romenskiy2012/jemalloc
1
0
Fork 0
mirror of https://github.com/jemalloc/jemalloc.git synced 2026-06-17 01:15:39 +03:00
Code Issues Projects Releases Packages Wiki Activity

Cap the base-block growth heuristic

Browse source 
base_block_alloc() grows new base blocks along the page size-class series to
reduce the number of disjoint VM ranges. This works well when new base blocks
are rare.

Under high thread churn, many threads can miss the base free pool in parallel
while allocating metadata.  base_extent_alloc() drops base->mtx after mapping a
new block, but before splitting and inserting the rest into the reuse pool.
Therefore, each parallel miss can map its own block and each completed
allocation then advances base->pind_last.  The result is that small metadata
requests can drive the growth heuristic to increasingly large mmap() sizes, far
beyond the actual allocation demand.

Cap the heuristic growth size at 128 MiB.  This preserves the usual amortization
benefit while bounding the rare pathological case where parallel misses rapidly
advance the growth series. Large individual requests are still honored because
min_block_size continues to override the cap.
This commit is contained in:
guangli-dai 2026-06-12 11:52:11 -07:00 • committed by Guangli Dai
parent 4de3a4c3d1
commit 2f4db8cf5d
1 changed files with 16 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

17
src/base.c
View file

@ -19,6 +19,17 @@ JET_EXTERN ehooks_t *base_ehooks_get_for_metadata(base_t *base);
#define BASE_AUTO_THP_THRESHOLD 2
#define BASE_AUTO_THP_THRESHOLD_A0 5
/*
* Cap the base-block growth heuristic in base_block_alloc(). The
* growth heuristic reduces the number of disjoint VM ranges when new base
* blocks are rare, but high thread churn can cause many parallel misses for
* metadata allocations. Without a cap, those misses can advance
* base->pind_last causing small requests to mmap multi-TiB blocks and exhaust
* the address space. Large individual requests still use min_block_size and
* can exceed this cap.
*/
#define BASE_BLOCK_GROWTH_MAX ((size_t)128 << 20) /* 128 MiB */
/******************************************************************************/
/* Data. */
@ -376,7 +387,11 @@ base_block_alloc(tsdn_t *tsdn, base_t *base, ehooks_t *ehooks, unsigned ind,
pszind_t pind_next = (*pind_last + 1 < sz_psz2ind(SC_LARGE_MAXCLASS))
? *pind_last + 1
: *pind_last;
size_t next_block_size = base_block_size_ceil(sz_pind2sz(pind_next));
size_t next_block_size = base_block_size_ceil(sz_pind2sz(pind_next));
size_t max_block_size = base_block_size_ceil(BASE_BLOCK_GROWTH_MAX);
next_block_size = (next_block_size < max_block_size)
? next_block_size
: max_block_size;
size_t block_size = (min_block_size > next_block_size)
? min_block_size
: next_block_size;