admin/jemalloc-jemalloc
1
0
Fork 0
mirror of https://github.com/jemalloc/jemalloc synced 2026-06-08 15:01:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove medium size classes.

Browse Source 
Remove medium size classes, because concurrent dirty page purging is
no longer capable of purging inactive dirty pages inside active runs
(due to recent arena/bin locking changes).

Enhance tcache to support caching large objects, so that the same range
of size classes is still cached, despite the removal of medium size
class support.
This commit is contained in:
Jason Evans
2010-03-17 16:27:39 -07:00
parent e69bee01de
commit dafde14e08
12 changed files with 514 additions and 483 deletions
Download Patch File Download Diff File Expand all files Collapse all files
jemalloc/src/jemalloc.c
+19 -51
View File
@@ -52,17 +52,9 @@
* | | | 3584 |
* | | | 3840 |
* |========================================|
* | Medium | 4 KiB |
* | | 6 KiB |
* | Large | 4 KiB |
* | | 8 KiB |
* | | ... |
* | | 28 KiB |
* | | 30 KiB |
* | | 32 KiB |
* |========================================|
* | Large | 36 KiB |
* | | 40 KiB |
* | | 44 KiB |
* | | 12 KiB |
* | | ... |
* | | 1012 KiB |
* | | 1016 KiB |
@@ -76,9 +68,8 @@
*
* Different mechanisms are used accoding to category:
*
* Small/medium : Each size class is segregated into its own set of runs.
* Each run maintains a bitmap of which regions are
* free/allocated.
* Small: Each size class is segregated into its own set of runs. Each run
* maintains a bitmap of which regions are free/allocated.
*
* Large : Each allocation is backed by a dedicated run. Metadata are stored
* in the associated arena chunk header maps.
@@ -252,6 +243,11 @@ stats_print_atexit(void)
if (arena != NULL) {
tcache_t *tcache;
/*
* tcache_stats_merge() locks bins, so if any code is
* introduced that acquires both arena and bin locks in
* the opposite order, deadlocks may result.
*/
malloc_mutex_lock(&arena->lock);
ql_foreach(tcache, &arena->tcache_ql, link) {
tcache_stats_merge(tcache, arena);
@@ -510,14 +506,10 @@ MALLOC_OUT:
case 'k':
/*
* Chunks always require at least one
* header page, plus enough room to
* hold a run for the largest medium
* size class (one page more than the
* size).
* header page, plus one data page.
*/
if ((1U << (opt_lg_chunk - 1)) >=
(2U << PAGE_SHIFT) + (1U <<
opt_lg_medium_max))
(2U << PAGE_SHIFT))
opt_lg_chunk--;
break;
case 'K':
@@ -533,15 +525,17 @@ MALLOC_OUT:
opt_prof_leak = true;
break;
#endif
#ifdef JEMALLOC_TCACHE
case 'm':
if (opt_lg_medium_max > PAGE_SHIFT)
opt_lg_medium_max--;
if (opt_lg_tcache_maxclass >= 0)
opt_lg_tcache_maxclass--;
break;
case 'M':
if (opt_lg_medium_max + 1 <
opt_lg_chunk)
opt_lg_medium_max++;
if (opt_lg_tcache_maxclass + 1 <
(sizeof(size_t) << 3))
opt_lg_tcache_maxclass++;
break;
#endif
case 'n':
opt_narenas_lshift--;
break;
@@ -725,33 +719,7 @@ MALLOC_OUT:
* For SMP systems, create more than one arena per CPU by
* default.
*/
#ifdef JEMALLOC_TCACHE
if (opt_tcache
# ifdef JEMALLOC_PROF
/*
* Profile data storage concurrency is directly linked to
* the number of arenas, so only drop the number of arenas
* on behalf of enabled tcache if profiling is disabled.
*/
&& opt_prof == false
# endif
) {
/*
* Only large object allocation/deallocation is
* guaranteed to acquire an arena mutex, so we can get
* away with fewer arenas than without thread caching.
*/
opt_narenas_lshift += 1;
} else {
#endif
/*
* All allocations must acquire an arena mutex, so use
* plenty of arenas.
*/
opt_narenas_lshift += 2;
#ifdef JEMALLOC_TCACHE
}
#endif
opt_narenas_lshift += 2;
}
/* Determine how many arenas to use. */