project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
fd5f9e43c3
server-skynet-source-3rd-je.../src/tcache.c

534 lines
13 KiB
C
Raw Normal View History

Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
#define JEMALLOC_TCACHE_C_
Restructure source tree.
2010-02-12 06:45:59 +08:00
#include "jemalloc/internal/jemalloc_internal.h"
Streamline tcache-related malloc/free fast paths. tcache_get() is inlined, so do the config_tcache check inside tcache_get() and simplify its callers. Make arena_malloc() an inline function, since it is part of the malloc() fast path. Remove conditional logic that cause build issues if --disable-tcache was specified.
2012-02-14 04:29:49 +08:00
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
/******************************************************************************/
/* Data. */
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
bool opt_tcache = true;
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
ssize_t opt_lg_tcache_max = LG_TCACHE_MAXCLASS_DEFAULT;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
tcache_bin_info_t *tcache_bin_info;
static unsigned stack_nelms; /* Total stack elms per tcache. */
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
size_t nhbins;
size_t tcache_maxclass;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcaches_t *tcaches;
/* Index of first element within tcaches that has never been used. */
static unsigned tcaches_past;
/* Head of singly linked list tracking available tcaches elements. */
static tcaches_t *tcaches_avail;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
/******************************************************************************/
Make arena_salloc() an inline function.
2012-04-20 09:28:03 +08:00
size_t tcache_salloc(const void *ptr)
{
return (arena_salloc(ptr, false));
}
Optimize malloc() and free() fast paths. Embed the bin index for small page runs into the chunk page map, in order to omit [...] in the following dependent load sequence: ptr-->mapelm-->[run-->bin-->]bin_info Move various non-critcal code out of the inlined function chain into helper functions (tcache_event_hard(), arena_dalloc_small(), and locking).
2012-05-02 15:30:36 +08:00
void
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_event_hard(tsd_t *tsd, tcache_t *tcache)
Optimize malloc() and free() fast paths. Embed the bin index for small page runs into the chunk page map, in order to omit [...] in the following dependent load sequence: ptr-->mapelm-->[run-->bin-->]bin_info Move various non-critcal code out of the inlined function chain into helper functions (tcache_event_hard(), arena_dalloc_small(), and locking).
2012-05-02 15:30:36 +08:00
{
Normalize size classes. Normalize size classes to use the same number of size classes per size doubling (currently hard coded to 4), across the intire range of size classes. Small size classes already used this spacing, but in order to support this change, additional small size classes now fill [4 KiB .. 16 KiB). Large size classes range from [16 KiB .. 4 MiB). Huge size classes now support non-multiples of the chunk size in order to fill (4 MiB .. 16 MiB).
2014-10-06 08:54:10 +08:00
index_t binind = tcache->next_gc_bin;
Optimize malloc() and free() fast paths. Embed the bin index for small page runs into the chunk page map, in order to omit [...] in the following dependent load sequence: ptr-->mapelm-->[run-->bin-->]bin_info Move various non-critcal code out of the inlined function chain into helper functions (tcache_event_hard(), arena_dalloc_small(), and locking).
2012-05-02 15:30:36 +08:00
tcache_bin_t *tbin = &tcache->tbins[binind];
tcache_bin_info_t *tbin_info = &tcache_bin_info[binind];
if (tbin->low_water > 0) {
/*
* Flush (ceiling) 3/4 of the objects below the low water mark.
*/
if (binind < NBINS) {
If MALLOCX_ARENA(a) is specified, use it during tcache fill.
2015-02-14 07:28:56 +08:00
tcache_bin_flush_small(tsd, tcache, tbin, binind,
tbin->ncached - tbin->low_water + (tbin->low_water
>> 2));
Optimize malloc() and free() fast paths. Embed the bin index for small page runs into the chunk page map, in order to omit [...] in the following dependent load sequence: ptr-->mapelm-->[run-->bin-->]bin_info Move various non-critcal code out of the inlined function chain into helper functions (tcache_event_hard(), arena_dalloc_small(), and locking).
2012-05-02 15:30:36 +08:00
} else {
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_bin_flush_large(tsd, tbin, binind, tbin->ncached
- tbin->low_water + (tbin->low_water >> 2), tcache);
Optimize malloc() and free() fast paths. Embed the bin index for small page runs into the chunk page map, in order to omit [...] in the following dependent load sequence: ptr-->mapelm-->[run-->bin-->]bin_info Move various non-critcal code out of the inlined function chain into helper functions (tcache_event_hard(), arena_dalloc_small(), and locking).
2012-05-02 15:30:36 +08:00
}
/*
* Reduce fill count by 2X. Limit lg_fill_div such that the
* fill count is always at least 1.
*/
if ((tbin_info->ncached_max >> (tbin->lg_fill_div+1)) >= 1)
tbin->lg_fill_div++;
} else if (tbin->low_water < 0) {
/*
* Increase fill count by 2X. Make sure lg_fill_div stays
* greater than 0.
*/
if (tbin->lg_fill_div > 1)
tbin->lg_fill_div--;
}
tbin->low_water = tbin->ncached;
tcache->next_gc_bin++;
if (tcache->next_gc_bin == nhbins)
tcache->next_gc_bin = 0;
tcache->ev_cnt = 0;
}
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
void *
If MALLOCX_ARENA(a) is specified, use it during tcache fill.
2015-02-14 07:28:56 +08:00
tcache_alloc_small_hard(tsd_t *tsd, arena_t *arena, tcache_t *tcache,
tcache_bin_t *tbin, index_t binind)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
void *ret;
If MALLOCX_ARENA(a) is specified, use it during tcache fill.
2015-02-14 07:28:56 +08:00
arena_tcache_fill_small(arena, tbin, binind, config_prof ?
tcache->prof_accumbytes : 0);
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
if (config_prof)
tcache->prof_accumbytes = 0;
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
ret = tcache_alloc_easy(tbin);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
return (ret);
}
void
If MALLOCX_ARENA(a) is specified, use it during tcache fill.
2015-02-14 07:28:56 +08:00
tcache_bin_flush_small(tsd_t *tsd, tcache_t *tcache, tcache_bin_t *tbin,
index_t binind, unsigned rem)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_t *arena;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
void *ptr;
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
unsigned i, nflush, ndeferred;
Fix a thread cache stats merging bug. When a thread cache flushes objects to their arenas due to an abundance of cached objects, it merges the allocation request count for the associated size class, and increments a flush counter. If none of the flushed objects came from the thread's assigned arena, then the merging wouldn't happen (though the counter would typically eventually be merged), nor would the flush counter be incremented (a hard bug). Fix this via extra conditional code just after the flush loop.
2011-03-15 03:56:51 +08:00
bool merged_stats = false;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
assert(binind < NBINS);
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
assert(rem <= tbin->ncached);
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena = arena_choose(tsd, NULL);
assert(arena != NULL);
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
for (nflush = tbin->ncached - rem; nflush > 0; nflush = ndeferred) {
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
/* Lock the arena bin associated with the first object. */
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(
tbin->avail[0]);
Integrate whole chunks into unused dirty page purging machinery. Extend per arena unused dirty page purging to manage unused dirty chunks in aaddtion to unused dirty runs. Rather than immediately unmapping deallocated chunks (or purging them in the --disable-munmap case), store them in a separate set of trees, chunks_[sz]ad_dirty. Preferrentially allocate dirty chunks. When excessive unused dirty pages accumulate, purge runs and chunks in ingegrated LRU order (and unmap chunks in the --enable-munmap case). Refactor extent_node_t to provide accessor functions.
2015-02-16 10:04:46 +08:00
arena_t *bin_arena = extent_node_arena_get(&chunk->node);
Fix a regression in tcache_bin_flush_small(). Fix a serious regression in tcache_bin_flush_small() that was introduced by 1cb181ed632e7573fb4eab194e4d216867222d27 (Implement explicit tcache support.).
2015-02-12 16:09:37 +08:00
arena_bin_t *bin = &bin_arena->bins[binind];
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
if (config_prof && bin_arena == arena) {
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
if (arena_prof_accum(arena, tcache->prof_accumbytes))
prof_idump();
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
tcache->prof_accumbytes = 0;
Fix a run initialization race condition. Initialize small run header before dropping arena->lock, arena_chunk_purge() relies on valid small run headers during run iteration. Add some assertions.
2010-03-16 13:25:23 +08:00
}
malloc_mutex_lock(&bin->lock);
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
if (config_stats && bin_arena == arena) {
Convert to uniform style: cond == false --> !cond
2014-10-04 01:16:09 +08:00
assert(!merged_stats);
Fix a thread cache stats merging bug. When a thread cache flushes objects to their arenas due to an abundance of cached objects, it merges the allocation request count for the associated size class, and increments a flush counter. If none of the flushed objects came from the thread's assigned arena, then the merging wouldn't happen (though the counter would typically eventually be merged), nor would the flush counter be incremented (a hard bug). Fix this via extra conditional code just after the flush loop.
2011-03-15 03:56:51 +08:00
merged_stats = true;
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
bin->stats.nflushes++;
bin->stats.nrequests += tbin->tstats.nrequests;
tbin->tstats.nrequests = 0;
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
}
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
ndeferred = 0;
for (i = 0; i < nflush; i++) {
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
ptr = tbin->avail[i];
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
assert(ptr != NULL);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
Integrate whole chunks into unused dirty page purging machinery. Extend per arena unused dirty page purging to manage unused dirty chunks in aaddtion to unused dirty runs. Rather than immediately unmapping deallocated chunks (or purging them in the --disable-munmap case), store them in a separate set of trees, chunks_[sz]ad_dirty. Preferrentially allocate dirty chunks. When excessive unused dirty pages accumulate, purge runs and chunks in ingegrated LRU order (and unmap chunks in the --enable-munmap case). Refactor extent_node_t to provide accessor functions.
2015-02-16 10:04:46 +08:00
if (extent_node_arena_get(&chunk->node) == bin_arena) {
Omit chunk header in arena chunk map. Omit the first map_bias elements of the map in arena_chunk_t. This avoids barely spilling over into an extra chunk header page for common chunk sizes.
2010-10-02 08:35:43 +08:00
size_t pageind = ((uintptr_t)ptr -
Clean up *PAGE* macros. s/PAGE_SHIFT/LG_PAGE/g and s/PAGE_SIZE/PAGE/g. Remove remnants of the dynamic-page-shift code. Rename the "arenas.pagesize" mallctl to "arenas.page". Remove the "arenas.chunksize" mallctl, which is redundant with "opt.lg_chunk".
2012-04-02 22:04:34 +08:00
(uintptr_t)chunk) >> LG_PAGE;
Refactor chunk map. Break the chunk map into two separate arrays, in order to improve cache locality. This is related to issue #23.
2014-08-30 04:34:40 +08:00
arena_chunk_map_bits_t *bitselm =
arena_bitselm_get(chunk, pageind);
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_dalloc_bin_junked_locked(bin_arena, chunk,
Add configure options. Add: --with-lg-page --with-lg-page-sizes --with-lg-size-class-group --with-lg-quantum Get rid of STATIC_PAGE_SHIFT, in favor of directly setting LG_PAGE. Fix various edge conditions exposed by the configure options.
2014-10-10 08:54:06 +08:00
ptr, bitselm);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
} else {
/*
* This object was allocated via a different
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
* arena bin than the one that is currently
* locked. Stash the object, so that it can be
* handled in a future pass.
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
*/
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
tbin->avail[ndeferred] = ptr;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
ndeferred++;
}
}
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
malloc_mutex_unlock(&bin->lock);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
Convert to uniform style: cond == false --> !cond
2014-10-04 01:16:09 +08:00
if (config_stats && !merged_stats) {
Fix a thread cache stats merging bug. When a thread cache flushes objects to their arenas due to an abundance of cached objects, it merges the allocation request count for the associated size class, and increments a flush counter. If none of the flushed objects came from the thread's assigned arena, then the merging wouldn't happen (though the counter would typically eventually be merged), nor would the flush counter be incremented (a hard bug). Fix this via extra conditional code just after the flush loop.
2011-03-15 03:56:51 +08:00
/*
* The flush loop didn't happen to flush to this thread's
* arena, so the stats didn't get merged. Manually do so now.
*/
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_bin_t *bin = &arena->bins[binind];
Fix a thread cache stats merging bug. When a thread cache flushes objects to their arenas due to an abundance of cached objects, it merges the allocation request count for the associated size class, and increments a flush counter. If none of the flushed objects came from the thread's assigned arena, then the merging wouldn't happen (though the counter would typically eventually be merged), nor would the flush counter be incremented (a hard bug). Fix this via extra conditional code just after the flush loop.
2011-03-15 03:56:51 +08:00
malloc_mutex_lock(&bin->lock);
bin->stats.nflushes++;
bin->stats.nrequests += tbin->tstats.nrequests;
tbin->tstats.nrequests = 0;
malloc_mutex_unlock(&bin->lock);
}
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
memmove(tbin->avail, &tbin->avail[tbin->ncached - rem],
rem * sizeof(void *));
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
tbin->ncached = rem;
Dynamically adjust tcache fill count. Dynamically adjust tcache fill count (number of objects allocated per tcache refill) such that if GC has to flush inactive objects, the fill count gradually decreases. Conversely, if refills occur while the fill count is depressed, the fill count gradually increases back to its maximum value.
2011-03-21 15:18:17 +08:00
if ((int)tbin->ncached < tbin->low_water)
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
tbin->low_water = tbin->ncached;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
void
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_bin_flush_large(tsd_t *tsd, tcache_bin_t *tbin, index_t binind,
unsigned rem, tcache_t *tcache)
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
{
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_t *arena;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
void *ptr;
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
unsigned i, nflush, ndeferred;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
bool merged_stats = false;
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
assert(binind < nhbins);
assert(rem <= tbin->ncached);
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena = arena_choose(tsd, NULL);
assert(arena != NULL);
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
for (nflush = tbin->ncached - rem; nflush > 0; nflush = ndeferred) {
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
/* Lock the arena associated with the first object. */
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(
tbin->avail[0]);
Integrate whole chunks into unused dirty page purging machinery. Extend per arena unused dirty page purging to manage unused dirty chunks in aaddtion to unused dirty runs. Rather than immediately unmapping deallocated chunks (or purging them in the --disable-munmap case), store them in a separate set of trees, chunks_[sz]ad_dirty. Preferrentially allocate dirty chunks. When excessive unused dirty pages accumulate, purge runs and chunks in ingegrated LRU order (and unmap chunks in the --enable-munmap case). Refactor extent_node_t to provide accessor functions.
2015-02-16 10:04:46 +08:00
arena_t *locked_arena = extent_node_arena_get(&chunk->node);
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
UNUSED bool idump;
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
if (config_prof)
idump = false;
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
malloc_mutex_lock(&locked_arena->lock);
if ((config_prof || config_stats) && locked_arena == arena) {
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
if (config_prof) {
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
idump = arena_prof_accum_locked(arena,
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
tcache->prof_accumbytes);
tcache->prof_accumbytes = 0;
}
if (config_stats) {
merged_stats = true;
arena->stats.nrequests_large +=
tbin->tstats.nrequests;
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
arena->stats.lstats[binind - NBINS].nrequests +=
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
tbin->tstats.nrequests;
tbin->tstats.nrequests = 0;
}
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
}
ndeferred = 0;
for (i = 0; i < nflush; i++) {
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
ptr = tbin->avail[i];
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
assert(ptr != NULL);
chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
Integrate whole chunks into unused dirty page purging machinery. Extend per arena unused dirty page purging to manage unused dirty chunks in aaddtion to unused dirty runs. Rather than immediately unmapping deallocated chunks (or purging them in the --disable-munmap case), store them in a separate set of trees, chunks_[sz]ad_dirty. Preferrentially allocate dirty chunks. When excessive unused dirty pages accumulate, purge runs and chunks in ingegrated LRU order (and unmap chunks in the --enable-munmap case). Refactor extent_node_t to provide accessor functions.
2015-02-16 10:04:46 +08:00
if (extent_node_arena_get(&chunk->node) ==
locked_arena) {
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_dalloc_large_junked_locked(locked_arena,
chunk, ptr);
Add configure options. Add: --with-lg-page --with-lg-page-sizes --with-lg-size-class-group --with-lg-quantum Get rid of STATIC_PAGE_SHIFT, in favor of directly setting LG_PAGE. Fix various edge conditions exposed by the configure options.
2014-10-10 08:54:06 +08:00
} else {
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
/*
* This object was allocated via a different
* arena than the one that is currently locked.
* Stash the object, so that it can be handled
* in a future pass.
*/
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
tbin->avail[ndeferred] = ptr;
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
ndeferred++;
}
}
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
malloc_mutex_unlock(&locked_arena->lock);
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
if (config_prof && idump)
prof_idump();
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
}
Convert to uniform style: cond == false --> !cond
2014-10-04 01:16:09 +08:00
if (config_stats && !merged_stats) {
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
/*
* The flush loop didn't happen to flush to this thread's
* arena, so the stats didn't get merged. Manually do so now.
*/
malloc_mutex_lock(&arena->lock);
arena->stats.nrequests_large += tbin->tstats.nrequests;
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
arena->stats.lstats[binind - NBINS].nrequests +=
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
tbin->tstats.nrequests;
tbin->tstats.nrequests = 0;
malloc_mutex_unlock(&arena->lock);
}
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
memmove(tbin->avail, &tbin->avail[tbin->ncached - rem],
rem * sizeof(void *));
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
tbin->ncached = rem;
Dynamically adjust tcache fill count. Dynamically adjust tcache fill count (number of objects allocated per tcache refill) such that if GC has to flush inactive objects, the fill count gradually decreases. Conversely, if refills occur while the fill count is depressed, the fill count gradually increases back to its maximum value.
2011-03-21 15:18:17 +08:00
if ((int)tbin->ncached < tbin->low_water)
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
tbin->low_water = tbin->ncached;
}
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
void
tcache_arena_associate(tcache_t *tcache, arena_t *arena)
{
if (config_stats) {
/* Link into list of extant tcaches. */
malloc_mutex_lock(&arena->lock);
ql_elm_new(tcache, link);
ql_tail_insert(&arena->tcache_ql, tcache, link);
malloc_mutex_unlock(&arena->lock);
}
}
Refactor/fix arenas manipulation. Abstract arenas access to use arena_get() (or a0get() where appropriate) rather than directly reading e.g. arenas[ind]. Prior to the addition of the arenas.extend mallctl, the worst possible outcome of directly accessing arenas was a stale read, but arenas.extend may allocate and assign a new array to arenas. Add a tsd-based arenas_cache, which amortizes arenas reads. This introduces some subtle bootstrapping issues, with tsd_boot() now being split into tsd_boot[01]() to support tsd wrapper allocation bootstrapping, as well as an arenas_cache_bypass tsd variable which dynamically terminates allocation of arenas_cache itself. Promote a0malloc(), a0calloc(), and a0free() to be generally useful for internal allocation, and use them in several places (more may be appropriate). Abstract arena->nthreads management and fix a missing decrement during thread destruction (recent tsd refactoring left arenas_cleanup() unused). Change arena_choose() to propagate OOM, and handle OOM in all callers. This is important for providing consistent allocation behavior when the MALLOCX_ARENA() flag is being used. Prior to this fix, it was possible for an OOM to result in allocation silently allocating from a different arena than the one specified.
2014-10-08 14:14:57 +08:00
void
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_arena_reassociate(tcache_t *tcache, arena_t *oldarena, arena_t *newarena)
Refactor/fix arenas manipulation. Abstract arenas access to use arena_get() (or a0get() where appropriate) rather than directly reading e.g. arenas[ind]. Prior to the addition of the arenas.extend mallctl, the worst possible outcome of directly accessing arenas was a stale read, but arenas.extend may allocate and assign a new array to arenas. Add a tsd-based arenas_cache, which amortizes arenas reads. This introduces some subtle bootstrapping issues, with tsd_boot() now being split into tsd_boot[01]() to support tsd wrapper allocation bootstrapping, as well as an arenas_cache_bypass tsd variable which dynamically terminates allocation of arenas_cache itself. Promote a0malloc(), a0calloc(), and a0free() to be generally useful for internal allocation, and use them in several places (more may be appropriate). Abstract arena->nthreads management and fix a missing decrement during thread destruction (recent tsd refactoring left arenas_cleanup() unused). Change arena_choose() to propagate OOM, and handle OOM in all callers. This is important for providing consistent allocation behavior when the MALLOCX_ARENA() flag is being used. Prior to this fix, it was possible for an OOM to result in allocation silently allocating from a different arena than the one specified.
2014-10-08 14:14:57 +08:00
{
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_arena_dissociate(tcache, oldarena);
tcache_arena_associate(tcache, newarena);
Refactor/fix arenas manipulation. Abstract arenas access to use arena_get() (or a0get() where appropriate) rather than directly reading e.g. arenas[ind]. Prior to the addition of the arenas.extend mallctl, the worst possible outcome of directly accessing arenas was a stale read, but arenas.extend may allocate and assign a new array to arenas. Add a tsd-based arenas_cache, which amortizes arenas reads. This introduces some subtle bootstrapping issues, with tsd_boot() now being split into tsd_boot[01]() to support tsd wrapper allocation bootstrapping, as well as an arenas_cache_bypass tsd variable which dynamically terminates allocation of arenas_cache itself. Promote a0malloc(), a0calloc(), and a0free() to be generally useful for internal allocation, and use them in several places (more may be appropriate). Abstract arena->nthreads management and fix a missing decrement during thread destruction (recent tsd refactoring left arenas_cleanup() unused). Change arena_choose() to propagate OOM, and handle OOM in all callers. This is important for providing consistent allocation behavior when the MALLOCX_ARENA() flag is being used. Prior to this fix, it was possible for an OOM to result in allocation silently allocating from a different arena than the one specified.
2014-10-08 14:14:57 +08:00
}
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
void
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_arena_dissociate(tcache_t *tcache, arena_t *arena)
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
{
if (config_stats) {
/* Unlink from list of extant tcaches. */
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
malloc_mutex_lock(&arena->lock);
if (config_debug) {
bool in_ql = false;
tcache_t *iter;
ql_foreach(iter, &arena->tcache_ql, link) {
if (iter == tcache) {
in_ql = true;
break;
}
}
assert(in_ql);
}
ql_remove(&arena->tcache_ql, tcache, link);
tcache_stats_merge(tcache, arena);
malloc_mutex_unlock(&arena->lock);
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
}
}
outline rare tcache_get codepaths
2014-04-16 04:28:37 +08:00
tcache_t *
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
tcache_get_hard(tsd_t *tsd)
outline rare tcache_get codepaths
2014-04-16 04:28:37 +08:00
{
Refactor/fix arenas manipulation. Abstract arenas access to use arena_get() (or a0get() where appropriate) rather than directly reading e.g. arenas[ind]. Prior to the addition of the arenas.extend mallctl, the worst possible outcome of directly accessing arenas was a stale read, but arenas.extend may allocate and assign a new array to arenas. Add a tsd-based arenas_cache, which amortizes arenas reads. This introduces some subtle bootstrapping issues, with tsd_boot() now being split into tsd_boot[01]() to support tsd wrapper allocation bootstrapping, as well as an arenas_cache_bypass tsd variable which dynamically terminates allocation of arenas_cache itself. Promote a0malloc(), a0calloc(), and a0free() to be generally useful for internal allocation, and use them in several places (more may be appropriate). Abstract arena->nthreads management and fix a missing decrement during thread destruction (recent tsd refactoring left arenas_cleanup() unused). Change arena_choose() to propagate OOM, and handle OOM in all callers. This is important for providing consistent allocation behavior when the MALLOCX_ARENA() flag is being used. Prior to this fix, it was possible for an OOM to result in allocation silently allocating from a different arena than the one specified.
2014-10-08 14:14:57 +08:00
arena_t *arena;
outline rare tcache_get codepaths
2014-04-16 04:28:37 +08:00
Convert to uniform style: cond == false --> !cond
2014-10-04 01:16:09 +08:00
if (!tcache_enabled_get()) {
Fix tsd cleanup regressions. Fix tsd cleanup regressions that were introduced in 5460aa6f6676c7f253bfcb75c028dfd38cae8aaf (Convert all tsd variables to reside in a single tsd structure.). These regressions were twofold: 1) tsd_tryget() should never (and need never) return NULL. Rename it to tsd_fetch() and simplify all callers. 2) tsd_*_set() must only be called when tsd is in the nominal state, because cleanup happens during the nominal-->purgatory transition, and re-initialization must not happen while in the purgatory state. Add tsd_nominal() and use it as needed. Note that tsd_*{p,}_get() can still be used as long as no re-initialization that would require cleanup occurs. This means that e.g. the thread_allocated counter can be updated unconditionally.
2014-10-05 02:12:53 +08:00
if (tsd_nominal(tsd))
tcache_enabled_set(false); /* Memoize. */
outline rare tcache_get codepaths
2014-04-16 04:28:37 +08:00
return (NULL);
}
Refactor/fix arenas manipulation. Abstract arenas access to use arena_get() (or a0get() where appropriate) rather than directly reading e.g. arenas[ind]. Prior to the addition of the arenas.extend mallctl, the worst possible outcome of directly accessing arenas was a stale read, but arenas.extend may allocate and assign a new array to arenas. Add a tsd-based arenas_cache, which amortizes arenas reads. This introduces some subtle bootstrapping issues, with tsd_boot() now being split into tsd_boot[01]() to support tsd wrapper allocation bootstrapping, as well as an arenas_cache_bypass tsd variable which dynamically terminates allocation of arenas_cache itself. Promote a0malloc(), a0calloc(), and a0free() to be generally useful for internal allocation, and use them in several places (more may be appropriate). Abstract arena->nthreads management and fix a missing decrement during thread destruction (recent tsd refactoring left arenas_cleanup() unused). Change arena_choose() to propagate OOM, and handle OOM in all callers. This is important for providing consistent allocation behavior when the MALLOCX_ARENA() flag is being used. Prior to this fix, it was possible for an OOM to result in allocation silently allocating from a different arena than the one specified.
2014-10-08 14:14:57 +08:00
arena = arena_choose(tsd, NULL);
if (unlikely(arena == NULL))
return (NULL);
Add configure options. Add: --with-lg-page --with-lg-page-sizes --with-lg-size-class-group --with-lg-quantum Get rid of STATIC_PAGE_SHIFT, in favor of directly setting LG_PAGE. Fix various edge conditions exposed by the configure options.
2014-10-10 08:54:06 +08:00
return (tcache_create(tsd, arena));
outline rare tcache_get codepaths
2014-04-16 04:28:37 +08:00
}
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
tcache_t *
Add configure options. Add: --with-lg-page --with-lg-page-sizes --with-lg-size-class-group --with-lg-quantum Get rid of STATIC_PAGE_SHIFT, in favor of directly setting LG_PAGE. Fix various edge conditions exposed by the configure options.
2014-10-10 08:54:06 +08:00
tcache_create(tsd_t *tsd, arena_t *arena)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
tcache_t *tcache;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
size_t size, stack_offset;
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
unsigned i;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Use offsetof() when sizing dynamic structures. Base dynamic structure size on offsetof(), rather than subtracting the size of the dynamic structure member. Results could differ on systems with strict data structure alignment requirements.
2010-10-02 09:02:43 +08:00
size = offsetof(tcache_t, tbins) + (sizeof(tcache_bin_t) * nhbins);
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
/* Naturally align the pointer stacks. */
size = PTR_CEILING(size);
stack_offset = size;
size += stack_nelms * sizeof(void *);
Add configure options. Add: --with-lg-page --with-lg-page-sizes --with-lg-size-class-group --with-lg-quantum Get rid of STATIC_PAGE_SHIFT, in favor of directly setting LG_PAGE. Fix various edge conditions exposed by the configure options.
2014-10-10 08:54:06 +08:00
/* Avoid false cacheline sharing. */
size = sa2u(size, CACHELINE);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache = ipallocztm(tsd, size, CACHELINE, true, false, true, a0get());
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
if (tcache == NULL)
return (NULL);
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
tcache_arena_associate(tcache, arena);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
assert((TCACHE_NSLOTS_SMALL_MAX & 1U) == 0);
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
for (i = 0; i < nhbins; i++) {
Dynamically adjust tcache fill count. Dynamically adjust tcache fill count (number of objects allocated per tcache refill) such that if GC has to flush inactive objects, the fill count gradually decreases. Conversely, if refills occur while the fill count is depressed, the fill count gradually increases back to its maximum value.
2011-03-21 15:18:17 +08:00
tcache->tbins[i].lg_fill_div = 1;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
tcache->tbins[i].avail = (void **)((uintptr_t)tcache +
(uintptr_t)stack_offset);
stack_offset += tcache_bin_info[i].ncached_max * sizeof(void *);
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
}
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
return (tcache);
}
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
static void
tcache_destroy(tsd_t *tsd, tcache_t *tcache)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_t *arena;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
unsigned i;
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena = arena_choose(tsd, NULL);
tcache_arena_dissociate(tcache, arena);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
for (i = 0; i < NBINS; i++) {
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
tcache_bin_t *tbin = &tcache->tbins[i];
If MALLOCX_ARENA(a) is specified, use it during tcache fill.
2015-02-14 07:28:56 +08:00
tcache_bin_flush_small(tsd, tcache, tbin, i, 0);
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
if (config_stats && tbin->tstats.nrequests != 0) {
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
arena_bin_t *bin = &arena->bins[i];
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
malloc_mutex_lock(&bin->lock);
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
bin->stats.nrequests += tbin->tstats.nrequests;
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
malloc_mutex_unlock(&bin->lock);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
}
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
for (; i < nhbins; i++) {
tcache_bin_t *tbin = &tcache->tbins[i];
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
tcache_bin_flush_large(tsd, tbin, i, 0, tcache);
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
if (config_stats && tbin->tstats.nrequests != 0) {
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
malloc_mutex_lock(&arena->lock);
arena->stats.nrequests_large += tbin->tstats.nrequests;
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
arena->stats.lstats[i - NBINS].nrequests +=
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
tbin->tstats.nrequests;
malloc_mutex_unlock(&arena->lock);
}
}
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
if (config_prof && tcache->prof_accumbytes > 0 &&
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
arena_prof_accum(arena, tcache->prof_accumbytes))
Fix a prof-related locking order bug. Fix a locking order bug that could cause deadlock during fork if heap profiling were enabled.
2013-02-07 03:59:30 +08:00
prof_idump();
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
Implement metadata statistics. There are three categories of metadata: - Base allocations are used for bootstrap-sensitive internal allocator data structures. - Arena chunk headers comprise pages which track the states of the non-metadata pages. - Internal allocations differ from application-originated allocations in that they are for internal use, and that they are omitted from heap profiles. The metadata statistics comprise the metadata categories as follows: - stats.metadata: All metadata -- base + arena chunk headers + internal allocations. - stats.arenas.<i>.metadata.mapped: Arena chunk headers. - stats.arenas.<i>.metadata.allocated: Internal allocations. This is reported separately from the other metadata statistics because it overlaps with the allocated and active statistics, whereas the other metadata statistics do not. Base allocations are not reported separately, though their magnitude can be computed by subtracting the arena-specific metadata. This resolves #163.
2014-11-28 03:22:36 +08:00
idalloctm(tsd, tcache, false, true);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
void
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
tcache_cleanup(tsd_t *tsd)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
tcache_t *tcache;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
if (!config_tcache)
return;
if ((tcache = tsd_tcache_get(tsd)) != NULL) {
tcache_destroy(tsd, tcache);
tsd_tcache_set(tsd, NULL);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
}
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
void
tcache_enabled_cleanup(tsd_t *tsd)
{
/* Do nothing. */
}
Fix a data race for large allocation stats counters. Reported by Pat Lynch.
2013-10-22 06:00:06 +08:00
/* Caller must own arena->lock. */
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
void
tcache_stats_merge(tcache_t *tcache, arena_t *arena)
{
unsigned i;
Fix a data race for large allocation stats counters. Reported by Pat Lynch.
2013-10-22 06:00:06 +08:00
cassert(config_stats);
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
/* Merge and reset tcache stats. */
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
for (i = 0; i < NBINS; i++) {
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
arena_bin_t *bin = &arena->bins[i];
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
tcache_bin_t *tbin = &tcache->tbins[i];
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
malloc_mutex_lock(&bin->lock);
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
bin->stats.nrequests += tbin->tstats.nrequests;
Push locks into arena bins. For bin-related allocation, protect data structures with bin locks rather than arena locks. Arena locks remain for run allocation/deallocation and other miscellaneous operations. Restructure statistics counters to maintain per bin allocated/nmalloc/ndalloc, but continue to provide arena-wide statistics via aggregation in the ctl code.
2010-03-14 12:32:56 +08:00
malloc_mutex_unlock(&bin->lock);
Simplify tcache object caching. Use chains of cached objects, rather than using arrays of pointers. Since tcache_bin_t is no longer dynamically sized, convert tcache_t's tbin to an array of structures, rather than an array of pointers. This implicitly removes tcache_bin_{create,destroy}(), which further simplifies the fast path for malloc/free. Use cacheline alignment for tcache_t allocations. Remove runtime configuration option for number of tcache bin slots, and replace it with a boolean option for enabling/disabling tcache. Limit the number of tcache objects to the lesser of TCACHE_NSLOTS_MAX and 2X the number of regions per run for the size class. For GC-triggered flush, discard 3/4 of the objects below the low water mark, rather than 1/2.
2010-03-08 07:34:14 +08:00
tbin->tstats.nrequests = 0;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
for (; i < nhbins; i++) {
Simplify small size class infrastructure. Program-generate small size class tables for all valid combinations of LG_TINY_MIN, LG_QUANTUM, and PAGE_SHIFT. Use the appropriate table to generate all relevant data structures, and remove the distinction between tiny/quantum/cacheline/subpage bins. Remove --enable-dynamic-page-shift. This option didn't prove useful in practice, and it prevented optimizations. Add Tilera architecture support.
2012-02-29 08:50:47 +08:00
malloc_large_stats_t *lstats = &arena->stats.lstats[i - NBINS];
Remove medium size classes. 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.
2010-03-18 07:27:39 +08:00
tcache_bin_t *tbin = &tcache->tbins[i];
arena->stats.nrequests_large += tbin->tstats.nrequests;
lstats->nrequests += tbin->tstats.nrequests;
tbin->tstats.nrequests = 0;
}
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
}
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
bool
tcaches_create(tsd_t *tsd, unsigned *r_ind)
{
tcache_t *tcache;
tcaches_t *elm;
if (tcaches == NULL) {
tcaches = base_alloc(sizeof(tcache_t *) *
(MALLOCX_TCACHE_MAX+1));
if (tcaches == NULL)
return (true);
}
if (tcaches_avail == NULL && tcaches_past > MALLOCX_TCACHE_MAX)
return (true);
tcache = tcache_create(tsd, a0get());
if (tcache == NULL)
return (true);
if (tcaches_avail != NULL) {
elm = tcaches_avail;
tcaches_avail = tcaches_avail->next;
elm->tcache = tcache;
Test and fix tcache ID recycling.
2015-02-11 01:03:48 +08:00
*r_ind = elm - tcaches;
Implement explicit tcache support. Add the MALLOCX_TCACHE() and MALLOCX_TCACHE_NONE macros, which can be used in conjunction with the *allocx() API. Add the tcache.create, tcache.flush, and tcache.destroy mallctls. This resolves #145.
2015-01-30 07:30:47 +08:00
} else {
elm = &tcaches[tcaches_past];
elm->tcache = tcache;
*r_ind = tcaches_past;
tcaches_past++;
}
return (false);
}
static void
tcaches_elm_flush(tsd_t *tsd, tcaches_t *elm)
{
if (elm->tcache == NULL)
return;
tcache_destroy(tsd, elm->tcache);
elm->tcache = NULL;
}
void
tcaches_flush(tsd_t *tsd, unsigned ind)
{
tcaches_elm_flush(tsd, &tcaches[ind]);
}
void
tcaches_destroy(tsd_t *tsd, unsigned ind)
{
tcaches_t *elm = &tcaches[ind];
tcaches_elm_flush(tsd, elm);
elm->next = tcaches_avail;
tcaches_avail = elm;
}
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
bool
Convert all tsd variables to reside in a single tsd structure.
2014-09-23 12:09:23 +08:00
tcache_boot(void)
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
{
Always initialize tcache data structures. Always initialize tcache data structures if the tcache configuration option is enabled, regardless of opt_tcache. This fixes "thread.tcache.enabled" mallctl manipulation in the case when opt_tcache is false.
2012-04-07 03:41:55 +08:00
unsigned i;
Refactor jemalloc.c into multiple source files. Fix a stats bug in large object curruns accounting. Replace tcache_bin_fill() with arena_tcache_fill(), and fix a bug in an OOM error path. Fix API name mangling to coexist with __attribute__((malloc)).
2010-01-17 01:53:50 +08:00
Always initialize tcache data structures. Always initialize tcache data structures if the tcache configuration option is enabled, regardless of opt_tcache. This fixes "thread.tcache.enabled" mallctl manipulation in the case when opt_tcache is false.
2012-04-07 03:41:55 +08:00
/*
* If necessary, clamp opt_lg_tcache_max, now that arena_maxclass is
* known.
*/
if (opt_lg_tcache_max < 0 || (1U << opt_lg_tcache_max) < SMALL_MAXCLASS)
tcache_maxclass = SMALL_MAXCLASS;
else if ((1U << opt_lg_tcache_max) > arena_maxclass)
tcache_maxclass = arena_maxclass;
else
tcache_maxclass = (1U << opt_lg_tcache_max);
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
Always initialize tcache data structures. Always initialize tcache data structures if the tcache configuration option is enabled, regardless of opt_tcache. This fixes "thread.tcache.enabled" mallctl manipulation in the case when opt_tcache is false.
2012-04-07 03:41:55 +08:00
nhbins = NBINS + (tcache_maxclass >> LG_PAGE);
/* Initialize tcache_bin_info. */
tcache_bin_info = (tcache_bin_info_t *)base_alloc(nhbins *
sizeof(tcache_bin_info_t));
if (tcache_bin_info == NULL)
return (true);
stack_nelms = 0;
for (i = 0; i < NBINS; i++) {
if ((arena_bin_info[i].nregs << 1) <= TCACHE_NSLOTS_SMALL_MAX) {
tcache_bin_info[i].ncached_max =
(arena_bin_info[i].nregs << 1);
} else {
tcache_bin_info[i].ncached_max =
TCACHE_NSLOTS_SMALL_MAX;
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
}
Always initialize tcache data structures. Always initialize tcache data structures if the tcache configuration option is enabled, regardless of opt_tcache. This fixes "thread.tcache.enabled" mallctl manipulation in the case when opt_tcache is false.
2012-04-07 03:41:55 +08:00
stack_nelms += tcache_bin_info[i].ncached_max;
}
for (; i < nhbins; i++) {
tcache_bin_info[i].ncached_max = TCACHE_NSLOTS_LARGE;
stack_nelms += tcache_bin_info[i].ncached_max;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
}
Use bitmaps to track small regions. The previous free list implementation, which embedded singly linked lists in available regions, had the unfortunate side effect of causing many cache misses during thread cache fills. Fix this in two places: - arena_run_t: Use a new bitmap implementation to track which regions are available. Furthermore, revert to preferring the lowest available region (as jemalloc did with its old bitmap-based approach). - tcache_t: Move read-only tcache_bin_t metadata into tcache_bin_info_t, and add a contiguous array of pointers to tcache_t in order to track cached objects. This substantially increases the size of tcache_t, but results in much higher data locality for common tcache operations. As a side benefit, it is again possible to efficiently flush the least recently used cached objects, so this change changes flushing from MRU to LRU. The new bitmap implementation uses a multi-level summary approach to make finding the lowest available region very fast. In practice, bitmaps only have one or two levels, though the implementation is general enough to handle extremely large bitmaps, mainly so that large page sizes can still be entertained. Fix tcache_bin_flush_large() to always flush statistics, in the same way that tcache_bin_flush_small() was recently fixed. Use JEMALLOC_DEBUG rather than NDEBUG. Add dassert(), and use it for debug-only asserts.
2011-03-17 01:30:13 +08:00
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
return (false);
}
Reference in New Issue Copy Permalink