Files
hakmem/core/page_arena.h

248 lines
8.4 KiB
C
Raw Normal View History

Phase 24 PageArena/HotSpanBox: Mid/VM page reuse cache (structural limit identified) Summary: - Implemented PageArena (Box PA1-PA3) for Mid-Large (8-52KB) / L25 (64KB-2MB) - Integration: Pool TLS Arena + L25 alloc/refill paths - Result: Minimal impact (+4.7% Mid, 0% VM page-fault reduction) - Conclusion: Structural limit - existing Arena/Pool/L25 already optimized Implementation: 1. Box PA1: Hot Page Cache (4KB pages, LIFO stack, 1024 slots) - core/page_arena.c: hot_page_alloc/free with mutex protection - TLS cache for 4KB pages 2. Box PA2: Warm Span Cache (64KB-2MB spans, size-bucketed) - 64KB/128KB/2MB span caches (256/128/64 slots) - Size-class based allocation 3. Box PA3: Cold Path (mmap fallback) - page_arena_alloc_pages/aligned with fallback to direct mmap Integration Points: 4. Pool TLS Arena (core/pool_tls_arena.c) - chunk_ensure(): Lazy init + page_arena_alloc_pages() hook - arena_cleanup_thread(): Return chunks to PageArena if enabled - Exponential growth preserved (1MB → 8MB) 5. L25 Pool (core/hakmem_l25_pool.c) - l25_alloc_new_run(): Lazy init + page_arena_alloc_aligned() hook - refill_freelist(): PageArena allocation for bundles - 2MB run carving preserved ENV Variables: - HAKMEM_PAGE_ARENA_ENABLE=1 (default: 0, OFF) - HAKMEM_PAGE_ARENA_HOT_SIZE=1024 (default: 1024) - HAKMEM_PAGE_ARENA_WARM_64K=256 (default: 256) - HAKMEM_PAGE_ARENA_WARM_128K=128 (default: 128) - HAKMEM_PAGE_ARENA_WARM_2M=64 (default: 64) Benchmark Results: - Mid-Large MT (4T, 40K iter, 2KB): - OFF: 84,535 page-faults, 726K ops/s - ON: 84,534 page-faults, 760K ops/s (+4.7% ops, -0.001% faults) - VM Mixed (200K iter): - OFF: 102,134 page-faults, 257K ops/s - ON: 102,134 page-faults, 255K ops/s (0% change) Root Cause Analysis: - Hypothesis: 50-66% page-fault reduction (80-100K → 30-40K) - Actual: <1% page-fault reduction, minimal performance impact - Reason: Structural limit - existing Arena/Pool/L25 already highly optimized - 1MB chunk sizes with high-density linear carving - TLS ring + exponential growth minimize mmap calls - PageArena becomes double-buffering layer with no benefit - Remaining page-faults from kernel zero-clear + app access patterns Lessons Learned: 1. Mid/Large allocators already page-optimal via Arena/Pool design 2. Middle-layer caching ineffective when base layer already optimized 3. Page-fault reduction requires app-level access pattern changes 4. Tiny layer (Phase 23) remains best target for frontend optimization Next Steps: - Defer PageArena (low ROI, structural limit reached) - Focus on upper layers (allocation pattern analysis, size distribution) - Consider app-side access pattern optimization 🤖 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-17 03:22:27 +09:00
// page_arena.h - Phase 24: PageArena/HotSpanBox (Mid-Large page-fault optimization)
//
// Goal: Reduce Mid/VM page-faults by 50-66% (80-100K → 30-40K)
// Target: +30-50% performance for Mid-Large (8-52KB) / L25 (64KB-2MB)
//
// Design:
// Box PA1: Hot Page Cache (4KB pages, LIFO stack, 1024 slots)
// Box PA2: Warm Span Cache (64KB-2MB spans, size-bucketed, 448 slots)
// Box PA3: Cold Path (mmap fallback when cache misses)
//
// Integration:
// - Pool TLS: chunk_ensure() → page_arena_alloc_pages()
// - L25: l25_alloc_new_run() → page_arena_alloc_aligned()
// - L25: refill_freelist() → page_arena_alloc_aligned()
//
// ENV Variables:
// HAKMEM_PAGE_ARENA_ENABLE=1 # Enable PageArena (default: 0, OFF)
// HAKMEM_PAGE_ARENA_HOT_SIZE=1024 # Hot page cache size (default: 1024)
// HAKMEM_PAGE_ARENA_WARM_64K=256 # Warm 64KB span cache size (default: 256)
// HAKMEM_PAGE_ARENA_WARM_128K=128 # Warm 128KB span cache size (default: 128)
// HAKMEM_PAGE_ARENA_WARM_2M=64 # Warm 2MB span cache size (default: 64)
#ifndef HAK_PAGE_ARENA_H
#define HAK_PAGE_ARENA_H
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
Phase 24 PageArena/HotSpanBox: Mid/VM page reuse cache (structural limit identified) Summary: - Implemented PageArena (Box PA1-PA3) for Mid-Large (8-52KB) / L25 (64KB-2MB) - Integration: Pool TLS Arena + L25 alloc/refill paths - Result: Minimal impact (+4.7% Mid, 0% VM page-fault reduction) - Conclusion: Structural limit - existing Arena/Pool/L25 already optimized Implementation: 1. Box PA1: Hot Page Cache (4KB pages, LIFO stack, 1024 slots) - core/page_arena.c: hot_page_alloc/free with mutex protection - TLS cache for 4KB pages 2. Box PA2: Warm Span Cache (64KB-2MB spans, size-bucketed) - 64KB/128KB/2MB span caches (256/128/64 slots) - Size-class based allocation 3. Box PA3: Cold Path (mmap fallback) - page_arena_alloc_pages/aligned with fallback to direct mmap Integration Points: 4. Pool TLS Arena (core/pool_tls_arena.c) - chunk_ensure(): Lazy init + page_arena_alloc_pages() hook - arena_cleanup_thread(): Return chunks to PageArena if enabled - Exponential growth preserved (1MB → 8MB) 5. L25 Pool (core/hakmem_l25_pool.c) - l25_alloc_new_run(): Lazy init + page_arena_alloc_aligned() hook - refill_freelist(): PageArena allocation for bundles - 2MB run carving preserved ENV Variables: - HAKMEM_PAGE_ARENA_ENABLE=1 (default: 0, OFF) - HAKMEM_PAGE_ARENA_HOT_SIZE=1024 (default: 1024) - HAKMEM_PAGE_ARENA_WARM_64K=256 (default: 256) - HAKMEM_PAGE_ARENA_WARM_128K=128 (default: 128) - HAKMEM_PAGE_ARENA_WARM_2M=64 (default: 64) Benchmark Results: - Mid-Large MT (4T, 40K iter, 2KB): - OFF: 84,535 page-faults, 726K ops/s - ON: 84,534 page-faults, 760K ops/s (+4.7% ops, -0.001% faults) - VM Mixed (200K iter): - OFF: 102,134 page-faults, 257K ops/s - ON: 102,134 page-faults, 255K ops/s (0% change) Root Cause Analysis: - Hypothesis: 50-66% page-fault reduction (80-100K → 30-40K) - Actual: <1% page-fault reduction, minimal performance impact - Reason: Structural limit - existing Arena/Pool/L25 already highly optimized - 1MB chunk sizes with high-density linear carving - TLS ring + exponential growth minimize mmap calls - PageArena becomes double-buffering layer with no benefit - Remaining page-faults from kernel zero-clear + app access patterns Lessons Learned: 1. Mid/Large allocators already page-optimal via Arena/Pool design 2. Middle-layer caching ineffective when base layer already optimized 3. Page-fault reduction requires app-level access pattern changes 4. Tiny layer (Phase 23) remains best target for frontend optimization Next Steps: - Defer PageArena (low ROI, structural limit reached) - Focus on upper layers (allocation pattern analysis, size distribution) - Consider app-side access pattern optimization 🤖 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-17 03:22:27 +09:00
#include <pthread.h>
#include "hakmem_build_flags.h"
Code Cleanup: Remove false positives, redundant validations, and reduce verbose logging Following the C7 stride upgrade fix (commit 23c0d9541), this commit performs comprehensive cleanup to improve code quality and reduce debug noise. ## Changes ### 1. Disable False Positive Checks (tiny_nextptr.h) - **Disabled**: NXT_MISALIGN validation block with `#if 0` - **Reason**: Produces false positives due to slab base offsets (2048, 65536) not being stride-aligned, causing all blocks to appear "misaligned" - **TODO**: Reimplement to check stride DISTANCE between consecutive blocks instead of absolute alignment to stride boundaries ### 2. Remove Redundant Geometry Validations **hakmem_tiny_refill_p0.inc.h (P0 batch refill)** - Removed 25-line CARVE_GEOMETRY_FIX validation block - Replaced with NOTE explaining redundancy - **Reason**: Stride table is now correct in tiny_block_stride_for_class(), defense-in-depth validation adds overhead without benefit **ss_legacy_backend_box.c (legacy backend)** - Removed 18-line LEGACY_FIX_GEOMETRY validation block - Replaced with NOTE explaining redundancy - **Reason**: Shared_pool validates geometry at acquisition time ### 3. Reduce Verbose Logging **hakmem_shared_pool.c (sp_fix_geometry_if_needed)** - Made SP_FIX_GEOMETRY logging conditional on `!HAKMEM_BUILD_RELEASE` - **Reason**: Geometry fixes are expected during stride upgrades, no need to log in release builds ### 4. Verification - Build: ✅ Successful (LTO warnings expected) - Test: ✅ 10K iterations (1.87M ops/s, no crashes) - NXT_MISALIGN false positives: ✅ Eliminated ## Files Modified - core/tiny_nextptr.h - Disabled false positive NXT_MISALIGN check - core/hakmem_tiny_refill_p0.inc.h - Removed redundant CARVE validation - core/box/ss_legacy_backend_box.c - Removed redundant LEGACY validation - core/hakmem_shared_pool.c - Made SP_FIX_GEOMETRY logging debug-only ## Impact - **Code clarity**: Removed 43 lines of redundant validation code - **Debug noise**: Reduced false positive diagnostics - **Performance**: Eliminated overhead from redundant geometry checks - **Maintainability**: Single source of truth for geometry validation 🧹 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-21 23:00:24 +09:00
#include <stdio.h> // debug logging
Phase 24 PageArena/HotSpanBox: Mid/VM page reuse cache (structural limit identified) Summary: - Implemented PageArena (Box PA1-PA3) for Mid-Large (8-52KB) / L25 (64KB-2MB) - Integration: Pool TLS Arena + L25 alloc/refill paths - Result: Minimal impact (+4.7% Mid, 0% VM page-fault reduction) - Conclusion: Structural limit - existing Arena/Pool/L25 already optimized Implementation: 1. Box PA1: Hot Page Cache (4KB pages, LIFO stack, 1024 slots) - core/page_arena.c: hot_page_alloc/free with mutex protection - TLS cache for 4KB pages 2. Box PA2: Warm Span Cache (64KB-2MB spans, size-bucketed) - 64KB/128KB/2MB span caches (256/128/64 slots) - Size-class based allocation 3. Box PA3: Cold Path (mmap fallback) - page_arena_alloc_pages/aligned with fallback to direct mmap Integration Points: 4. Pool TLS Arena (core/pool_tls_arena.c) - chunk_ensure(): Lazy init + page_arena_alloc_pages() hook - arena_cleanup_thread(): Return chunks to PageArena if enabled - Exponential growth preserved (1MB → 8MB) 5. L25 Pool (core/hakmem_l25_pool.c) - l25_alloc_new_run(): Lazy init + page_arena_alloc_aligned() hook - refill_freelist(): PageArena allocation for bundles - 2MB run carving preserved ENV Variables: - HAKMEM_PAGE_ARENA_ENABLE=1 (default: 0, OFF) - HAKMEM_PAGE_ARENA_HOT_SIZE=1024 (default: 1024) - HAKMEM_PAGE_ARENA_WARM_64K=256 (default: 256) - HAKMEM_PAGE_ARENA_WARM_128K=128 (default: 128) - HAKMEM_PAGE_ARENA_WARM_2M=64 (default: 64) Benchmark Results: - Mid-Large MT (4T, 40K iter, 2KB): - OFF: 84,535 page-faults, 726K ops/s - ON: 84,534 page-faults, 760K ops/s (+4.7% ops, -0.001% faults) - VM Mixed (200K iter): - OFF: 102,134 page-faults, 257K ops/s - ON: 102,134 page-faults, 255K ops/s (0% change) Root Cause Analysis: - Hypothesis: 50-66% page-fault reduction (80-100K → 30-40K) - Actual: <1% page-fault reduction, minimal performance impact - Reason: Structural limit - existing Arena/Pool/L25 already highly optimized - 1MB chunk sizes with high-density linear carving - TLS ring + exponential growth minimize mmap calls - PageArena becomes double-buffering layer with no benefit - Remaining page-faults from kernel zero-clear + app access patterns Lessons Learned: 1. Mid/Large allocators already page-optimal via Arena/Pool design 2. Middle-layer caching ineffective when base layer already optimized 3. Page-fault reduction requires app-level access pattern changes 4. Tiny layer (Phase 23) remains best target for frontend optimization Next Steps: - Defer PageArena (low ROI, structural limit reached) - Focus on upper layers (allocation pattern analysis, size distribution) - Consider app-side access pattern optimization 🤖 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-17 03:22:27 +09:00
// ============================================================================
// Box PA1: Hot Page Cache (4KB pages)
// ============================================================================
#define PA_HOT_PAGE_DEFAULT_SIZE 1024 // 1024 slots = 4MB cache
typedef struct {
void** pages; // Dynamic array of 4KB pages
int capacity; // Max slots (power of 2)
int count; // Current occupancy
pthread_mutex_t lock; // Lock for MT safety
// Metrics (debug only)
#if !HAKMEM_BUILD_RELEASE
uint64_t hits; // Alloc hits
uint64_t misses; // Alloc misses (fallback to mmap)
uint64_t frees; // Free pushes
uint64_t evictions; // Free evictions (cache full)
#endif
} HotPageCache;
// Initialize hot page cache (called at startup or lazy init)
void hot_page_cache_init(HotPageCache* cache, int capacity);
// Shutdown hot page cache (called at cleanup)
void hot_page_cache_shutdown(HotPageCache* cache);
// Allocate 4KB page from hot cache (returns NULL if cache miss)
void* hot_page_alloc(HotPageCache* cache);
// Free 4KB page to hot cache (may evict if cache full)
void hot_page_free(HotPageCache* cache, void* page);
// ============================================================================
// Box PA2: Warm Span Cache (64KB-2MB spans)
// ============================================================================
#define PA_WARM_64K_DEFAULT_SIZE 256 // 256 slots = 16MB cache
#define PA_WARM_128K_DEFAULT_SIZE 128 // 128 slots = 16MB cache
#define PA_WARM_2M_DEFAULT_SIZE 64 // 64 slots = 128MB cache
typedef struct {
// 64KB spans
void** spans_64k;
int capacity_64k;
int count_64k;
// 128KB spans
void** spans_128k;
int capacity_128k;
int count_128k;
// 2MB spans
void** spans_2m;
int capacity_2m;
int count_2m;
pthread_mutex_t lock; // Lock for MT safety
// Metrics (debug only)
#if !HAKMEM_BUILD_RELEASE
uint64_t hits_64k;
uint64_t hits_128k;
uint64_t hits_2m;
uint64_t misses;
uint64_t frees_64k;
uint64_t frees_128k;
uint64_t frees_2m;
uint64_t evictions;
#endif
} WarmSpanCache;
// Initialize warm span cache (called at startup or lazy init)
void warm_span_cache_init(WarmSpanCache* cache, int cap_64k, int cap_128k, int cap_2m);
// Shutdown warm span cache (called at cleanup)
void warm_span_cache_shutdown(WarmSpanCache* cache);
// Allocate aligned span from warm cache (returns NULL if cache miss)
void* warm_span_alloc(WarmSpanCache* cache, size_t size);
// Free aligned span to warm cache (may evict if cache full)
void warm_span_free(WarmSpanCache* cache, void* span, size_t size);
// ============================================================================
// Box PA3: Unified PageArena (combines PA1 + PA2 + Cold Path)
// ============================================================================
typedef struct {
HotPageCache hot; // Box PA1: 4KB pages
WarmSpanCache warm; // Box PA2: 64KB-2MB spans
// Metrics (debug only)
#if !HAKMEM_BUILD_RELEASE
uint64_t total_allocs;
uint64_t total_frees;
uint64_t mmap_calls; // Cold path fallback count
#endif
} PageArena;
// Global page arena (TLS per thread, or single global instance)
extern __thread PageArena g_page_arena;
// ============================================================================
// ENV Control (cached, lazy init)
// ============================================================================
// Enable flag (default: 0, OFF)
static inline int page_arena_enabled(void) {
static int g_enable = -1;
if (__builtin_expect(g_enable == -1, 0)) {
const char* e = getenv("HAKMEM_PAGE_ARENA_ENABLE");
g_enable = (e && *e && *e != '0') ? 1 : 0;
#if !HAKMEM_BUILD_RELEASE
if (g_enable) {
fprintf(stderr, "[PageArena-INIT] page_arena_enabled() = %d\n", g_enable);
fflush(stderr);
}
#endif
}
return g_enable;
}
// Hot page cache size (default: 1024)
static inline int page_arena_hot_size(void) {
static int g_size = -1;
if (__builtin_expect(g_size == -1, 0)) {
const char* e = getenv("HAKMEM_PAGE_ARENA_HOT_SIZE");
g_size = (e && *e) ? atoi(e) : PA_HOT_PAGE_DEFAULT_SIZE;
if (g_size < 64) g_size = 64;
if (g_size > 4096) g_size = 4096;
#if !HAKMEM_BUILD_RELEASE
fprintf(stderr, "[PageArena-INIT] hot_size = %d\n", g_size);
fflush(stderr);
#endif
}
return g_size;
}
// Warm 64KB span cache size (default: 256)
static inline int page_arena_warm_64k_size(void) {
static int g_size = -1;
if (__builtin_expect(g_size == -1, 0)) {
const char* e = getenv("HAKMEM_PAGE_ARENA_WARM_64K");
g_size = (e && *e) ? atoi(e) : PA_WARM_64K_DEFAULT_SIZE;
if (g_size < 16) g_size = 16;
if (g_size > 1024) g_size = 1024;
#if !HAKMEM_BUILD_RELEASE
fprintf(stderr, "[PageArena-INIT] warm_64k_size = %d\n", g_size);
fflush(stderr);
#endif
}
return g_size;
}
// Warm 128KB span cache size (default: 128)
static inline int page_arena_warm_128k_size(void) {
static int g_size = -1;
if (__builtin_expect(g_size == -1, 0)) {
const char* e = getenv("HAKMEM_PAGE_ARENA_WARM_128K");
g_size = (e && *e) ? atoi(e) : PA_WARM_128K_DEFAULT_SIZE;
if (g_size < 8) g_size = 8;
if (g_size > 512) g_size = 512;
#if !HAKMEM_BUILD_RELEASE
fprintf(stderr, "[PageArena-INIT] warm_128k_size = %d\n", g_size);
fflush(stderr);
#endif
}
return g_size;
}
// Warm 2MB span cache size (default: 64)
static inline int page_arena_warm_2m_size(void) {
static int g_size = -1;
if (__builtin_expect(g_size == -1, 0)) {
const char* e = getenv("HAKMEM_PAGE_ARENA_WARM_2M");
g_size = (e && *e) ? atoi(e) : PA_WARM_2M_DEFAULT_SIZE;
if (g_size < 4) g_size = 4;
if (g_size > 256) g_size = 256;
#if !HAKMEM_BUILD_RELEASE
fprintf(stderr, "[PageArena-INIT] warm_2m_size = %d\n", g_size);
fflush(stderr);
#endif
}
return g_size;
}
// ============================================================================
// Public API (Box PA3: Unified PageArena)
// ============================================================================
// Initialize PageArena (called at thread start or lazy on first access)
void page_arena_init(PageArena* arena);
// Shutdown PageArena (called at thread exit)
void page_arena_shutdown(PageArena* arena);
// Allocate pages (any size, uses hot/warm/cold based on size)
// Returns: Pointer to allocated pages, or NULL if failed
void* page_arena_alloc_pages(PageArena* arena, size_t size);
// Allocate aligned span (64KB/128KB/2MB aligned)
// Returns: Pointer to allocated span, or NULL if failed
void* page_arena_alloc_aligned(PageArena* arena, size_t size, size_t alignment);
// Free pages (any size, uses hot/warm based on size)
void page_arena_free_pages(PageArena* arena, void* ptr, size_t size);
// Free aligned span (64KB/128KB/2MB aligned)
void page_arena_free_aligned(PageArena* arena, void* ptr, size_t size);
// Print stats (debug only)
void page_arena_print_stats(PageArena* arena);
#endif // HAK_PAGE_ARENA_H