
     memset(&g_tiny_class_stats, 0, sizeof(g_tiny_class_stats));
@@ -15,11 +30,13 @@ void tiny_class_stats_reset_thread(void) {
 
 void tiny_class_stats_snapshot_thread(TinyClassStatsThread* out) {
     if (!out) return;
+    tiny_class_stats_record_mem_once();
     memcpy(out, &g_tiny_class_stats, sizeof(*out));
 }
 
 void tiny_class_stats_snapshot_global(TinyClassStatsThread* out) {
     if (!out) return;
+    tiny_class_stats_record_mem_once();
     for (int i = 0; i < TINY_NUM_CLASSES; i++) {
         out->uc_miss[i] = atomic_load_explicit(&g_tiny_class_stats_uc_miss_global[i],
                                                memory_order_relaxed);
@@ -27,6 +44,10 @@ void tiny_class_stats_snapshot_global(TinyClassStatsThread* out) {
                                                 memory_order_relaxed);
         out->shared_lock[i] = atomic_load_explicit(&g_tiny_class_stats_shared_lock_global[i],
                                                    memory_order_relaxed);
+        out->tls_carve_attempt[i] = atomic_load_explicit(
+            &g_tiny_class_stats_tls_carve_attempt_global[i], memory_order_relaxed);
+        out->tls_carve_success[i] = atomic_load_explicit(
+            &g_tiny_class_stats_tls_carve_success_global[i], memory_order_relaxed);
     }
 }
 
@@ -34,14 +55,18 @@ static void tiny_class_stats_dump_common(FILE* out,
                                          const char* tag,
                                          const TinyClassStatsThread* stats) {
     if (!(out && stats)) return;
-    fprintf(out, "%s class uc_miss warm_hit shared_lock\n", tag ? tag : "[STATS]");
+    fprintf(out, "%s class uc_miss warm_hit shared_lock tls_carve_attempt tls_carve_success\n",
+            tag ? tag : "[STATS]");
     for (int c = 0; c < TINY_NUM_CLASSES; c++) {
-        if (stats->uc_miss[c] || stats->warm_hit[c] || stats->shared_lock[c]) {
+        if (stats->uc_miss[c] || stats->warm_hit[c] || stats->shared_lock[c] ||
-            fprintf(out, "  C%d: %llu %llu %llu\n",
+            stats->tls_carve_attempt[c] || stats->tls_carve_success[c]) {
+            fprintf(out, "  C%d: %llu %llu %llu %llu %llu\n",
                     c,
                     (unsigned long long)stats->uc_miss[c],
                     (unsigned long long)stats->warm_hit[c],
-                    (unsigned long long)stats->shared_lock[c]);
+                    (unsigned long long)stats->shared_lock[c],
+                    (unsigned long long)stats->tls_carve_attempt[c],
+                    (unsigned long long)stats->tls_carve_success[c]);
         }
     }
 }

core/box/tiny_class_stats_box.h

@@ -16,6 +16,8 @@ typedef struct TinyClassStatsThread {
     uint64_t uc_miss[TINY_NUM_CLASSES];    // unified_cache_refill() hits
     uint64_t warm_hit[TINY_NUM_CLASSES];   // warm pool successes
     uint64_t shared_lock[TINY_NUM_CLASSES]; // shared pool lock acquisitions (hook as needed)
+    uint64_t tls_carve_attempt[TINY_NUM_CLASSES]; // Warm/TLS carve attempts
+    uint64_t tls_carve_success[TINY_NUM_CLASSES]; // Warm/TLS carve successes
 } TinyClassStatsThread;
 
 extern __thread TinyClassStatsThread g_tiny_class_stats;
@@ -24,6 +26,8 @@ extern __thread TinyClassStatsThread g_tiny_class_stats;
 extern _Atomic uint64_t g_tiny_class_stats_uc_miss_global[TINY_NUM_CLASSES];
 extern _Atomic uint64_t g_tiny_class_stats_warm_hit_global[TINY_NUM_CLASSES];
 extern _Atomic uint64_t g_tiny_class_stats_shared_lock_global[TINY_NUM_CLASSES];
+extern _Atomic uint64_t g_tiny_class_stats_tls_carve_attempt_global[TINY_NUM_CLASSES];
+extern _Atomic uint64_t g_tiny_class_stats_tls_carve_success_global[TINY_NUM_CLASSES];
 
 static inline void tiny_class_stats_on_uc_miss(int ci) {
     if (ci >= 0 && ci < TINY_NUM_CLASSES) {
@@ -49,6 +53,22 @@ static inline void tiny_class_stats_on_shared_lock(int ci) {
     }
 }
 
+static inline void tiny_class_stats_on_tls_carve_attempt(int ci) {
+    if (ci >= 0 && ci < TINY_NUM_CLASSES) {
+        g_tiny_class_stats.tls_carve_attempt[ci]++;
+        atomic_fetch_add_explicit(&g_tiny_class_stats_tls_carve_attempt_global[ci],
+                                  1, memory_order_relaxed);
+    }
+}
+
+static inline void tiny_class_stats_on_tls_carve_success(int ci) {
+    if (ci >= 0 && ci < TINY_NUM_CLASSES) {
+        g_tiny_class_stats.tls_carve_success[ci]++;
+        atomic_fetch_add_explicit(&g_tiny_class_stats_tls_carve_success_global[ci],
+                                  1, memory_order_relaxed);
+    }
+}
+
 // Optional: reset per-thread counters (cold path only).
 void tiny_class_stats_reset_thread(void);
 

core/box/tiny_mem_stats_box.c

@@ -0,0 +1,65 @@
+// tiny_mem_stats_box.c - Memory accounting helpers for Tiny front components
+
+#include "tiny_mem_stats_box.h"
+
+#include <stdatomic.h>
+#include <sys/types.h>
+#include <stdio.h>
+
+_Atomic long long g_tiny_mem_unified_cache_bytes = 0;
+_Atomic long long g_tiny_mem_warm_pool_bytes = 0;
+_Atomic long long g_tiny_mem_page_box_bytes = 0;
+_Atomic long long g_tiny_mem_tls_magazine_bytes = 0;
+_Atomic long long g_tiny_mem_policy_stats_bytes = 0;
+
+static inline void tiny_mem_stats_add(_Atomic long long* target, ssize_t bytes) {
+    if (!target || bytes == 0) {
+        return;
+    }
+    atomic_fetch_add_explicit(target, (long long)bytes, memory_order_relaxed);
+}
+
+void tiny_mem_stats_add_unified(ssize_t bytes) {
+    tiny_mem_stats_add(&g_tiny_mem_unified_cache_bytes, bytes);
+}
+
+void tiny_mem_stats_add_warm(ssize_t bytes) {
+    tiny_mem_stats_add(&g_tiny_mem_warm_pool_bytes, bytes);
+}
+
+void tiny_mem_stats_add_pagebox(ssize_t bytes) {
+    tiny_mem_stats_add(&g_tiny_mem_page_box_bytes, bytes);
+}
+
+void tiny_mem_stats_add_tls_magazine(ssize_t bytes) {
+    tiny_mem_stats_add(&g_tiny_mem_tls_magazine_bytes, bytes);
+}
+
+void tiny_mem_stats_add_policy_stats(ssize_t bytes) {
+    tiny_mem_stats_add(&g_tiny_mem_policy_stats_bytes, bytes);
+}
+
+void tiny_mem_stats_dump(void) {
+    long long unified = atomic_load_explicit(&g_tiny_mem_unified_cache_bytes,
+                                             memory_order_relaxed);
+    long long warm = atomic_load_explicit(&g_tiny_mem_warm_pool_bytes,
+                                          memory_order_relaxed);
+    long long pagebox = atomic_load_explicit(&g_tiny_mem_page_box_bytes,
+                                             memory_order_relaxed);
+    long long tls_mag = atomic_load_explicit(&g_tiny_mem_tls_magazine_bytes,
+                                             memory_order_relaxed);
+    long long policy_stats = atomic_load_explicit(&g_tiny_mem_policy_stats_bytes,
+                                                  memory_order_relaxed);
+
+    long long total = unified + warm + pagebox + tls_mag + policy_stats;
+
+    fprintf(stderr,
+            "[TINY_MEM_STATS] unified_cache=%lldKB warm_pool=%lldKB page_box=%lldKB "
+            "tls_mag=%lldKB policy_stats=%lldKB total=%lldKB\n",
+            unified / 1024,
+            warm / 1024,
+            pagebox / 1024,
+            tls_mag / 1024,
+            policy_stats / 1024,
+            total / 1024);
+}

core/box/tiny_mem_stats_box.h

@@ -0,0 +1,38 @@
+// tiny_mem_stats_box.h - Lightweight memory accounting for Tiny front boxes
+//
+// Purpose:
+//   - Provide coarse-grained byte counters for major Tiny front allocations
+//     (Unified Cache buffers, Warm Pool TLS state, Page Box TLS state,
+//      TLS magazine/front caches, and policy/stats tables).
+//   - Keep overhead near-zero: helpers are simple fetch-adds, typically called
+//     at init time when the structures are allocated.
+//
+// Usage:
+//   - Call tiny_mem_stats_add_*() at allocation/free sites (positive/negative).
+//   - Call tiny_mem_stats_dump() when HAKMEM_TINY_MEM_DUMP is set to emit one
+//     summary line to stderr (values reported in KB).
+
+#ifndef TINY_MEM_STATS_BOX_H
+#define TINY_MEM_STATS_BOX_H
+
+#include <stddef.h>
+#include <stdint.h>
+#include <sys/types.h>
+
+// Byte counters (signed to allow subtracting on free paths)
+extern _Atomic long long g_tiny_mem_unified_cache_bytes;
+extern _Atomic long long g_tiny_mem_warm_pool_bytes;
+extern _Atomic long long g_tiny_mem_page_box_bytes;
+extern _Atomic long long g_tiny_mem_tls_magazine_bytes;
+extern _Atomic long long g_tiny_mem_policy_stats_bytes;
+
+void tiny_mem_stats_add_unified(ssize_t bytes);
+void tiny_mem_stats_add_warm(ssize_t bytes);
+void tiny_mem_stats_add_pagebox(ssize_t bytes);
+void tiny_mem_stats_add_tls_magazine(ssize_t bytes);
+void tiny_mem_stats_add_policy_stats(ssize_t bytes);
+
+// Dump one line summary (values in KB) if hooked by caller.
+void tiny_mem_stats_dump(void);
+
+#endif // TINY_MEM_STATS_BOX_H

core/box/tiny_page_box.c

@@ -1,6 +1,5 @@
 #include "tiny_page_box.h"
 
 // TLS state definitions for Tiny Page Box
-__thread TinyPageBoxState g_tiny_page_box_state[TINY_NUM_CLASSES];
+__thread TinyPageBoxContext g_tiny_page_box[TINY_NUM_CLASSES];
 __thread int g_tiny_page_box_init_done = 0;
-

core/box/tiny_page_box.h

@@ -9,7 +9,7 @@
 //   - API is generic over class_idx (0-7), but enabled-classes are controlled
 //     by ENV so that we can start with C7 only and later extend to C5/C6.
 //   - When enabled for a class:
-//       tiny_page_box_refill(class_idx, out, max) will try to supply up to
+//       tiny_page_box_refill(class_idx, tls, out, max) will try to supply up to
 //       `max` BASE pointers using per-page freelist before falling back.
 //   - When disabled for a class: the box returns 0 and caller uses legacy path.
 //
@@ -37,6 +37,7 @@
 #include "../superslab/superslab_types.h"   // For TinySlabMeta, SuperSlab
 #include "../box/tiny_next_ptr_box.h"       // For tiny_next_read()
 #include "../hakmem_tiny_superslab.h"       // For tiny_stride_for_class(), base helpers, superslab_ref_inc/dec
+#include "../box/tiny_mem_stats_box.h"      // For coarse memory accounting
 
 // Superslab active counter（Release Guard Box と整合性を取るためのカウンタ更新）
 extern void ss_active_add(SuperSlab* ss, uint32_t n);
@@ -61,19 +62,28 @@ typedef struct TinyPageDesc {
 //   - enabled: このクラスで Page Box を使うかどうか
 //   - num_pages: 現在保持しているページ数（0〜TINY_PAGE_BOX_MAX_PAGES）
 //   - pages[]: TLS が掴んだ C7/C5/C6 ページの ring（小さなバッファ）
-typedef struct TinyPageBoxState {
+typedef struct TinyPageBoxContext {
     uint8_t      enabled;    // 1=Page Box enabled for this class, 0=disabled
     uint8_t      num_pages;  // 有効な pages[] エントリ数
     uint8_t      _pad[2];
     TinyPageDesc pages[TINY_PAGE_BOX_MAX_PAGES];
-} TinyPageBoxState;
+} TinyPageBoxContext;
 
-// TLS/state: one TinyPageBoxState per class（per-thread Box）
+// TLS/state: one TinyPageBoxContext per class（per-thread Box）
-extern __thread TinyPageBoxState g_tiny_page_box_state[TINY_NUM_CLASSES];
+extern __thread TinyPageBoxContext g_tiny_page_box[TINY_NUM_CLASSES];
 
 // One-shot init guard（per-thread）
 extern __thread int g_tiny_page_box_init_done;
 
+static inline int tiny_page_box_log_enabled(void) {
+    static int g_page_box_log = -1;
+    if (__builtin_expect(g_page_box_log == -1, 0)) {
+        const char* e = getenv("HAKMEM_TINY_PAGEBOX_LOG");
+        g_page_box_log = (e && *e && *e != '0') ? 1 : 0;
+    }
+    return g_page_box_log;
+}
+
 // Helper: parse class list from ENV and set enabled flags.
 // Default behaviour (ENV unset/empty) is to enable class 7 only.
 static inline void tiny_page_box_init_once(void) {
@@ -82,13 +92,14 @@ static inline void tiny_page_box_init_once(void) {
     }
 
     // Clear all state
-    memset(g_tiny_page_box_state, 0, sizeof(g_tiny_page_box_state));
+    memset(g_tiny_page_box, 0, sizeof(g_tiny_page_box));
+    tiny_mem_stats_add_pagebox((ssize_t)sizeof(g_tiny_page_box));
 
     const char* env = getenv("HAKMEM_TINY_PAGE_BOX_CLASSES");
     if (!env || !*env) {
         // Default: enable mid-size classes (C5–C7)
         for (int c = 5; c <= 7 && c < TINY_NUM_CLASSES; c++) {
-            g_tiny_page_box_state[c].enabled = 1;
+            g_tiny_page_box[c].enabled = 1;
         }
     } else {
         // Parse simple comma-separated list of integers: "5,6,7"
@@ -107,7 +118,7 @@ static inline void tiny_page_box_init_once(void) {
                 p++;
             }
             if (val >= 0 && val < TINY_NUM_CLASSES) {
-                g_tiny_page_box_state[val].enabled = 1;
+                g_tiny_page_box[val].enabled = 1;
             }
         }
     }
@@ -123,7 +134,7 @@ static inline int tiny_page_box_is_enabled(int class_idx) {
     if (class_idx < 0 || class_idx >= TINY_NUM_CLASSES) {
         return 0;
     }
-    return g_tiny_page_box_state[class_idx].enabled != 0;
+    return g_tiny_page_box[class_idx].enabled != 0;
 }
 
 // Forward declaration for TLS slab state（tiny_tls.h から参照）
@@ -133,7 +144,7 @@ extern __thread TinyTLSSlab g_tls_slabs[TINY_NUM_CLASSES];
 // ここで Page Box が利用可能なページとして登録しておくことで、
 // 後続の unified_cache_refill() から Superslab/Warm Pool に落ちる前に
 // 「既に TLS が掴んでいるページ」を優先的に使えるようにする。
-static inline void tiny_page_box_on_new_slab(TinyTLSSlab* tls)
+static inline void tiny_page_box_on_new_slab(int class_idx, TinyTLSSlab* tls)
 {
     if (!tls) {
         return;
@@ -143,6 +154,10 @@ static inline void tiny_page_box_on_new_slab(TinyTLSSlab* tls)
         tiny_page_box_init_once();
     }
 
+    if (class_idx < 0 || class_idx >= TINY_NUM_CLASSES) {
+        return;
+    }
+
     SuperSlab* ss = tls->ss;
     TinySlabMeta* meta = tls->meta;
     uint8_t* base = tls->slab_base;
@@ -152,12 +167,11 @@ static inline void tiny_page_box_on_new_slab(TinyTLSSlab* tls)
         return;
     }
 
-    int class_idx = (int)meta->class_idx;
+    if (meta->class_idx != (uint8_t)class_idx) {
-    if (class_idx < 0 || class_idx >= TINY_NUM_CLASSES) {
         return;
     }
 
-    TinyPageBoxState* st = &g_tiny_page_box_state[class_idx];
+    TinyPageBoxContext* st = &g_tiny_page_box[class_idx];
     if (!st->enabled) {
         return;
     }
@@ -200,9 +214,11 @@ static inline void tiny_page_box_on_new_slab(TinyTLSSlab* tls)
     superslab_ref_inc(ss);
 
 #if !HAKMEM_BUILD_RELEASE
-    // Debug: Track Page Box stats per-class
+    // Debug: Track Page Box stats per-class（ENV: HAKMEM_TINY_PAGEBOX_LOG=0 で抑制）
+    if (tiny_page_box_log_enabled()) {
         fprintf(stderr, "[PAGE_BOX_REG] class=%d num_pages=%u capacity=%u carved=%u\n",
                 class_idx, st->num_pages, meta->capacity, meta->carved);
+    }
 #endif
 }
 
@@ -219,9 +235,11 @@ static inline void tiny_page_box_on_new_slab(TinyTLSSlab* tls)
 //   - Superslab/Shared Pool 呼び出し頻度を徐々に観測・調整できる。
 
 static inline int tiny_page_box_refill(int class_idx,
+                                       TinyTLSSlab* tls,
                                        void** out,
                                        int max_out)
 {
+    (void)tls; // reserved for future per-TLS hints
     if (!tiny_page_box_is_enabled(class_idx)) {
         return 0;
     }
@@ -233,7 +251,7 @@ static inline int tiny_page_box_refill(int class_idx,
         return 0;
     }
 
-    TinyPageBoxState* st = &g_tiny_page_box_state[class_idx];
+    TinyPageBoxContext* st = &g_tiny_page_box[class_idx];
     if (st->num_pages == 0) {
         return 0;
     }

core/box/tiny_policy_learner_box.c

@@ -4,39 +4,78 @@
 #include "tiny_class_policy_box.h"
 #include "tiny_class_stats_box.h"
 #include <stdint.h>
+#include <stdio.h>
 
-// Simple OBSERVE/LEARN rule:
+// Simple OBSERVE/LEARN rule (auto profile only):
-//   - Choose top-2 classes by shared_pool_lock and enable Page Box for them.
+//   - C7 は常に ON (page + warm, cap=8)
-//   - Always keep existing warm_enabled / warm_cap (policy table is already seeded).
+//   - それ以外のクラスから score = shared_lock*4 + uc_miss の上位2つだけ page/warm を ON
+//   - warm_cap は C5–C7:8, それ以外:4
+//   - スコアが 0 なら何も変更しない
 void tiny_policy_learner_tick(void) {
+    if (!tiny_class_policy_is_auto()) {
+        return;
+    }
+
     TinyClassStatsThread snap = {0};
     tiny_class_stats_snapshot_global(&snap);
 
+    // 事前に全クラスを OFF ベースに初期化（cap はデフォルト値に）
+    for (int c = 0; c < TINY_NUM_CLASSES; c++) {
+        TinyClassPolicy* p = &g_tiny_class_policy[c];
+        p->page_box_enabled = 0;
+        p->warm_enabled = 0;
+        p->warm_cap = (c >= 5) ? 8 : 4;
+        p->tls_carve_enabled = 0;
+    }
+
+    // C7 は常に ON
+    g_tiny_class_policy[7].page_box_enabled = 1;
+    g_tiny_class_policy[7].warm_enabled = 1;
+    g_tiny_class_policy[7].warm_cap = 8;
+    g_tiny_class_policy[7].tls_carve_enabled = 1;
+
+    // C7 を除く上位2クラスをスコアで選択
     int top1 = -1, top2 = -1;
     uint64_t v1 = 0, v2 = 0;
     for (int i = 0; i < TINY_NUM_CLASSES; i++) {
-        uint64_t v = snap.shared_lock[i];
+        if (i == 7) continue;
-        if (v > v1) {
+        uint64_t score = snap.shared_lock[i] * 4 + snap.uc_miss[i];
+        if (score > v1) {
             top2 = top1;
             v2 = v1;
             top1 = i;
-            v1 = v;
+            v1 = score;
-        } else if (v > v2) {
+        } else if (score > v2) {
             top2 = i;
-            v2 = v;
+            v2 = score;
         }
     }
 
-    // Nothing observed yet → leave policy untouched
+    // スコアが全く無い場合は C7 だけ維持
     if (v1 == 0) {
         return;
     }
 
-    for (int c = 0; c < TINY_NUM_CLASSES; c++) {
+    if (top1 >= 0) {
-        TinyClassPolicy* p = &g_tiny_class_policy[c];
+        TinyClassPolicy* p = &g_tiny_class_policy[top1];
-        if (c == top1 || c == top2) {
         p->page_box_enabled = 1;
         p->warm_enabled = 1;
+        p->tls_carve_enabled = 1;
+    }
+    if (top2 >= 0 && v2 > 0) {
+        TinyClassPolicy* p = &g_tiny_class_policy[top2];
+        p->page_box_enabled = 1;
+        p->warm_enabled = 1;
+        p->tls_carve_enabled = 1;
+    }
+
+    // 1-shot ログ（最多 4 回まで）
+    static _Atomic uint32_t auto_logs = 0;
+    if (tiny_policy_log_enabled()) {
+        uint32_t n = atomic_fetch_add_explicit(&auto_logs, 1, memory_order_relaxed);
+        if (n < 4) {
+            fprintf(stderr, "[POLICY_AUTO_UPDATE] profile=auto (top=%d/%d)\n", top1, top2);
+            tiny_class_policy_dump(NULL);
         }
     }
 }

core/box/tiny_tls_carve_one_block_box.h

@@ -7,6 +7,7 @@
 #include "../tiny_debug_api.h"        // tiny_refill_failfast_level(), tiny_failfast_abort_ptr()
 #include "c7_meta_used_counter_box.h" // C7 meta->used telemetry (Release/Debug共通)
 #include "tiny_next_ptr_box.h"
+#include "tiny_class_stats_box.h"
 #include "../superslab/superslab_inline.h"
 #include <stdatomic.h>
 #include <signal.h>
@@ -41,6 +42,8 @@ tiny_tls_carve_one_block(TinyTLSSlab* tls, int class_idx)
     if (meta->class_idx != (uint8_t)class_idx) return res;
     if (tls->slab_idx < 0 || tls->slab_idx >= ss_slabs_capacity(tls->ss)) return res;
 
+    tiny_class_stats_on_tls_carve_attempt(class_idx);
+
     // Freelist pop
     if (meta->freelist) {
 #if !HAKMEM_BUILD_RELEASE
@@ -61,6 +64,7 @@ tiny_tls_carve_one_block(TinyTLSSlab* tls, int class_idx)
         meta->used++;
         c7_meta_used_note(meta->class_idx, C7_META_USED_SRC_TLS);
         ss_active_add(tls->ss, 1);
+        tiny_class_stats_on_tls_carve_success(class_idx);
         res.block = block;
         res.path = TINY_TLS_CARVE_PATH_FREELIST;
         return res;
@@ -93,6 +97,7 @@ tiny_tls_carve_one_block(TinyTLSSlab* tls, int class_idx)
         meta->used++;
         c7_meta_used_note(meta->class_idx, C7_META_USED_SRC_TLS);
         ss_active_add(tls->ss, 1);
+        tiny_class_stats_on_tls_carve_success(class_idx);
         res.block = block;
         res.path = TINY_TLS_CARVE_PATH_LINEAR;
         return res;

core/box/warm_pool_prefill_box.h

@@ -9,6 +9,7 @@
 #include <stdint.h>
 #include <stdatomic.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include "../hakmem_tiny_config.h"
 #include "../hakmem_tiny_superslab.h"
 #include "../tiny_tls.h"
@@ -18,8 +19,18 @@
 
 extern _Atomic uintptr_t g_c7_stage3_magic_ss;
 
+static inline int warm_prefill_log_enabled(void) {
+    static int g_warm_log = -1;
+    if (__builtin_expect(g_warm_log == -1, 0)) {
+        const char* e = getenv("HAKMEM_TINY_WARM_LOG");
+        g_warm_log = (e && *e && *e != '0') ? 1 : 0;
+    }
+    return g_warm_log;
+}
+
 static inline void warm_prefill_log_c7_meta(const char* tag, TinyTLSSlab* tls) {
     if (!tls || !tls->ss) return;
+    if (!warm_prefill_log_enabled()) return;
 #if HAKMEM_BUILD_RELEASE
     static _Atomic uint32_t rel_logs = 0;
     uint32_t n = atomic_fetch_add_explicit(&rel_logs, 1, memory_order_relaxed);
@@ -116,7 +127,7 @@ static inline int warm_pool_do_prefill(int class_idx, TinyTLSSlab* tls, int warm
         }
 
         // C7 safety: prefer only pristine slabs (used=0 carved=0 freelist=NULL)
-        if (class_idx == 7) {
+        if (class_idx == 7 && warm_prefill_log_enabled()) {
             TinySlabMeta* meta = &tls->ss->slabs[tls->slab_idx];
             if (meta->class_idx == 7 &&
                 (meta->used > 0 || meta->carved > 0 || meta->freelist != NULL)) {
@@ -162,7 +173,7 @@ static inline int warm_pool_do_prefill(int class_idx, TinyTLSSlab* tls, int warm
             warm_pool_rel_c7_prefill_slab();
         }
         #else
-        if (class_idx == 7) {
+        if (class_idx == 7 && warm_prefill_log_enabled()) {
             static __thread int dbg_c7_prefill_logs = 0;
             if (dbg_c7_prefill_logs < 8) {
                 TinySlabMeta* meta = &tls->ss->slabs[tls->slab_idx];

core/box/warm_pool_stats_box.h

@@ -23,31 +23,19 @@ extern __thread TinyWarmPoolStats g_warm_pool_stats[TINY_NUM_CLASSES];
 // Record a warm pool hit
 // Called when warm_pool_pop() succeeds and carve produces blocks
 static inline void warm_pool_record_hit(int class_idx) {
-#if HAKMEM_DEBUG_COUNTERS
     g_warm_pool_stats[class_idx].hits++;
-#else
-    (void)class_idx;
-#endif
 }
 
 // Record a warm pool miss
 // Called when warm_pool_pop() returns NULL (pool empty)
 static inline void warm_pool_record_miss(int class_idx) {
-#if HAKMEM_DEBUG_COUNTERS
     g_warm_pool_stats[class_idx].misses++;
-#else
-    (void)class_idx;
-#endif
 }
 
 // Record a warm pool prefill event
 // Called when pool is empty and we do secondary prefill
 static inline void warm_pool_record_prefilled(int class_idx) {
-#if HAKMEM_DEBUG_COUNTERS
     g_warm_pool_stats[class_idx].prefilled++;
-#else
-    (void)class_idx;
-#endif
 }
 
 #endif // HAK_WARM_POOL_STATS_BOX_H

core/front/malloc_tiny_fast.h

@@ -36,6 +36,7 @@
 #include "../hakmem_tiny.h"         // For hak_tiny_size_to_class
 #include "../box/tiny_front_hot_box.h"  // Phase 4-Step2: Hot Path Box
 #include "../box/tiny_front_cold_box.h" // Phase 4-Step2: Cold Path Box
+#include "../box/tiny_c7_hotpath_box.h" // Optional: C7 専用ホットパス
 
 // Helper: current thread id (low 32 bits) for owner check
 #ifndef TINY_SELF_U32_LOCAL_DEFINED
@@ -98,6 +99,11 @@ static inline void* malloc_tiny_fast(size_t size) {
     // 1. size → class_idx (inline table lookup, 1-2 instructions)
     int class_idx = hak_tiny_size_to_class(size);
 
+    // Optional: C7 専用ホットパス（環境変数 HAKMEM_TINY_C7_HOT でON）
+    if (__builtin_expect(class_idx == 7 && tiny_c7_hot_enabled(), 0)) {
+        return tiny_c7_alloc_hot(size);
+    }
+
     // 2. Phase 4-Step2: Hot/Cold Path Box
     // Try hot path first (cache hit, 1 branch)
     void* ptr = tiny_hot_alloc_fast(class_idx);
@@ -235,6 +241,14 @@ static inline int free_tiny_fast(void* ptr) {
     }
 #endif
 
+    // Optional: C7 専用ホットパス（キャッシュのみで完了させる）
+    if (__builtin_expect(class_idx == 7 && tiny_c7_hot_enabled(), 0)) {
+        if (tiny_c7_free_hot(base)) {
+            return 1;
+        }
+        // fallthrough to unified cache push on failure
+    }
+
     int pushed = unified_cache_push(class_idx, HAK_BASE_FROM_RAW(base));
     if (__builtin_expect(pushed, 1)) {
         return 1;  // Success

core/front/tiny_unified_cache.c

@@ -17,6 +17,7 @@
 #undef WARM_POOL_REL_DEFINE
 #include "../box/c7_meta_used_counter_box.h"    // Box: C7 meta->used increment counters
 #include "../box/warm_pool_prefill_box.h"    // Box: Warm Pool Prefill (secondary optimization)
+#include "../box/tiny_mem_stats_box.h"       // Box: Tiny front memory accounting
 #include "../hakmem_env_cache.h"             // Priority-2: ENV cache (eliminate syscalls)
 #include "../box/tiny_page_box.h"           // Tiny-Plus Page Box (C5–C7 initial hook)
 #include "../box/ss_tls_bind_box.h"         // Box: TLS Bind (SuperSlab -> TLS binding)
@@ -205,6 +206,8 @@ void unified_cache_init(void) {
             continue;  // Skip this class, try others
         }
 
+        tiny_mem_stats_add_unified((ssize_t)(cap * sizeof(void*)));
+
         g_unified_cache[cls].capacity = (uint16_t)cap;
         g_unified_cache[cls].mask = (uint16_t)(cap - 1);
         g_unified_cache[cls].head = 0;
@@ -522,6 +525,7 @@ hak_base_ptr_t unified_cache_refill(int class_idx) {
     int warm_enabled = policy ? policy->warm_enabled : 0;
     int warm_cap = policy ? policy->warm_cap : 0;
     int page_enabled = policy ? policy->page_box_enabled : 0;
+    TinyTLSSlab* tls = &g_tls_slabs[class_idx];
 
     // ✅ Phase 11+: Ensure cache is initialized (lazy init for cold path)
     if (!cache->slots) {
@@ -562,12 +566,15 @@ hak_base_ptr_t unified_cache_refill(int class_idx) {
     void* out[512];
     int produced = 0;
     int tls_carved = 0;  // Debug bookkeeping: track TLS carve experiment hits
+#if HAKMEM_BUILD_RELEASE
+    (void)tls_carved;
+#endif
 
     // ========== PAGE BOX HOT PATH（Tiny-Plus 層）: Try page box FIRST ==========
     // 将来的に C7 専用の page-level freelist 管理をここに統合する。
     // いまは stub 実装で常に 0 を返すが、Box 境界としての接続だけ先に行う。
     if (page_enabled && tiny_page_box_is_enabled(class_idx)) {
-        int page_produced = tiny_page_box_refill(class_idx, out, room);
+        int page_produced = tiny_page_box_refill(class_idx, tls, out, room);
         if (page_produced > 0) {
             // Store blocks into cache and return first
             void* first = out[0];
@@ -625,12 +632,19 @@ hak_base_ptr_t unified_cache_refill(int class_idx) {
         #endif
         SuperSlab* warm_ss = tiny_warm_pool_pop(class_idx);
         if (warm_ss) {
+            int allow_tls_bind = policy && policy->tls_carve_enabled;
+            int allow_tls_carve = allow_tls_bind;
+            int warm_mode = 0;
             if (class_idx == 7) {
                 #if !HAKMEM_BUILD_RELEASE
                 warm_pool_dbg_c7_hit();
                 #endif
-                int warm_mode = warm_tls_bind_mode_c7();
+                warm_mode = warm_tls_bind_mode_c7();
-                if (warm_mode >= 1) {
+                allow_tls_bind = (warm_mode >= 1);
+                allow_tls_carve = (warm_mode == 2);
+            }
+
+            if (allow_tls_bind) {
                 int cap = ss_slabs_capacity(warm_ss);
                 int slab_idx = -1;
 
@@ -643,27 +657,33 @@ hak_base_ptr_t unified_cache_refill(int class_idx) {
                 }
 
                 if (slab_idx >= 0) {
-                        TinyTLSSlab* tls = &g_tls_slabs[class_idx];
                     uint32_t tid = (uint32_t)(uintptr_t)pthread_self();
                     if (ss_tls_bind_one(class_idx, tls, warm_ss, slab_idx, tid)) {
+                        if (class_idx == 7) {
                             warm_tls_bind_log_success(warm_ss, slab_idx);
+                        }
 
-                            // Mode 2: carve a single block via TLS fast path
+                        // Mode 2: carve a single block via TLS fast path (policy enabled classes)
-                            if (warm_mode == 2) {
+                        if (allow_tls_carve) {
                             #if !HAKMEM_BUILD_RELEASE
+                            if (class_idx == 7) {
                                 warm_pool_dbg_c7_tls_attempt();
+                            }
                             #endif
                             TinyTLSCarveOneResult tls_carve =
                                 tiny_tls_carve_one_block(tls, class_idx);
                             if (tls_carve.block) {
+                                if (class_idx == 7) {
                                     warm_tls_bind_log_tls_carve(warm_ss, slab_idx, tls_carve.block);
                                     #if !HAKMEM_BUILD_RELEASE
                                     warm_pool_dbg_c7_tls_success();
                                     #endif
+                                }
                                 out[0] = tls_carve.block;
                                 produced = 1;
                                 tls_carved = 1;
                             } else {
+                                if (class_idx == 7) {
                                     warm_tls_bind_log_tls_fail(warm_ss, slab_idx);
                                     #if !HAKMEM_BUILD_RELEASE
                                     warm_pool_dbg_c7_tls_fail();
@@ -774,8 +794,6 @@ hak_base_ptr_t unified_cache_refill(int class_idx) {
         warm_pool_record_miss(class_idx);
     }
 
-    TinyTLSSlab* tls = &g_tls_slabs[class_idx];
-
     // Step 1: Ensure SuperSlab available via normal refill
     // Enhanced: Use Warm Pool Prefill Box for secondary prefill when pool is empty
     if (warm_enabled) {

core/front/tiny_warm_pool.h

@@ -10,6 +10,7 @@
 #include <stdint.h>
 #include "../hakmem_tiny_config.h"
 #include "../superslab/superslab_types.h"
+#include "../box/tiny_mem_stats_box.h"
 
 // ============================================================================
 // Warm Pool Design
@@ -74,6 +75,7 @@ static inline void tiny_warm_pool_init_once(void) {
         for (int i = 0; i < TINY_NUM_CLASSES; i++) {
             g_tiny_warm_pool[i].count = 0;
         }
+        tiny_mem_stats_add_warm((ssize_t)(sizeof(g_tiny_warm_pool) + sizeof(g_warm_pool_stats)));
         initialized = 1;
     }
 }

core/hakmem_shared_pool.c

@@ -7,6 +7,7 @@
 #include "box/tls_sll_drain_box.h"        // Box TLS SLL Drain (tiny_tls_sll_drain)
 #include "box/tls_slab_reuse_guard_box.h" // Box TLS Slab Reuse Guard (P0.3)
 #include "hakmem_policy.h"                // FrozenPolicy (learning layer)
+#include "box/shared_pool_box.h"          // Logical cap for bench profile
 
 #include <stdlib.h>
 #include <string.h>
@@ -287,6 +288,7 @@ shared_pool_init(void)
 {
     // Idempotent init; safe to call from multiple early paths.
     // pthread_mutex_t with static initializer is already valid.
+    shared_pool_box_init(NULL, NULL);
     pthread_mutex_lock(&g_shared_pool.alloc_lock);
     if (g_shared_pool.capacity == 0 && g_shared_pool.slabs == NULL) {
         shared_pool_ensure_capacity_unlocked(16);

core/hakmem_shared_pool_acquire.c

@@ -12,6 +12,10 @@
 #include "front/tiny_warm_pool.h"  // Warm Pool: Prefill during registry scans
 #include "box/ss_slab_reset_box.h" // Box: Reset slab metadata on reuse (C7 guard)
 #include "box/tiny_class_stats_box.h" // OBSERVE: per-class shared lock stats
+#include "box/ss_stats_box.h" // OBSERVE: Superslab/slab event counters
+#include "box/ss_budget_box.h" // Budget guard for Superslab growth (larson_guard)
+#include "box/super_reg_box.h" // Logical limit for registry scan
+#include "box/shared_pool_box.h" // Logical cap for shared pool slots (bench profile)
 
 #include <stdint.h>
 #include <stdlib.h>
@@ -22,8 +26,8 @@
 _Atomic uintptr_t g_c7_stage3_magic_ss = 0;
 
 static inline void sp_lock_with_stats(int class_idx) {
+    pthread_mutex_lock(&g_shared_pool.alloc_lock);
     tiny_class_stats_on_shared_lock(class_idx);
-    sp_lock_with_stats(class_idx);
 }
 
 static inline void c7_log_meta_state(const char* tag, SuperSlab* ss, int slab_idx) {
@@ -159,6 +163,37 @@ static inline int c7_reset_and_log_if_needed(SuperSlab* ss,
     return 0;
 }
 
+static inline void sp_reset_superslab_all_slabs(SuperSlab* ss,
+                                                int class_idx,
+                                                int from_lru) {
+    if (!ss) {
+        return;
+    }
+    int cap = ss_slabs_capacity(ss);
+    ss->slab_bitmap = 0;
+    ss->nonempty_mask = 0;
+    ss->freelist_mask = 0;
+    ss->empty_mask = 0;
+    ss->empty_count = 0;
+    ss->active_slabs = 0;
+    ss->hot_count = 0;
+    ss->cold_count = 0;
+    for (int s = 0; s < cap; s++) {
+        ss_slab_reset_meta_for_tiny(ss, s, class_idx);
+    }
+    ss_stats_on_ss_scan(class_idx, 0, 1);
+    static _Atomic uint32_t rel_stage3_reset_logs = 0;
+    uint32_t n = atomic_fetch_add_explicit(&rel_stage3_reset_logs, 1, memory_order_relaxed);
+    if (n < 4) {
+        fprintf(stderr,
+                "[REL_STAGE3_RESET] class=%d ss=%p from_lru=%d cap=%d\n",
+                class_idx,
+                (void*)ss,
+                from_lru,
+                cap);
+    }
+}
+
 // ============================================================================
 // Performance Measurement: Shared Pool Lock Contention (ENV-gated)
 // ============================================================================
@@ -208,10 +243,13 @@ sp_acquire_from_empty_scan(int class_idx, SuperSlab** ss_out, int* slab_idx_out,
         return -1;
     }
 
-    extern SuperSlab* g_super_reg_by_class[TINY_NUM_CLASSES][SUPER_REG_PER_CLASS];
     extern int g_super_reg_class_size[TINY_NUM_CLASSES];
 
     int reg_size = (class_idx < TINY_NUM_CLASSES) ? g_super_reg_class_size[class_idx] : 0;
+    int reg_cap = super_reg_effective_per_class();
+    if (reg_cap > 0 && reg_size > reg_cap) {
+        reg_size = reg_cap;
+    }
     // Priority-2: Use cached ENV
     int scan_limit = HAK_ENV_SS_EMPTY_SCAN_LIMIT();
     if (scan_limit > reg_size) scan_limit = reg_size;
@@ -229,7 +267,7 @@ sp_acquire_from_empty_scan(int class_idx, SuperSlab** ss_out, int* slab_idx_out,
     int primary_slab_idx = -1;
 
     for (int i = 0; i < scan_limit; i++) {
-        SuperSlab* ss = g_super_reg_by_class[class_idx][i];
+        SuperSlab* ss = super_reg_by_class_at(class_idx, i);
         if (!(ss && ss->magic == SUPERSLAB_MAGIC)) continue;
         // P-Tier: Skip DRAINING tier SuperSlabs
         if (!ss_tier_is_hot(ss)) continue;
@@ -769,6 +807,26 @@ stage2_scan:
                                   1, memory_order_relaxed);
     }
 
+    // bench プロファイルでは Shared Pool の論理上限を軽くかけておく
+    uint32_t total_limit = shared_pool_effective_total_slots();
+    if (total_limit > 0 && g_shared_pool.total_count >= total_limit) {
+        if (g_lock_stats_enabled == 1) {
+            atomic_fetch_add(&g_lock_release_count, 1);
+        }
+        pthread_mutex_unlock(&g_shared_pool.alloc_lock);
+        return -1;
+    }
+    uint32_t class_limit = shared_pool_effective_class_slots(class_idx);
+    if (class_limit > 0 &&
+        class_idx < TINY_NUM_CLASSES_SS &&
+        (uint32_t)g_shared_pool.class_active_slots[class_idx] >= class_limit) {
+        if (g_lock_stats_enabled == 1) {
+            atomic_fetch_add(&g_lock_release_count, 1);
+        }
+        pthread_mutex_unlock(&g_shared_pool.alloc_lock);
+        return -1;
+    }
+
     // ========== Stage 3: Get new SuperSlab ==========
     // Try LRU cache first, then mmap
     SuperSlab* new_ss = NULL;
@@ -786,6 +844,13 @@ stage2_scan:
 
     // Stage 3b: If LRU miss, allocate new SuperSlab
     if (!new_ss) {
+        if (!ss_budget_on_alloc(class_idx)) {
+            if (g_lock_stats_enabled == 1) {
+                atomic_fetch_add(&g_lock_release_count, 1);
+            }
+            pthread_mutex_unlock(&g_shared_pool.alloc_lock);
+            return -1;
+        }
         // Release the alloc_lock to avoid deadlock with registry during superslab_allocate
         if (g_lock_stats_enabled == 1) {
             atomic_fetch_add(&g_lock_release_count, 1);
@@ -834,27 +899,10 @@ stage2_scan:
         g_shared_pool.total_count++;
     }
 
-    // C7: LRU 再利用・新規確保いずれでも、空スラブに完全リセットしてから返す
+    // Stage3 から返す前に、LRU 再利用分は必ず空スラブ化する。
-    if (class_idx == 7 && new_ss) {
+    // C7 以外でも from_lru の場合は全スラブをリセットしておく。
-        int cap = ss_slabs_capacity(new_ss);
+    if (new_ss && (from_lru || class_idx == 7)) {
-        new_ss->slab_bitmap = 0;
+        sp_reset_superslab_all_slabs(new_ss, class_idx, from_lru);
-        new_ss->nonempty_mask = 0;
-        new_ss->freelist_mask = 0;
-        new_ss->empty_mask = 0;
-        new_ss->empty_count = 0;
-        new_ss->active_slabs = 0;
-        new_ss->hot_count = 0;
-        new_ss->cold_count = 0;
-        for (int s = 0; s < cap; s++) {
-            ss_slab_reset_meta_for_tiny(new_ss, s, class_idx);
-        }
-        static _Atomic uint32_t rel_stage3_reset_logs = 0;
-        uint32_t n = atomic_fetch_add_explicit(&rel_stage3_reset_logs, 1, memory_order_relaxed);
-        if (n < 4) {
-            fprintf(stderr,
-                    "[REL_C7_STAGE3_RESET] ss=%p from_lru=%d cap=%d\n",
-                    (void*)new_ss, from_lru, cap);
-        }
     }
 
     #if !HAKMEM_BUILD_RELEASE

core/hakmem_shared_pool_release.c

@@ -7,6 +7,8 @@
 #include "superslab/superslab_inline.h"  // superslab_ref_get guard for TLS pins
 #include "box/ss_release_guard_box.h"    // Box: SuperSlab Release Guard
 #include "box/ss_slab_reset_box.h"       // Box: Reset slab metadata on reuse path
+#include "box/ss_stats_box.h"            // Observability: Superslab/slab counters
+#include "box/ss_budget_box.h"           // Budget guard (global/class caps)
 
 #include <stdlib.h>
 #include <stdio.h>
@@ -217,6 +219,8 @@ shared_pool_release_slab(SuperSlab* ss, int slab_idx)
         slab_meta->class_idx = 255;  // UNASSIGNED
         // P1.1: Mark class_map as UNASSIGNED when releasing slab
         ss->class_map[slab_idx] = 255;
+        // Reset slab metadata to a pristine state for all classes (C0–C7)
+        ss_slab_reset_meta_for_tiny(ss, slab_idx, -1);
 
         if (ss->active_slabs > 0) {
             ss->active_slabs--;
@@ -284,6 +288,8 @@ shared_pool_release_slab(SuperSlab* ss, int slab_idx)
 
             // Free SuperSlab immediately (bypasses normal active_slots==0 check)
             extern void superslab_free(SuperSlab* ss);
+            ss_stats_on_ss_free_class(class_idx);
+            ss_budget_on_free(class_idx);
             superslab_free(ss);
             return;
         }
@@ -321,6 +327,8 @@ shared_pool_release_slab(SuperSlab* ss, int slab_idx)
         // If so, we must NOT free the SS.
         if (ss_release_guard_superslab_can_free(ss)) {
             extern void superslab_free(SuperSlab* ss);
+            ss_stats_on_ss_free_class(class_idx);
+            ss_budget_on_free(class_idx);
             superslab_free(ss);
         } else {
             #if !HAKMEM_BUILD_RELEASE

core/hakmem_super_registry.c

@@ -4,17 +4,20 @@
 #include "box/ss_addr_map_box.h"     // Phase 9-1: SuperSlab address map
 #include "box/ss_cold_start_box.inc.h"  // Phase 11+: Cold Start prewarm defaults
 #include "hakmem_env_cache.h"        // Priority-2: ENV cache (eliminate syscalls)
+#include <stdlib.h>
 #include <string.h>
 #include <stdio.h>
 #include <sys/mman.h>  // munmap for incompatible SuperSlab eviction
 
-// Global registry storage
+// Global registry storage (allocated via SuperRegBox)
-SuperRegEntry g_super_reg[SUPER_REG_SIZE];
+static SuperRegEntry* reg_entries(void) {
+    return super_reg_entries();
+}
+
 pthread_mutex_t g_super_reg_lock = PTHREAD_MUTEX_INITIALIZER;
 int g_super_reg_initialized = 0;
 
 // Per-class registry storage (Phase 6: Registry Optimization)
-SuperSlab* g_super_reg_by_class[TINY_NUM_CLASSES][SUPER_REG_PER_CLASS];
 int g_super_reg_class_size[TINY_NUM_CLASSES];
 
 // Phase 9: Lazy Deallocation - LRU Cache Storage
@@ -28,11 +31,23 @@ static _Atomic int g_ss_prewarm_bypass = 0;
 void hak_super_registry_init(void) {
     if (g_super_reg_initialized) return;
 
+    super_reg_init(NULL, NULL);
+
+    SuperRegEntry* entries = reg_entries();
+    int reg_cap = super_reg_effective_size();
+    if (!entries) {
+        fprintf(stderr, "[SUPER_REG] init failed: no registry entries\n");
+        abort();
+    }
     // Zero-initialize all entries (hash table)
-    memset(g_super_reg, 0, sizeof(g_super_reg));
+    memset(entries, 0, (size_t)reg_cap * sizeof(SuperRegEntry));
 
     // Zero-initialize per-class registry (Phase 6: Registry Optimization)
-    memset(g_super_reg_by_class, 0, sizeof(g_super_reg_by_class));
+    SuperSlab** by_class = super_reg_by_class_slots();
+    int stride = super_reg_by_class_stride();
+    if (by_class && stride > 0) {
+        memset(by_class, 0, (size_t)TINY_NUM_CLASSES * (size_t)stride * sizeof(SuperSlab*));
+    }
     memset(g_super_reg_class_size, 0, sizeof(g_super_reg_class_size));
 
     // Memory fence to ensure initialization is visible to all threads
@@ -62,12 +77,22 @@ int hak_super_register(uintptr_t base, SuperSlab* ss) {
     const int dbg = 0;
 #endif
 
+    SuperRegEntry* entries = reg_entries();
+    if (!entries) {
+        pthread_mutex_unlock(&g_super_reg_lock);
+        return 0;
+    }
+
     int h = hak_super_hash(base, lg);
+    const int mask = super_reg_effective_mask();
+    const int probe_limit = super_reg_effective_size() > SUPER_MAX_PROBE
+                                ? SUPER_MAX_PROBE
+                                : super_reg_effective_size();
 
     // Step 1: Register in hash table (for address → SuperSlab lookup)
     int hash_registered = 0;
-    for (int i = 0; i < SUPER_MAX_PROBE; i++) {
+    for (int i = 0; i < probe_limit; i++) {
-        SuperRegEntry* e = &g_super_reg[(h + i) & SUPER_REG_MASK];
+        SuperRegEntry* e = &entries[(h + i) & mask];
 
         if (atomic_load_explicit(&e->base, memory_order_acquire) == 0) {
             // Found empty slot
@@ -84,7 +109,7 @@ int hak_super_register(uintptr_t base, SuperSlab* ss) {
             hash_registered = 1;
             if (dbg == 1) {
                 fprintf(stderr, "[SUPER_REG] register base=%p lg=%d slot=%d magic=%llx\n",
-                        (void*)base, lg, (h + i) & SUPER_REG_MASK,
+                        (void*)base, lg, (h + i) & mask,
                         (unsigned long long)ss->magic);
             }
             break;
@@ -131,12 +156,22 @@ void hak_super_unregister(uintptr_t base) {
 
     // Step 1: Find and remove from hash table
     SuperSlab* ss = NULL;  // Save SuperSlab pointer for per-class removal
+    SuperRegEntry* entries = reg_entries();
+    if (!entries) {
+        pthread_mutex_unlock(&g_super_reg_lock);
+        return;
+    }
+
     for (int lg = 20; lg <= 21; lg++) {
         int h = hak_super_hash(base, lg);
+        const int mask = super_reg_effective_mask();
+        const int probe_limit = super_reg_effective_size() > SUPER_MAX_PROBE
+                                    ? SUPER_MAX_PROBE
+                                    : super_reg_effective_size();
 
         // Linear probing to find matching entry
-        for (int i = 0; i < SUPER_MAX_PROBE; i++) {
+        for (int i = 0; i < probe_limit; i++) {
-            SuperRegEntry* e = &g_super_reg[(h + i) & SUPER_REG_MASK];
+            SuperRegEntry* e = &entries[(h + i) & mask];
 
             if (atomic_load_explicit(&e->base, memory_order_acquire) == base && e->lg_size == lg) {
                 // Found entry to remove
@@ -775,30 +810,37 @@ void hak_ss_prewarm_init(void) {
 void hak_super_registry_stats(SuperRegStats* stats) {
     if (!stats) return;
 
-    stats->total_slots = SUPER_REG_SIZE;
+    int eff_size = super_reg_effective_size();
+    int eff_mask = super_reg_effective_mask();
+    SuperRegEntry* reg = reg_entries();
+
+    stats->total_slots = eff_size;
     stats->used_slots = 0;
     stats->max_probe_depth = 0;
+    if (!reg || eff_size <= 0) {
+        return;
+    }
 
     pthread_mutex_lock(&g_super_reg_lock);
 
     // Count used slots
-    for (int i = 0; i < SUPER_REG_SIZE; i++) {
+    for (int i = 0; i < eff_size; i++) {
-        if (atomic_load_explicit(&g_super_reg[i].base, memory_order_acquire) != 0) {
+        if (atomic_load_explicit(&reg[i].base, memory_order_acquire) != 0) {
             stats->used_slots++;
         }
     }
 
     // Calculate max probe depth
-    for (int i = 0; i < SUPER_REG_SIZE; i++) {
+    for (int i = 0; i < eff_size; i++) {
-        if (atomic_load_explicit(&g_super_reg[i].base, memory_order_acquire) != 0) {
+        if (atomic_load_explicit(&reg[i].base, memory_order_acquire) != 0) {
-            uintptr_t base = atomic_load_explicit(&g_super_reg[i].base, memory_order_acquire);
+            uintptr_t base = atomic_load_explicit(&reg[i].base, memory_order_acquire);
-            int lg = g_super_reg[i].lg_size;  // Phase 8.3: Use stored lg_size
+            int lg = reg[i].lg_size;  // Phase 8.3: Use stored lg_size
             int h = hak_super_hash(base, lg);
 
             // Find actual probe depth for this entry
             for (int j = 0; j < SUPER_MAX_PROBE; j++) {
-                int idx = (h + j) & SUPER_REG_MASK;
+                int idx = (h + j) & eff_mask;
-                if (atomic_load_explicit(&g_super_reg[idx].base, memory_order_acquire) == base && g_super_reg[idx].lg_size == lg) {
+                if (atomic_load_explicit(&reg[idx].base, memory_order_acquire) == base && reg[idx].lg_size == lg) {
                     if (j > stats->max_probe_depth) {
                         stats->max_probe_depth = j;
                     }

core/hakmem_super_registry.h

@@ -19,6 +19,7 @@
 #include <stdint.h>
 #include "hakmem_tiny_superslab.h"  // For SuperSlab and SUPERSLAB_MAGIC
 #include "box/ss_addr_map_box.h"    // Phase 9-1: O(1) hash table lookup
+#include "box/super_reg_box.h"      // Phase X: profile-aware logical registry sizing
 
 // Registry configuration
 // Increased from 4096 to 32768 to avoid registry exhaustion under
@@ -36,7 +37,7 @@
 #define SUPER_REG_PER_CLASS 16384  // Per-class registry capacity (increased for high-churn workloads)
 
 // Registry entry: base address → SuperSlab pointer mapping
-typedef struct {
+typedef struct SuperRegEntry {
     _Atomic(uintptr_t) base;  // Aligned base address (1MB or 2MB, 0 = empty slot) [atomic for proper sync]
     _Atomic(SuperSlab*) ss;   // Atomic SuperSlab pointer (MT-safe, prevents TOCTOU race)
     uint8_t   lg_size;        // Phase 8.3: ACE - SuperSlab size (20=1MB, 21=2MB)
@@ -44,7 +45,6 @@ typedef struct {
 } SuperRegEntry;
 
 // Global registry (lock-free reads, mutex-protected writes)
-extern SuperRegEntry g_super_reg[SUPER_REG_SIZE];
 extern pthread_mutex_t g_super_reg_lock;
 extern int g_super_reg_initialized;
 
@@ -56,7 +56,6 @@ extern int g_super_reg_initialized;
 #ifndef TINY_NUM_CLASSES
 #define TINY_NUM_CLASSES 8  // Fallback if hakmem_tiny.h not included yet
 #endif
-extern SuperSlab* g_super_reg_by_class[TINY_NUM_CLASSES][SUPER_REG_PER_CLASS];
 extern int g_super_reg_class_size[TINY_NUM_CLASSES];
 
 // ============================================================================
@@ -111,7 +110,7 @@ void hak_super_registry_init(void);
 // Hash function for aligned addresses (variable size)
 static inline int hak_super_hash(uintptr_t base, int lg_size) {
     // Phase 8.3: ACE - Variable size hash (lg_size = 20 for 1MB, 21 for 2MB)
-    return (int)((base >> lg_size) & SUPER_REG_MASK);
+    return (int)((base >> lg_size) & super_reg_effective_mask());
 }
 
 // Lookup SuperSlab by pointer (lock-free, thread-safe)
@@ -127,12 +126,18 @@ static inline SuperSlab* hak_super_lookup(void* ptr) {
     // Fallback: If hash map misses (e.g., map not populated yet), probe the
     // legacy registry table to avoid NULL for valid SuperSlabs.
     if (__builtin_expect(ss == NULL, 0)) {
+        SuperRegEntry* reg = super_reg_entries();
+        if (!reg) return NULL;
         uintptr_t p = (uintptr_t)ptr;
         for (int lg = SUPERSLAB_LG_MIN; lg <= SUPERSLAB_LG_MAX; lg++) {
             uintptr_t base = p & ~(((uintptr_t)1 << lg) - 1);
             int h = hak_super_hash(base, lg);
-            for (int i = 0; i < SUPER_MAX_PROBE; i++) {
+            int eff_mask = super_reg_effective_mask();
-                SuperRegEntry* e = &g_super_reg[(h + i) & SUPER_REG_MASK];
+            int probe_limit = super_reg_effective_size() > SUPER_MAX_PROBE
+                                  ? SUPER_MAX_PROBE
+                                  : super_reg_effective_size();
+            for (int i = 0; i < probe_limit; i++) {
+                SuperRegEntry* e = &reg[(h + i) & eff_mask];
                 uintptr_t reg_base = atomic_load_explicit(&e->base, memory_order_acquire);
                 if (reg_base == 0) {
                     break;  // empty slot

core/hakmem_tiny.c

@@ -26,6 +26,7 @@
 #include "tiny_tls_guard.h"
 #include "tiny_ready.h"
 #include "box/c7_meta_used_counter_box.h"
+#include "box/super_reg_box.h"
 #include "hakmem_tiny_tls_list.h"
 #include "hakmem_tiny_remote_target.h" // Phase 2C-1: Remote target queue
 #include "hakmem_tiny_bg_spill.h"      // Phase 2C-2: Background spill queue
@@ -124,7 +125,7 @@ static void* __attribute__((cold, noinline)) hak_tiny_alloc_slow(size_t size, in
 // Box: adopt_gate_try (implementation moved from header for robust linkage)
 // --------------------------------------------------------------------------- 
 #include "box/adopt_gate_box.h"
-extern SuperSlab* g_super_reg_by_class[TINY_NUM_CLASSES][SUPER_REG_PER_CLASS];
+#include "box/super_reg_box.h"
 extern int g_super_reg_class_size[TINY_NUM_CLASSES];
 extern unsigned long long g_adopt_gate_calls[];
 extern unsigned long long g_adopt_gate_success[];
@@ -137,6 +138,10 @@ SuperSlab* adopt_gate_try(int class_idx, TinyTLSSlab* tls) {
     if (ss) { g_adopt_gate_success[class_idx]++; return ss; }
     g_reg_scan_attempts[class_idx]++;
     int reg_size = g_super_reg_class_size[class_idx];
+    int reg_cap = super_reg_effective_per_class();
+    if (reg_cap > 0 && reg_size > reg_cap) {
+        reg_size = reg_cap;
+    }
     int scan_limit = tiny_reg_scan_max();
     if (scan_limit > reg_size) scan_limit = reg_size;
     uint32_t self_tid = tiny_self_u32();
@@ -156,7 +161,7 @@ SuperSlab* adopt_gate_try(int class_idx, TinyTLSSlab* tls) {
     }
 
     for (int i = 0; i < scan_limit; i++) {
-        SuperSlab* cand = g_super_reg_by_class[class_idx][i];
+        SuperSlab* cand = super_reg_by_class_at(class_idx, i);
         if (!(cand && cand->magic == SUPERSLAB_MAGIC)) continue;
         // Fast path: use nonempty_mask / freelist_mask to locate candidates in O(1)
         uint32_t mask = cand->nonempty_mask;

core/hakmem_tiny_lifecycle.inc

@@ -12,6 +12,7 @@
 // Cold/maintenance path - not performance critical.
 #include "tiny_tls_guard.h"
 #include "box/ss_slab_meta_box.h"  // Phase 3d-A: SlabMeta Box boundary
+#include "hakmem_super_registry.h"
 
 // Phase 12: Helper to derive a representative class index for a SuperSlab
 // from per-slab metadata (all slabs are empty when used in trim).
@@ -96,8 +97,11 @@ void hak_tiny_trim(void) {
     }
 
     // Walk the registry and collect empty SuperSlabs by class
-    for (int i = 0; i < SUPER_REG_SIZE; i++) {
+    SuperRegEntry* reg = super_reg_entries();
-        SuperRegEntry* e = &g_super_reg[i];
+    int reg_cap = super_reg_effective_size();
+    if (!reg || reg_cap <= 0) return;
+    for (int i = 0; i < reg_cap; i++) {
+        SuperRegEntry* e = &reg[i];
         uintptr_t base = atomic_load_explicit((_Atomic uintptr_t*)&e->base, memory_order_acquire);
         if (base == 0) continue;
         SuperSlab* ss = e->ss;

core/hakmem_tiny_magazine.c

@@ -7,6 +7,7 @@
 #include "hakmem_prof.h"
 #include "hakmem_internal.h"
 #include "box/tiny_next_ptr_box.h"  // Box API: Next pointer read/write
+#include "box/tiny_mem_stats_box.h"
 #include <pthread.h>
 
 static inline uint32_t tiny_self_u32_guard(void) {
@@ -36,6 +37,14 @@ int g_mag_cap_limit = TINY_TLS_MAG_CAP;
 int g_mag_cap_override[TINY_NUM_CLASSES] = {0};     // HAKMEM_TINY_MAG_CAP_C{0..7}
 
 __thread int g_tls_small_mags_inited = 0;
+static __thread int g_tls_mag_mem_recorded = 0;
+
+static inline void tiny_mag_record_mem_once(void) {
+    if (!g_tls_mag_mem_recorded) {
+        tiny_mem_stats_add_tls_magazine((ssize_t)sizeof(g_tls_mags));
+        g_tls_mag_mem_recorded = 1;
+    }
+}
 
 // tiny_default_cap() and tiny_cap_max_for_class() now defined as inline functions
 // in hakmem_tiny_config.h for centralized configuration
@@ -49,6 +58,7 @@ int tiny_effective_cap(int class_idx) {
 
 void tiny_small_mags_init_once(void) {
     if (__builtin_expect(g_tls_small_mags_inited, 1)) return;
+    tiny_mag_record_mem_once();
     for (int k = 0; k <= 3; k++) {
         TinyTLSMag* m = &g_tls_mags[k];
         if (m->cap == 0) {
@@ -65,6 +75,7 @@ void tiny_small_mags_init_once(void) {
 void tiny_mag_init_if_needed(int class_idx) {
     TinyTLSMag* mag = &g_tls_mags[class_idx];
     if (mag->cap == 0) {
+        tiny_mag_record_mem_once();
         int base = tiny_effective_cap(class_idx);
         int cap = (base < TINY_TLS_MAG_CAP) ? base : TINY_TLS_MAG_CAP;
         if (g_mag_cap_limit < cap) cap = g_mag_cap_limit;

core/hakmem_tiny_publish_box.inc

@@ -7,7 +7,7 @@
 
 // Tiny Page Box: C5〜C7 用 Tiny-Plus page pool（Superslab/Warm Pool より前段の箱）
 // tiny_tls_bind_slab() で新しい TLS Slab が bind されたタイミングで
-// tiny_page_box_on_new_slab() を呼び出し、Page Box 側の page pool を更新する。
+// tiny_page_box_on_new_slab(class_idx, tls) を呼び出し、Page Box 側の page pool を更新する。
 #include "box/tiny_page_box.h"
 
 // Mailbox box
@@ -369,8 +369,9 @@ static inline void tiny_tls_bind_slab(TinyTLSSlab* tls, SuperSlab* ss, int slab_
     tls->meta = &ss->slabs[slab_idx];
     tls->slab_base = tiny_slab_base_for(ss, slab_idx);
 
-    // Tiny Page Box にも新しい slab を通知しておく（C7 など有効クラスのみ）
+    // Tiny Page Box にも新しい slab を通知しておく（有効クラスのみ）
-    tiny_page_box_on_new_slab(tls);
+    int pb_class = tls->meta ? (int)tls->meta->class_idx : -1;
+    tiny_page_box_on_new_slab(pb_class, tls);
 }
 
 static inline uint32_t tiny_tls_default_refill(uint32_t cap) {

core/superslab_ace.c

@@ -4,6 +4,7 @@
 // Date: 2025-11-28
 
 #include "hakmem_tiny_superslab_internal.h"
+#include "hakmem_super_registry.h"
 
 // ============================================================================
 // ACE (Adaptive Cache Engine) State
@@ -140,8 +141,12 @@ void ace_observe_and_decide(int k) {
     int ss_count = 0;
     uint32_t total_live = 0;
 
-    for (int i = 0; i < SUPER_REG_SIZE; i++) {
+    SuperRegEntry* reg = super_reg_entries();
-        SuperRegEntry* e = &g_super_reg[i];
+    int reg_cap = super_reg_effective_size();
+    if (!reg || reg_cap <= 0) return;
+
+    for (int i = 0; i < reg_cap; i++) {
+        SuperRegEntry* e = &reg[i];
 
         // Atomic read (thread-safe)
         uintptr_t base = atomic_load_explicit(

core/superslab_stats.c

@@ -4,6 +4,7 @@
 // Date: 2025-11-28
 
 #include "hakmem_tiny_superslab_internal.h"
+#include <stdlib.h>
 
 // ============================================================================
 // Global Statistics
@@ -30,6 +31,11 @@ uint64_t g_ss_freed_by_class[8] = {0};
 _Atomic uint64_t g_ss_mmap_count = 0;
 _Atomic uint64_t g_final_fallback_mmap_count = 0;
 
+// Superslab/slab observability (Tiny-only; relaxed updates)
+_Atomic uint64_t g_ss_live_by_class[8]   = {0};
+_Atomic uint64_t g_ss_empty_events[8]    = {0};
+_Atomic uint64_t g_slab_live_events[8]   = {0};
+
 // ============================================================================
 // Statistics Functions
 // ============================================================================
@@ -56,6 +62,35 @@ void ss_stats_cache_store(void) {
     pthread_mutex_unlock(&g_superslab_lock);
 }
 
+void ss_stats_on_ss_alloc_class(int class_idx) {
+    if (class_idx >= 0 && class_idx < 8) {
+        atomic_fetch_add_explicit(&g_ss_live_by_class[class_idx], 1, memory_order_relaxed);
+    }
+}
+
+void ss_stats_on_ss_free_class(int class_idx) {
+    if (class_idx >= 0 && class_idx < 8) {
+        uint64_t prev = atomic_load_explicit(&g_ss_live_by_class[class_idx], memory_order_relaxed);
+        if (prev > 0) {
+            atomic_fetch_sub_explicit(&g_ss_live_by_class[class_idx], 1, memory_order_relaxed);
+        }
+    }
+}
+
+void ss_stats_on_ss_scan(int class_idx, int slab_live, int is_empty) {
+    if (class_idx < 0 || class_idx >= 8) {
+        return;
+    }
+    if (slab_live > 0) {
+        atomic_fetch_add_explicit(&g_slab_live_events[class_idx],
+                                  (uint64_t)slab_live,
+                                  memory_order_relaxed);
+    }
+    if (is_empty) {
+        atomic_fetch_add_explicit(&g_ss_empty_events[class_idx], 1, memory_order_relaxed);
+    }
+}
+
 // ============================================================================
 // Diagnostics
 // ============================================================================
@@ -164,3 +199,23 @@ void superslab_print_global_stats(void) {
     printf("Total bytes allocated: %lu MB\n", g_bytes_allocated / (1024 * 1024));
     pthread_mutex_unlock(&g_superslab_lock);
 }
+
+void ss_stats_dump_if_requested(void) {
+    const char* env = getenv("HAKMEM_SS_STATS_DUMP");
+    if (!env || !*env || *env == '0') {
+        return;
+    }
+    fprintf(stderr, "[SS_STATS] class live empty_events slab_live_events\n");
+    for (int c = 0; c < 8; c++) {
+        uint64_t live = atomic_load_explicit(&g_ss_live_by_class[c], memory_order_relaxed);
+        uint64_t empty = atomic_load_explicit(&g_ss_empty_events[c], memory_order_relaxed);
+        uint64_t slab_live = atomic_load_explicit(&g_slab_live_events[c], memory_order_relaxed);
+        if (live || empty || slab_live) {
+            fprintf(stderr, "  C%d: live=%llu empty=%llu slab_live=%llu\n",
+                    c,
+                    (unsigned long long)live,
+                    (unsigned long long)empty,
+                    (unsigned long long)slab_live);
+        }
+    }
+}

core/tiny_alloc_fast_push.d

@@ -17,8 +17,9 @@ core/tiny_alloc_fast_push.o: core/tiny_alloc_fast_push.c \
  core/box/../superslab/../tiny_box_geometry.h \
  core/box/../tiny_debug_ring.h core/box/../tiny_remote.h \
  core/box/../box/ss_addr_map_box.h \
- core/box/../box/../hakmem_build_flags.h core/box/../hakmem_tiny.h \
+ core/box/../box/../hakmem_build_flags.h core/box/../box/super_reg_box.h \
- core/box/../hakmem_trace.h core/box/../hakmem_tiny_mini_mag.h \
+ core/box/../hakmem_tiny.h core/box/../hakmem_trace.h \
+ core/box/../hakmem_tiny_mini_mag.h \
  core/box/../box/hak_lane_classify.inc.h core/box/../tiny_debug_api.h \
  core/box/../hakmem_tiny_integrity.h core/box/../ptr_track.h \
  core/box/../ptr_trace.h core/box/../hakmem_trace_master.h \
@@ -68,6 +69,7 @@ core/box/../tiny_debug_ring.h:
 core/box/../tiny_remote.h:
 core/box/../box/ss_addr_map_box.h:
 core/box/../box/../hakmem_build_flags.h:
+core/box/../box/super_reg_box.h:
 core/box/../hakmem_tiny.h:
 core/box/../hakmem_trace.h:
 core/box/../hakmem_tiny_mini_mag.h:

core/tiny_remote.c

@@ -10,18 +10,11 @@
 #endif
 #include <string.h>
 #include "tiny_remote.h"
+#include "box/remote_side_box.h"
 #include "hakmem_tiny_superslab.h"
 #include "tiny_debug_ring.h"
 
-#define REM_SIDE_LOG2 20
+static rem_side_entry* g_rem_side = NULL;
-#define REM_SIDE_SIZE (1u<<REM_SIDE_LOG2)
-
-typedef struct {
-    _Atomic(uintptr_t) key;   // node pointer
-    _Atomic(uintptr_t) val;   // next pointer
-} rem_side_entry;
-
-static rem_side_entry g_rem_side[REM_SIDE_SIZE];
 int g_remote_side_enable = 1; // default ON; can be disabled via env or 1T hint
 extern int g_debug_remote_guard;
 static _Atomic int g_remote_scribble_once = 0;
@@ -32,6 +25,21 @@ static inline uint32_t hmix(uintptr_t v);
 static inline uint32_t tiny_remote_stage_hash(const char* stage);
 static void tiny_remote_dump_backtrace(void);
 
+static inline uint32_t rem_side_mask(void) {
+    return remote_side_effective_mask();
+}
+
+static inline uint32_t rem_side_size(void) {
+    return remote_side_effective_size();
+}
+
+static inline rem_side_entry* rem_side_table_local(void) {
+    if (__builtin_expect(g_rem_side == NULL, 0)) {
+        g_rem_side = remote_side_table();
+    }
+    return g_rem_side;
+}
+
 #if !HAKMEM_BUILD_RELEASE
 #define REM_TRACK_TABLE_LOG2 20
 #define REM_TRACK_TABLE_SIZE (1u << REM_TRACK_TABLE_LOG2)
@@ -536,14 +544,18 @@ uint32_t tiny_remote_drain_threshold(void) {
 void tiny_remote_side_set(struct SuperSlab* ss, int slab_idx, void* node, uintptr_t next) {
     (void)ss; (void)slab_idx;
     if (!g_remote_side_enable) return;
+    rem_side_entry* table = rem_side_table_local();
+    if (!table) return;
     uintptr_t k = (uintptr_t)node;
     uintptr_t base = (uintptr_t)ss;
     size_t ss_size = (size_t)1ULL << ss->lg_size;
-    uint32_t i = hmix(k) & (REM_SIDE_SIZE - 1);
+    uint32_t mask = rem_side_mask();
-    for (uint32_t n=0; n<REM_SIDE_SIZE; n++, i=(i+1)&(REM_SIDE_SIZE-1)) {
+    uint32_t size = rem_side_size();
+    uint32_t i = hmix(k) & mask;
+    for (uint32_t n=0; n<size; n++, i=(i+1)&mask) {
         uintptr_t expect = 0;
-        if (atomic_compare_exchange_weak_explicit(&g_rem_side[i].key, &expect, k, memory_order_acq_rel, memory_order_relaxed)) {
+        if (atomic_compare_exchange_weak_explicit(&table[i].key, &expect, k, memory_order_acq_rel, memory_order_relaxed)) {
-            atomic_store_explicit(&g_rem_side[i].val, next, memory_order_release);
+            atomic_store_explicit(&table[i].val, next, memory_order_release);
             tiny_remote_sentinel_set(node);
             tiny_remote_watch_note("side_set", ss, slab_idx, node, 0xA233u, 0, 0);
             return;
@@ -583,12 +595,16 @@ void tiny_remote_side_set(struct SuperSlab* ss, int slab_idx, void* node, uintpt
 uintptr_t tiny_remote_side_get(struct SuperSlab* ss, int slab_idx, void* node) {
     (void)ss; (void)slab_idx;
     (void)g_remote_side_enable; // always true in caller
+    rem_side_entry* table = rem_side_table_local();
+    if (!table) return 0;
     uintptr_t k = (uintptr_t)node;
-    uint32_t i = hmix(k) & (REM_SIDE_SIZE - 1);
+    uint32_t mask = rem_side_mask();
-    for (uint32_t n=0; n<REM_SIDE_SIZE; n++, i=(i+1)&(REM_SIDE_SIZE-1)) {
+    uint32_t size = rem_side_size();
-        uintptr_t key = atomic_load_explicit(&g_rem_side[i].key, memory_order_acquire);
+    uint32_t i = hmix(k) & mask;
+    for (uint32_t n=0; n<size; n++, i=(i+1)&mask) {
+        uintptr_t key = atomic_load_explicit(&table[i].key, memory_order_acquire);
         if (key == k) {
-            return atomic_load_explicit(&g_rem_side[i].val, memory_order_acquire);
+            return atomic_load_explicit(&table[i].val, memory_order_acquire);
         }
         if (key == 0) break;
     }
@@ -606,13 +622,17 @@ uintptr_t tiny_remote_side_get(struct SuperSlab* ss, int slab_idx, void* node) {
 void tiny_remote_side_clear(struct SuperSlab* ss, int slab_idx, void* node) {
     (void)ss; (void)slab_idx;
     if (!g_remote_side_enable) return;
+    rem_side_entry* table = rem_side_table_local();
+    if (!table) return;
     uintptr_t k = (uintptr_t)node;
-    uint32_t i = hmix(k) & (REM_SIDE_SIZE - 1);
+    uint32_t mask = rem_side_mask();
-    for (uint32_t n = 0; n < REM_SIDE_SIZE; n++, i = (i + 1) & (REM_SIDE_SIZE - 1)) {
+    uint32_t size = rem_side_size();
-        uintptr_t key = atomic_load_explicit(&g_rem_side[i].key, memory_order_acquire);
+    uint32_t i = hmix(k) & mask;
+    for (uint32_t n = 0; n < size; n++, i = (i + 1) & mask) {
+        uintptr_t key = atomic_load_explicit(&table[i].key, memory_order_acquire);
         if (key == k) {
-            atomic_store_explicit(&g_rem_side[i].val, 0, memory_order_relaxed);
+            atomic_store_explicit(&table[i].val, 0, memory_order_relaxed);
-            atomic_store_explicit(&g_rem_side[i].key, 0, memory_order_release);
+            atomic_store_explicit(&table[i].key, 0, memory_order_release);
             tiny_remote_watch_clear(node);
             return;
         }
@@ -623,10 +643,14 @@ void tiny_remote_side_clear(struct SuperSlab* ss, int slab_idx, void* node) {
 int tiny_remote_side_contains(struct SuperSlab* ss, int slab_idx, void* node) {
     (void)ss; (void)slab_idx;
     if (!g_remote_side_enable) return 0;
+    rem_side_entry* table = rem_side_table_local();
+    if (!table) return 0;
     uintptr_t k = (uintptr_t)node;
-    uint32_t i = hmix(k) & (REM_SIDE_SIZE - 1);
+    uint32_t mask = rem_side_mask();
-    for (uint32_t n = 0; n < REM_SIDE_SIZE; n++, i = (i + 1) & (REM_SIDE_SIZE - 1)) {
+    uint32_t size = rem_side_size();
-        uintptr_t key = atomic_load_explicit(&g_rem_side[i].key, memory_order_acquire);
+    uint32_t i = hmix(k) & mask;
+    for (uint32_t n = 0; n < size; n++, i = (i + 1) & mask) {
+        uintptr_t key = atomic_load_explicit(&table[i].key, memory_order_acquire);
         if (key == k) {
             return 1;
         }
@@ -639,6 +663,7 @@ void tiny_remote_side_init_from_env(void) {
     static int g_side_init_once = 0;
     if (__builtin_expect(g_side_init_once, 0)) return;
     g_side_init_once = 1;
+    remote_side_init(NULL, NULL);
     const char* side_env = getenv("HAKMEM_TINY_REMOTE_SIDE");
     int enable = 1;
     if (side_env && *side_env) {
@@ -658,8 +683,12 @@ void tiny_remote_side_init_from_env(void) {
         fprintf(stderr, "[REMOTE_SIDE_INIT] enable=%d\n", enable);
     }
     if (!enable) return;
-    for (uint32_t i = 0; i < REM_SIDE_SIZE; i++) {
+    g_rem_side = remote_side_table();
-        atomic_store_explicit(&g_rem_side[i].key, 0, memory_order_relaxed);
+    rem_side_entry* table = rem_side_table_local();
-        atomic_store_explicit(&g_rem_side[i].val, 0, memory_order_relaxed);
+    if (!table) return;
+    uint32_t size = rem_side_size();
+    for (uint32_t i = 0; i < size; i++) {
+        atomic_store_explicit(&table[i].key, 0, memory_order_relaxed);
+        atomic_store_explicit(&table[i].val, 0, memory_order_relaxed);
     }
 }

docs/analysis/C7_FREE_HOTPATH.md

@@ -0,0 +1,35 @@
+C7 Free Hotpath (design memo)
+=============================
+
+Goals
+-----
+- Flatten the dominant C7 free path to minimise branches and helper hops.
+- Keep safety checks boxed; keep hot lane minimal.
+
+Current typical path (C7)
+-------------------------
+1. size→class LUT → `class_idx = 7`.
+2. free gate / route box decides Tiny vs Pool.
+3. Tiny free fast v2:
+   - Policy/env checks,
+   - TLS SLL push,
+   - Warm/UC interaction as needed.
+4. Multiple helper calls along the way (gate, policy, sll push).
+
+Target hot lane
+---------------
+1. Single policy snapshot for C7 (warm/page/tls on).
+2. Straight to TLS SLL push with minimal bookkeeping.
+3. Optional UC/Warm stats only in sampled mode.
+4. Rare branches (remote/free-list edge cases) stay in boxed slow path.
+
+Ideas to explore
+----------------
+- Add `hak_tiny_free_fast_v2_c7()` inline used when `class_idx==7`.
+- Fold gate/policy reads into one branch per free call.
+- Keep TLS SLL push inline, push remote/cross-thread cases behind unlikely branches.
+
+Validation
+----------
+- Compare C7-only ops/s before/after.
+- Ensure remote/free-list invariants stay enforced in the slow path.

docs/analysis/C7_HOTPATH_FLATTENING.md

@@ -0,0 +1,39 @@
+C7 Alloc Hotpath Flattening (design memo)
+=========================================
+
+Goals
+-----
+- Make C7 alloc as close to a straight line as possible.
+- Minimise branches/indirections on the steady hit path (UC/TLS/Warm already stable).
+- Keep Box boundaries intact; isolate feature gates to one lookup.
+
+Current shape (simplified)
+--------------------------
+1. size→class LUT → `class_idx = 7` for 1024B path.
+2. Route/Policy checks (tiny_route_get, tiny_policy_get) → gate UC/Warm/Page.
+3. UC pop: hit path shares code with miss/refill, includes stats/guards.
+4. TLS/Warm engagement happens behind UC miss boundary.
+5. Multiple helper calls on the hit path (gate box, policy box, UC helpers).
+
+Target shape
+------------
+1. size→class LUT (unchanged).
+2. One policy snapshot: `const TinyClassPolicy* pol = tiny_policy_get(7);`
+3. One route decision: C7 fast path assumes Tiny→UC→TLS/Warm enabled.
+4. Hit path specialised:
+   - Inline `tiny_unified_cache_pop_fast_c7()` that only touches the hot cache lines.
+   - Stats optional/sampled (avoid atomic on every hit).
+   - No feature/env reads.
+5. Miss path remains boxed and guarded; enters existing refill flow unchanged.
+
+Possible refactors
+------------------
+- Add `malloc_tiny_fast_c7_inline(...)` as a static inline used only when class==7.
+- Precompute `pol->warm_enabled/page_box_enabled` once per thread and reuse.
+- Split UC helpers into `*_hit_fast` vs `*_miss` to keep the hit CFG tiny.
+
+Trade-offs / checks
+-------------------
+- Keep the Box boundaries (Gate/Route/Policy) but allow an inline “fast lane” for C7.
+- Ensure Debug/Policy logging stays in the slow/miss path only.
+- Validate with IPC/ops after implementation; target +10–15% for C7-heavy mixes.

docs/analysis/CPU_HOTPATH_OVERVIEW.md

@@ -0,0 +1,53 @@
+CPU Hotpath Overview (bench profile)
+====================================
+
+Context
+-------
+- Build/profile: `HAKMEM_PROFILE=bench`, `HAKMEM_TINY_PROFILE=full`, `HAKMEM_WARM_TLS_BIND_C7=2`.
+- Workloads sampled:
+  - 16–1024B (`./bench_random_mixed_hakmem 1000000 256 42`)
+  - 129–1024B (`HAKMEM_BENCH_MIN_SIZE=129 HAKMEM_BENCH_MAX_SIZE=1024 ./bench_random_mixed_hakmem 1000000 256 42`)
+- Target: identify user‑space hot spots to guide C7 flattening work.
+
+Sampling attempt (perf)
+-----------------------
+- `perf record -g -e cycles:u` and `perf record -g -e cpu-clock:u` both fall back to `page-faults` on this host (likely perf_event_paranoid). The captures show:
+  - ~97% of page-fault samples in `__memset_avx2_unaligned_erms` during warmup/zeroing.
+  - Callers were `tiny_tls_sll_drain.part.0.constprop.0` and `adaptive_sizing_init` (warmup path).
+- No steady‑state cycle profile was available without elevated perf permissions. `perf data` removed after inspection (`rm perf.data`) to keep tree clean.
+
+What we can infer despite the limitation
+----------------------------------------
+- Warmup zeroing dominates page‑fault samples; steady‑state alloc/free is not represented.
+- Hot candidates for the next pass (from previous code inspection and bench intuition):
+  - `tiny_alloc_fast` / `malloc_tiny_fast` (C7 fast path)
+  - `hak_tiny_free_fast_v2`
+  - `tiny_unified_cache` hit path helpers
+  - `tls_sll_pop_impl` / `tiny_tls_sll_drain`
+
+Next measurement options
+------------------------
+- If perf cycles are still blocked:
+  - Use `perf stat -e cycles,instructions,branches,branch-misses -r 5 -- ...` to get aggregate IPC per workload.
+  - Add temporary user‑space counters (Box‑guarded) around C7 alloc/free hot sections to estimate per‑op cycles.
+  - Run perf with elevated permissions or lower `perf_event_paranoid` if available.
+
+Aggregate perf stat snapshot (bench profile)
+--------------------------------------------
+- Env: `HAKMEM_PROFILE=bench HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2`
+- Workloads (Release, 3× perf stat):
+  - 16–1024B: cycles≈109.8M, instructions≈233.2M → IPC≈2.12, branches≈49.4M, branch-miss≈2.89%
+  - 129–1024B: cycles≈109.6M, instructions≈230.3M → IPC≈2.10, branches≈48.8M, branch-miss≈2.90%
+  - 16–1024B with `HAKMEM_TINY_C7_HOT=1` (UC hit-only + flat TLS→UC→cold):
+    - cycles≈111.8M, instructions≈242.1M → IPC≈2.16, branches≈52.0M, branch-miss≈2.75%
+    - RSS≈7.1MB; throughput ≈47.4–47.6M ops/s (hot=1) vs ≈47.2M (hot=0) on the same runset.
+
+Action items flowing from this note
+-----------------------------------
+- Proceed with design notes for C7 alloc/free flattening and UC hit simplification based on code structure.
+- Keep warmup zeroing out of the steady‑state loop when profiling (consider `HAKMEM_BENCH_FAST_MODE` for future captures).
+
+Conclusion (current state)
+--------------------------
+- `HAKMEM_TINY_C7_HOT` は実験用フラグとして残し、デフォルト OFF のまま運用する。ON にしても branch-miss はわずかに改善する程度で、ops/s は同等〜微減。
+- ひとまず「安全＋そこそこ速い」現行経路を基準とし、さらなるフラット化は別途必要性を見て検討する。***

docs/analysis/LARGE_GLOBALS_OVERVIEW.md

@@ -0,0 +1,84 @@
+LARGE_GLOBALS_OVERVIEW
+======================
+
+概要
+----
+- `nm -S --size-sort bench_random_mixed_hakmem` で確認した巨大 BSS/静的領域の上位シンボルをメモ。
+- 目視での役割・要素数イメージと、直近の SS_STATS（短い run, ws=64, iters=10k, HAKMEM_SS_STATS_DUMP=1）のギャップを併記。
+- 目的: 次フェーズの「SuperReg/SharedPool/Remote を動的化 or 縮小」設計の入力にする。
+
+コマンド
+--------
+```bash
+nm -S --size-sort bench_random_mixed_hakmem | tail -n 120
+HAKMEM_SS_STATS_DUMP=1 ./bench_random_mixed_hakmem 10000 64 1 2> /tmp/ss_stats_sample.log
+```
+
+観測された大きめシンボル
+------------------------
+
+| Symbol | Size | 役割/箱 | 備考・ギャップ |
+| --- | --- | --- | --- |
+| `g_super_reg` | 0x1800000 ≈ 24.0 MB | Super Registry 全体 | SS_STATS では C2=1, C7=1 live と極小。大半が未使用の固定配列。 |
+| `g_rem_side` | 0x1000000 ≈ 16.0 MB | Remote Queue 側バッファ | スレッド数・ノード数に対してオーバーサイズ。bench ではほぼ未使用。 |
+| `g_shared_pool` | 0x238140 ≈ 2.23 MB | Shared Pool テーブル | live SS 2 枚に対し容量が大きい。class 別縮小余地あり。 |
+| `g_super_reg_by_class` | 0x100000 ≈ 1.0 MB | クラス別 SuperReg インデックス | クラス数 8 に対し 1MB 固定。動的化で圧縮可能。 |
+| `g_free_node_pool` | 0xC0000 ≈ 0.75 MB | Free ノードプール | Remote/Pool 用。小さくはないが上位ほどではない。 |
+| `g_mf2_page_registry.lto_priv.0` | 0x82810 ≈ 0.51 MB | MF2 ページレジストリ | MF2 経路用。 |
+| `g_tls_mags` | 0x40040 ≈ 0.26 MB | TLS Magazine 配列 | スレッド数ぶん前提。実使用は少数。 |
+| `g_site_rules` | 0x40040 ≈ 0.26 MB | Site rule テーブル | 固定長。 |
+| `g_mid_desc_mu` | 0x14000 ≈ 80 KB | Mid-size desc | 中規模。 |
+| `g_mid_tc_mu` | 0xA000 ≈ 40 KB | Mid-size TC | 中規模。 |
+| `g_pool.lto_priv.0` | 0x9680 ≈ 37 KB | Pool 配列 | 中規模。 |
+| `g_tiny_page_box` | 0xC40 ≈ 3.1 KB | Tiny Page Box 配列 | Tiny Front 系。微小。 |
+| `g_tls_hot_mag` | 0x2040 ≈ 8 KB | TLS Hot Magazine | 微小。 |
+| `g_fast_cache` | 0x2200 ≈ 8.6 KB | Fast cache | 微小。 |
+
+補足（SS_STATS サンプル）
+-------------------------
+- 短い run（ws=64, iters=10k, Release, HAKMEM_SS_STATS_DUMP=1）の結果:
+  - `[SS_STATS] class live empty_events slab_live_events`
+  - `C2: live=1 empty=0 slab_live=0`
+  - `C7: live=1 empty=1 slab_live=0`
+  - `[RSS] max_kb=29568`
+- 「巨大配列の容量」に対し「実際に live の Superslab」は 2 枚のみ。固定長 BSS が RSS を支配していることが確実。
+
+定義元と役割（コード位置）
+--------------------------
+- Super Registry (`core/hakmem_super_registry.{h,c}`)
+  - `g_super_reg[SUPER_REG_SIZE]` … ハッシュ登録（デフォルト 1,048,576 エントリ = 24MB、`SUPER_REG_SIZE` で調整可能）
+  - `g_super_reg_by_class[TINY_NUM_CLASSES][SUPER_REG_PER_CLASS]` … クラス別スキャン用（デフォルト 8×16384 = 128K スロット ≈1MB）
+  - `g_super_reg_class_size[]` … クラス別 live カウント
+  - `g_ss_lru_cache` … LRU 再利用キャッシュ（メモリは小さめ）
+- Shared Pool (`core/hakmem_shared_pool.{h,c}` + `_acquire.c` + `_release.c`)
+  - `g_shared_pool` … Superslab 配列、クラス別ヒント/活性/フリーリスト/メタ配列を同居させた大きめ struct（≈2.3MB）
+  - `g_shared_pool.ss_metadata[]` … Superslab ごとのメタデータ配列
+- Remote Queue (`core/tiny_remote.c`)
+  - `g_rem_side[REM_SIDE_SIZE]` … cross-thread free のハッシュ（`REM_SIDE_LOG2=20` → 1M エントリ ≈16MB）
+  - Debug 時の `g_rem_track[]` は release では落ちるのでサイズ影響なし
+- Free Node Pool (`core/pool_refill.c` など)
+  - `g_free_node_pool` … pool refilling 用のノードストック（≈0.75MB）
+- TLS / MF2 系
+  - `g_tls_mags` (`core/hakmem_tiny_magazine.c`) … TLS マガジン配列（≈0.26MB、スレッド数前提）
+  - `g_mf2_page_registry` (`core/mf2*`) … MF2 併用時のページレジストリ（≈0.5MB）
+  - `g_ss_addr_map` (`core/box/ss_addr_map_box.h`) … Superslab アドレス検索ハッシュ（サイズは中程度）
+
+ベンチ向け縮小の目安（案）
+--------------------------
+- SuperReg
+  - 現状: `SUPER_REG_SIZE=1,048,576`（24MB）、`SUPER_REG_PER_CLASS=16384`（1MB）
+  - Bench 目安: `SUPER_REG_SIZE_BENCH=65,536`（~1.5MB）、`SUPER_REG_PER_CLASS_BENCH=1024`（~64KB）
+- Shared Pool
+  - 現状: capacity は動的拡張だが初期サイズは大きめ（約 2.3MB）
+  - Bench 目安: 初期 capacity を 64〜128 に抑え、クラス別スロットも縮小
+- Remote Queue
+  - 現状: `REM_SIDE_LOG2=20` → 1M エントリ（16MB）
+  - Bench 目安: `REM_SIDE_LOG2=16`（64K エントリ ≈1MB）程度まで削減
+- Free Node Pool / TLS Mag / MF2
+  - Bench ではスレッド数や MF2 オンオフに応じて「初期化を遅延」「固定配列を半減」する余地あり。
+
+次の設計ステップ（Box 化の方向性）
+-----------------------------------
+- SuperReg/SharedPool/Remote を Box 化し、`HAKMEM_PROFILE`（prod/full/bench/larson_guard 等）で容量を切替できるようにする。
+- Bench 用の小型プロファイル（registry/pool/remote を 1/4〜1/8）を追加し、RSS を抑えた状態で mimalloc/system と比較する。
+- Superslab Budget Box と組み合わせ、live 枚数上限（予算）と「空 SS 再利用ポリシー」を分離して管理する。***

docs/analysis/SUPERSLAB_STATS_SNAPSHOT.md

@@ -0,0 +1,40 @@
+# Superslab Stats Snapshot (larson_guard, 2025-12-06)
+
+コマンド:  
+`HAKMEM_TINY_PROFILE=larson_guard HAKMEM_SS_STATS_DUMP=1 ./bench_allocators_hakmem larson 1 10000 1`
+
+抜粋ログ:
+```
+[SS_STATS] class live empty_events slab_live_events
+  C2: live=1 empty=0 slab_live=0
+```
+
+メモ: larson_guard では Superslab 枚数が予算近辺で頭打ちになり、暴走せずに完走することを確認。
+
+# Superslab Stats Snapshot (bench profile, 2025-12-06)
+
+コマンド:  
+`HAKMEM_PROFILE=bench HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2 HAKMEM_SS_STATS_DUMP=1 ./bench_random_mixed_hakmem 1000000 256 42`
+
+抜粋ログ:
+```
+[SS_STATS] class live empty_events slab_live_events
+  C2: live=1 empty=0 slab_live=0
+  C7: live=1 empty=1 slab_live=0
+[RSS] max_kb=7168
+```
+
+メモ: bench プロファイル（SuperReg/Remote 実配列縮小版）でも live Superslab は C2=1, C7=1 に収まり、RSS は ~7MB まで低減。***
+
+# Tiny Mem Stats Snapshot (bench profile, 2025-12-06)
+
+コマンド:  
+`HAKMEM_PROFILE=bench HAKMEM_TINY_PROFILE=full HAKMEM_WARM_TLS_BIND_C7=2 HAKMEM_TINY_MEM_DUMP=1 ./bench_random_mixed_hakmem 1000 8 1`
+
+抜粋ログ:
+```
+[TINY_MEM_STATS] unified_cache=36KB warm_pool=2KB page_box=3KB tls_mag=0KB policy_stats=0KB total=41KB
+[RSS] max_kb=7040
+```
+
+メモ: Tiny 層（UC/Warm/Page/TLS/Policy）だけなら数十 KB で、 bench プロファイルの RSS 低減は主に SuperReg/Remote の実配列縮小による。***

hakmem.d

@@ -16,12 +16,13 @@ hakmem.o: core/hakmem.c core/hakmem.h core/hakmem_build_flags.h \
  core/box/tiny_next_ptr_box.h core/hakmem_tiny_config.h \
  core/tiny_nextptr.h core/tiny_region_id.h core/tiny_box_geometry.h \
  core/ptr_track.h core/hakmem_super_registry.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/tiny_debug_api.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/box/tiny_layout_box.h core/box/../hakmem_tiny_config.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
- core/box/tiny_header_box.h core/box/tiny_layout_box.h \
+ core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
- core/box/../tiny_region_id.h core/hakmem_elo.h core/hakmem_ace_stats.h \
+ core/box/tiny_layout_box.h core/box/../tiny_region_id.h \
- core/hakmem_batch.h core/hakmem_evo.h core/hakmem_debug.h \
+ core/hakmem_elo.h core/hakmem_ace_stats.h core/hakmem_batch.h \
- core/hakmem_prof.h core/hakmem_syscall.h core/hakmem_ace_controller.h \
+ core/hakmem_evo.h core/hakmem_debug.h core/hakmem_prof.h \
+ core/hakmem_syscall.h core/hakmem_ace_controller.h \
  core/hakmem_ace_metrics.h core/hakmem_ace_ucb1.h \
  core/box/bench_fast_box.h core/ptr_trace.h core/hakmem_trace_master.h \
  core/hakmem_stats_master.h core/box/hak_kpi_util.inc.h \
@@ -86,6 +87,16 @@ hakmem.o: core/hakmem.c core/hakmem.h core/hakmem_build_flags.h \
  core/box/../front/../box/../front/tiny_unified_cache.h \
  core/box/../front/../box/tiny_layout_box.h \
  core/box/../front/../box/tiny_front_cold_box.h \
+ core/box/../front/../box/tiny_c7_hotpath_box.h \
+ core/box/../front/../box/c7_hotpath_env_box.h \
+ core/box/../front/../box/tiny_c7_uc_hit_box.h \
+ core/box/../front/../box/tiny_c7_warm_spill_box.h \
+ core/box/../front/../box/tiny_c7_stats_sample_box.h \
+ core/box/../front/../box/tiny_front_hot_box.h \
+ core/box/../front/../box/tiny_front_cold_box.h \
+ core/box/../front/../box/front_gate_box.h \
+ core/box/../front/../box/tls_sll_box.h \
+ core/box/../front/../box/ptr_conversion_box.h \
  core/box/tiny_alloc_gate_box.h core/box/tiny_route_box.h \
  core/box/tiny_front_config_box.h core/box/wrapper_env_box.h \
  core/box/../hakmem_internal.h core/box/../superslab/superslab_inline.h
@@ -131,6 +142,7 @@ core/ptr_track.h:
 core/hakmem_super_registry.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/tiny_debug_api.h:
 core/box/tiny_layout_box.h:
 core/box/../hakmem_tiny_config.h:
@@ -244,6 +256,16 @@ core/box/../front/../box/../tiny_region_id.h:
 core/box/../front/../box/../front/tiny_unified_cache.h:
 core/box/../front/../box/tiny_layout_box.h:
 core/box/../front/../box/tiny_front_cold_box.h:
+core/box/../front/../box/tiny_c7_hotpath_box.h:
+core/box/../front/../box/c7_hotpath_env_box.h:
+core/box/../front/../box/tiny_c7_uc_hit_box.h:
+core/box/../front/../box/tiny_c7_warm_spill_box.h:
+core/box/../front/../box/tiny_c7_stats_sample_box.h:
+core/box/../front/../box/tiny_front_hot_box.h:
+core/box/../front/../box/tiny_front_cold_box.h:
+core/box/../front/../box/front_gate_box.h:
+core/box/../front/../box/tls_sll_box.h:
+core/box/../front/../box/ptr_conversion_box.h:
 core/box/tiny_alloc_gate_box.h:
 core/box/tiny_route_box.h:
 core/box/tiny_front_config_box.h:

hakmem_shared_pool.d

@@ -12,9 +12,10 @@ hakmem_shared_pool.o: core/hakmem_shared_pool.c \
  core/box/tiny_next_ptr_box.h core/hakmem_tiny_config.h \
  core/tiny_nextptr.h core/tiny_region_id.h core/tiny_box_geometry.h \
  core/ptr_track.h core/hakmem_super_registry.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/hakmem_tiny.h core/hakmem_trace.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_tiny_mini_mag.h core/box/hak_lane_classify.inc.h \
+ core/hakmem_tiny.h core/hakmem_trace.h core/hakmem_tiny_mini_mag.h \
- core/box/ptr_type_box.h core/tiny_debug_api.h core/box/tiny_layout_box.h \
+ core/box/hak_lane_classify.inc.h core/box/ptr_type_box.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
  core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
  core/box/tiny_layout_box.h core/box/../tiny_region_id.h \
  core/box/ss_hot_cold_box.h core/box/pagefault_telemetry_box.h \
@@ -40,7 +41,8 @@ hakmem_shared_pool.o: core/hakmem_shared_pool.c \
  core/box/ss_hot_cold_box.h core/box/ss_release_guard_box.h \
  core/box/free_local_box.h core/box/ptr_type_box.h \
  core/box/free_publish_box.h core/hakmem_tiny.h core/tiny_region_id.h \
- core/box/tls_slab_reuse_guard_box.h core/hakmem_policy.h
+ core/box/tls_slab_reuse_guard_box.h core/hakmem_policy.h \
+ core/box/shared_pool_box.h
 core/hakmem_shared_pool_internal.h:
 core/hakmem_shared_pool.h:
 core/superslab/superslab_types.h:
@@ -69,6 +71,7 @@ core/ptr_track.h:
 core/hakmem_super_registry.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/hakmem_tiny.h:
 core/hakmem_trace.h:
 core/hakmem_tiny_mini_mag.h:
@@ -127,3 +130,4 @@ core/hakmem_tiny.h:
 core/tiny_region_id.h:
 core/box/tls_slab_reuse_guard_box.h:
 core/hakmem_policy.h:
+core/box/shared_pool_box.h:

hakmem_super_registry.d

@@ -7,9 +7,9 @@ hakmem_super_registry.o: core/hakmem_super_registry.c \
  core/superslab/../hakmem_tiny_config.h core/tiny_debug_ring.h \
  core/hakmem_build_flags.h core/tiny_remote.h \
  core/hakmem_tiny_superslab_constants.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/box/ss_allocation_box.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_tiny_superslab.h core/box/ss_cold_start_box.inc.h \
+ core/box/ss_allocation_box.h core/hakmem_tiny_superslab.h \
- core/hakmem_env_cache.h
+ core/box/ss_cold_start_box.inc.h core/hakmem_env_cache.h
 core/hakmem_super_registry.h:
 core/hakmem_tiny_superslab.h:
 core/superslab/superslab_types.h:
@@ -25,6 +25,7 @@ core/tiny_remote.h:
 core/hakmem_tiny_superslab_constants.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/box/ss_allocation_box.h:
 core/hakmem_tiny_superslab.h:
 core/box/ss_cold_start_box.inc.h:

hakmem_tiny_bg_spill.d

@@ -8,9 +8,10 @@ hakmem_tiny_bg_spill.o: core/hakmem_tiny_bg_spill.c \
  core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
  core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
  core/tiny_debug_ring.h core/tiny_remote.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/hakmem_tiny.h core/hakmem_trace.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_tiny_mini_mag.h core/box/hak_lane_classify.inc.h \
+ core/hakmem_tiny.h core/hakmem_trace.h core/hakmem_tiny_mini_mag.h \
- core/box/ptr_type_box.h core/tiny_debug_api.h core/box/tiny_layout_box.h \
+ core/box/hak_lane_classify.inc.h core/box/ptr_type_box.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
  core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
  core/box/tiny_layout_box.h core/box/../tiny_region_id.h
 core/hakmem_tiny_bg_spill.h:
@@ -34,6 +35,7 @@ core/tiny_debug_ring.h:
 core/tiny_remote.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/hakmem_tiny.h:
 core/hakmem_trace.h:
 core/hakmem_tiny_mini_mag.h:

hakmem_tiny_magazine.d

@@ -10,14 +10,15 @@ hakmem_tiny_magazine.o: core/hakmem_tiny_magazine.c \
  core/superslab/../hakmem_tiny_config.h core/tiny_debug_ring.h \
  core/tiny_remote.h core/hakmem_tiny_superslab_constants.h \
  core/hakmem_super_registry.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/hakmem_prof.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_internal.h core/hakmem.h core/hakmem_config.h \
+ core/hakmem_prof.h core/hakmem_internal.h core/hakmem.h \
- core/hakmem_features.h core/hakmem_sys.h core/hakmem_whale.h \
+ core/hakmem_config.h core/hakmem_features.h core/hakmem_sys.h \
- core/box/tiny_next_ptr_box.h core/hakmem_tiny_config.h \
+ core/hakmem_whale.h core/box/tiny_next_ptr_box.h \
- core/tiny_nextptr.h core/tiny_region_id.h core/tiny_box_geometry.h \
+ core/hakmem_tiny_config.h core/tiny_nextptr.h core/tiny_region_id.h \
- core/ptr_track.h core/tiny_debug_api.h core/box/tiny_layout_box.h \
+ core/tiny_box_geometry.h core/ptr_track.h core/tiny_debug_api.h \
- core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
+ core/box/tiny_layout_box.h core/box/../hakmem_tiny_config.h \
- core/box/tiny_layout_box.h core/box/../tiny_region_id.h
+ core/box/tiny_header_box.h core/box/tiny_layout_box.h \
+ core/box/../tiny_region_id.h core/box/tiny_mem_stats_box.h
 core/hakmem_tiny_magazine.h:
 core/hakmem_tiny.h:
 core/hakmem_build_flags.h:
@@ -40,6 +41,7 @@ core/hakmem_tiny_superslab_constants.h:
 core/hakmem_super_registry.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/hakmem_prof.h:
 core/hakmem_internal.h:
 core/hakmem.h:
@@ -59,3 +61,4 @@ core/box/../hakmem_tiny_config.h:
 core/box/tiny_header_box.h:
 core/box/tiny_layout_box.h:
 core/box/../tiny_region_id.h:
+core/box/tiny_mem_stats_box.h:

hakmem_tiny_query.d

@@ -10,8 +10,8 @@ hakmem_tiny_query.o: core/hakmem_tiny_query.c core/hakmem_tiny.h \
  core/superslab/../hakmem_tiny_config.h core/tiny_debug_ring.h \
  core/tiny_remote.h core/hakmem_tiny_superslab_constants.h \
  core/hakmem_super_registry.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/hakmem_config.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_features.h
+ core/hakmem_config.h core/hakmem_features.h
 core/hakmem_tiny.h:
 core/hakmem_build_flags.h:
 core/hakmem_trace.h:
@@ -34,5 +34,6 @@ core/hakmem_tiny_superslab_constants.h:
 core/hakmem_super_registry.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/hakmem_config.h:
 core/hakmem_features.h:

hakmem_tiny_sfc.d

@@ -9,21 +9,21 @@ hakmem_tiny_sfc.o: core/hakmem_tiny_sfc.c core/tiny_alloc_fast_sfc.inc.h \
  core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
  core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
  core/tiny_debug_ring.h core/tiny_remote.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/tiny_debug_api.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/box/tiny_layout_box.h core/box/../hakmem_tiny_config.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
- core/box/tiny_header_box.h core/box/tiny_layout_box.h \
+ core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
- core/box/../tiny_region_id.h core/hakmem_stats_master.h core/tiny_tls.h \
+ core/box/tiny_layout_box.h core/box/../tiny_region_id.h \
- core/box/tls_sll_box.h core/box/../hakmem_internal.h \
+ core/hakmem_stats_master.h core/tiny_tls.h core/box/tls_sll_box.h \
- core/box/../hakmem.h core/box/../hakmem_build_flags.h \
+ core/box/../hakmem_internal.h core/box/../hakmem.h \
- core/box/../hakmem_config.h core/box/../hakmem_features.h \
+ core/box/../hakmem_build_flags.h core/box/../hakmem_config.h \
- core/box/../hakmem_sys.h core/box/../hakmem_whale.h \
+ core/box/../hakmem_features.h core/box/../hakmem_sys.h \
- core/box/../box/ptr_type_box.h core/box/../hakmem_debug_master.h \
+ core/box/../hakmem_whale.h core/box/../box/ptr_type_box.h \
- core/box/../tiny_remote.h core/box/../hakmem_tiny_integrity.h \
+ core/box/../hakmem_debug_master.h core/box/../tiny_remote.h \
- core/box/../hakmem_tiny.h core/box/../ptr_track.h \
+ core/box/../hakmem_tiny_integrity.h core/box/../hakmem_tiny.h \
- core/box/../ptr_trace.h core/box/../hakmem_trace_master.h \
+ core/box/../ptr_track.h core/box/../ptr_trace.h \
- core/box/../hakmem_stats_master.h core/box/../tiny_debug_ring.h \
+ core/box/../hakmem_trace_master.h core/box/../hakmem_stats_master.h \
- core/box/ss_addr_map_box.h core/box/../superslab/superslab_inline.h \
+ core/box/../tiny_debug_ring.h core/box/ss_addr_map_box.h \
- core/box/tiny_ptr_bridge_box.h \
+ core/box/../superslab/superslab_inline.h core/box/tiny_ptr_bridge_box.h \
  core/box/../hakmem_tiny_superslab_internal.h \
  core/box/../hakmem_tiny_superslab.h core/box/../box/ss_hot_cold_box.h \
  core/box/../box/../superslab/superslab_types.h \
@@ -60,6 +60,7 @@ core/tiny_debug_ring.h:
 core/tiny_remote.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/tiny_debug_api.h:
 core/box/tiny_layout_box.h:
 core/box/../hakmem_tiny_config.h:

tiny_adaptive_sizing.d

@@ -10,10 +10,10 @@ tiny_adaptive_sizing.o: core/tiny_adaptive_sizing.c \
  core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
  core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
  core/tiny_debug_ring.h core/tiny_remote.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/tiny_debug_api.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/box/tiny_layout_box.h core/box/../hakmem_tiny_config.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
- core/box/tiny_header_box.h core/box/tiny_layout_box.h \
+ core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
- core/box/../tiny_region_id.h
+ core/box/tiny_layout_box.h core/box/../tiny_region_id.h
 core/tiny_adaptive_sizing.h:
 core/hakmem_tiny.h:
 core/hakmem_build_flags.h:
@@ -40,6 +40,7 @@ core/tiny_debug_ring.h:
 core/tiny_remote.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/tiny_debug_api.h:
 core/box/tiny_layout_box.h:
 core/box/../hakmem_tiny_config.h:

tiny_fastcache.d

@@ -8,9 +8,10 @@ tiny_fastcache.o: core/tiny_fastcache.c core/tiny_fastcache.h \
  core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
  core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
  core/tiny_debug_ring.h core/tiny_remote.h core/box/ss_addr_map_box.h \
- core/box/../hakmem_build_flags.h core/hakmem_tiny.h core/hakmem_trace.h \
+ core/box/../hakmem_build_flags.h core/box/super_reg_box.h \
- core/hakmem_tiny_mini_mag.h core/box/hak_lane_classify.inc.h \
+ core/hakmem_tiny.h core/hakmem_trace.h core/hakmem_tiny_mini_mag.h \
- core/box/ptr_type_box.h core/tiny_debug_api.h core/box/tiny_layout_box.h \
+ core/box/hak_lane_classify.inc.h core/box/ptr_type_box.h \
+ core/tiny_debug_api.h core/box/tiny_layout_box.h \
  core/box/../hakmem_tiny_config.h core/box/tiny_header_box.h \
  core/box/tiny_layout_box.h core/box/../tiny_region_id.h
 core/tiny_fastcache.h:
@@ -35,6 +36,7 @@ core/tiny_debug_ring.h:
 core/tiny_remote.h:
 core/box/ss_addr_map_box.h:
 core/box/../hakmem_build_flags.h:
+core/box/super_reg_box.h:
 core/hakmem_tiny.h:
 core/hakmem_trace.h:
 core/hakmem_tiny_mini_mag.h:

tiny_remote.d

@@ -1,11 +1,13 @@
 tiny_remote.o: core/tiny_remote.c core/tiny_remote.h \
- core/hakmem_tiny_superslab.h core/superslab/superslab_types.h \
+ core/box/remote_side_box.h core/hakmem_tiny_superslab.h \
- core/hakmem_tiny_superslab_constants.h core/superslab/superslab_inline.h \
+ core/superslab/superslab_types.h core/hakmem_tiny_superslab_constants.h \
- core/superslab/superslab_types.h core/superslab/../tiny_box_geometry.h \
+ core/superslab/superslab_inline.h core/superslab/superslab_types.h \
+ core/superslab/../tiny_box_geometry.h \
  core/superslab/../hakmem_tiny_superslab_constants.h \
  core/superslab/../hakmem_tiny_config.h core/tiny_debug_ring.h \
  core/hakmem_build_flags.h core/hakmem_tiny_superslab_constants.h
 core/tiny_remote.h:
+core/box/remote_side_box.h:
 core/hakmem_tiny_superslab.h:
 core/superslab/superslab_types.h:
 core/hakmem_tiny_superslab_constants.h: