P-Tier + Tiny Route Policy: Aggressive Superslab Management + Safe Routing

## Phase 1: Utilization-Aware Superslab Tiering (案B実装済)

- Add ss_tier_box.h: Classify SuperSlabs into HOT/DRAINING/FREE based on utilization
  - HOT (>25%): Accept new allocations
  - DRAINING (≤25%): Drain only, no new allocs
  - FREE (0%): Ready for eager munmap

- Enhanced shared_pool_release_slab():
  - Check tier transition after each slab release
  - If tier→FREE: Force remaining slots to EMPTY and call superslab_free() immediately
  - Bypasses LRU cache to prevent registry bloat from accumulating DRAINING SuperSlabs

- Test results (bench_random_mixed_hakmem):
  - 1M iterations:  ~1.03M ops/s (previously passed)
  - 10M iterations:  ~1.15M ops/s (previously: registry full error)
  - 50M iterations:  ~1.08M ops/s (stress test)

## Phase 2: Tiny Front Routing Policy (新規Box)

- Add tiny_route_box.h/c: Single 8-byte table for class→routing decisions
  - ROUTE_TINY_ONLY: Tiny front exclusive (no fallback)
  - ROUTE_TINY_FIRST: Try Tiny, fallback to Pool if fails
  - ROUTE_POOL_ONLY: Skip Tiny entirely

- Profiles via HAKMEM_TINY_PROFILE ENV:
  - "hot": C0-C3=TINY_ONLY, C4-C6=TINY_FIRST, C7=POOL_ONLY
  - "conservative" (default): All TINY_FIRST
  - "off": All POOL_ONLY (disable Tiny)
  - "full": All TINY_ONLY (microbench mode)

- A/B test results (ws=256, 100k ops random_mixed):
  - Default (conservative): ~2.90M ops/s
  - hot: ~2.65M ops/s (more conservative)
  - off: ~2.86M ops/s
  - full: ~2.98M ops/s (slightly best)

## Design Rationale

### Registry Pressure Fix (案B)
- Problem: DRAINING tier SS occupied registry indefinitely
- Solution: When total_active_blocks→0, immediately free to clear registry slot
- Result: No more "registry full" errors under stress

### Routing Policy Box (新)
- Problem: Tiny front optimization scattered across ENV/branches
- Solution: Centralize routing in single table, select profiles via ENV
- Benefit: Safe A/B testing without touching hot path code
- Future: Integrate with RSS budget/learning layers for dynamic profile switching

## Next Steps (性能最適化)
- Profile Tiny front internals (TLS SLL, FastCache, Superslab backend latency)
- Identify bottleneck between current ~2.9M ops/s and mimalloc ~100M ops/s
- Consider:
  - Reduce shared pool lock contention
  - Optimize unified cache hit rate
  - Streamline Superslab carving logic

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Moe Charm (CI)
2025-12-04 18:01:25 +09:00
parent 984cca41ef
commit d5e6ed535c
13 changed files with 647 additions and 25 deletions

View File

@ -30,6 +30,7 @@
#include "tiny_ptr_bridge_box.h" // Tiny Superslab Bridge
#include "../tiny_region_id.h" // Header 読み出し
#include "../front/malloc_tiny_fast.h" // 既存 Tiny Fast Path
#include "tiny_route_box.h" // Tiny Front Routing Policy
// 将来の拡張用コンテキスト:
// - size : 要求サイズ
@ -133,15 +134,59 @@ static inline int tiny_alloc_gate_validate(TinyAllocGateContext* ctx)
// Tiny Alloc Gatekeeper 本体:
// - malloc ラッパ (hak_wrappers) から呼ばれる Tiny fast alloc の入口。
// - 現状は malloc_tiny_fast(size) の薄いラッパで、診断 ON のときだけ
// 返された USER ポインタに対して Bridge + Layout 検査を追加。
// - ルーティングポリシーに基づき Tiny front / Pool fallback を振り分け、
// 診断 ON のときだけ返された USER ポインタに対して Bridge + Layout 検査を追加。
static inline void* tiny_alloc_gate_fast(size_t size)
{
// まずは従来どおり Tiny Fast Path で割り当てUSER ポインタを得る)
int class_idx = hak_tiny_size_to_class(size);
if (__builtin_expect(class_idx < 0 || class_idx >= TINY_NUM_CLASSES, 0)) {
// サイズが Tiny 管理外 → Pool/backend に任せるNULL で Gate を抜けさせる)
return NULL;
}
TinyRoutePolicy route = tiny_route_get(class_idx);
// Pool-only: Tiny front は完全スキップGate から見ると「Tiny では取れなかった」扱い)
if (__builtin_expect(route == ROUTE_POOL_ONLY, 0)) {
return NULL;
}
// まず Tiny Fast Path で割り当てUSER ポインタを得る)
void* user_ptr = malloc_tiny_fast(size);
// Layer 3aalloc 側): 取得したポインタが明らかに異常な場合は
// Debug ビルドで早期に検出して Fail-Fast。
// Tiny-only: その結果をそのまま返すNULL なら上位が扱う)
if (__builtin_expect(route == ROUTE_TINY_ONLY, 1)) {
#if !HAKMEM_BUILD_RELEASE
// Layer 3aalloc 側): 明らかに異常なポインタは debug ビルドで早期検出
if (user_ptr) {
uintptr_t addr = (uintptr_t)user_ptr;
if (__builtin_expect(addr < 4096, 0)) {
fprintf(stderr,
"[TINY_ALLOC_GATE_RANGE_INVALID] size=%zu user=%p\n",
size, user_ptr);
fflush(stderr);
abort();
}
}
if (__builtin_expect(tiny_alloc_gate_diag_enabled(), 0) && user_ptr) {
TinyAllocGateContext ctx;
ctx.size = size;
ctx.user = HAK_USER_FROM_RAW(user_ptr);
ctx.class_idx = class_idx;
ctx.base = HAK_BASE_FROM_RAW(NULL);
ctx.bridge.ss = NULL;
ctx.bridge.meta = NULL;
ctx.bridge.slab_idx = -1;
ctx.bridge.meta_cls = 0xffu;
(void)tiny_alloc_gate_validate(&ctx);
}
#endif
return user_ptr;
}
// ROUTE_TINY_FIRST: Tiny で取れなければ Pool/backend fallback を許可NULL で Gate 脱出)
#if !HAKMEM_BUILD_RELEASE
if (user_ptr) {
uintptr_t addr = (uintptr_t)user_ptr;
@ -152,20 +197,20 @@ static inline void* tiny_alloc_gate_fast(size_t size)
fflush(stderr);
abort();
}
}
if (__builtin_expect(tiny_alloc_gate_diag_enabled(), 0) && user_ptr) {
TinyAllocGateContext ctx;
ctx.size = size;
ctx.user = HAK_USER_FROM_RAW(user_ptr);
ctx.class_idx = hak_tiny_size_to_class(size);
ctx.base = HAK_BASE_FROM_RAW(NULL);
ctx.bridge.ss = NULL;
ctx.bridge.meta = NULL;
ctx.bridge.slab_idx = -1;
ctx.bridge.meta_cls = 0xffu;
if (__builtin_expect(tiny_alloc_gate_diag_enabled(), 0)) {
TinyAllocGateContext ctx;
ctx.size = size;
ctx.user = HAK_USER_FROM_RAW(user_ptr);
ctx.class_idx = class_idx;
ctx.base = HAK_BASE_FROM_RAW(NULL);
ctx.bridge.ss = NULL;
ctx.bridge.meta = NULL;
ctx.bridge.slab_idx = -1;
ctx.bridge.meta_cls = 0xffu;
(void)tiny_alloc_gate_validate(&ctx);
(void)tiny_alloc_gate_validate(&ctx);
}
}
#endif