project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Move small run metadata into the arena chunk header.

Browse Source 
Move small run metadata into the arena chunk header, with multiple
expected benefits:
- Lower run fragmentation due to reduced run sizes; runs are more likely
  to completely drain when there are fewer total regions.
- Improved cache behavior.  Prior to this change, run headers were
  always page-aligned, which put extra pressure on some CPU cache sets.
  The degree to which this was a problem was hardware dependent, but it
  likely hurt some even for the most advanced modern hardware.
- Buffer overruns/underruns are less likely to corrupt allocator
  metadata.
- Size classes between 4 KiB and 16 KiB become reasonable to support
  without any special handling, and the runs are small enough that dirty
  unused pages aren't a significant concern.
This commit is contained in:
Jason Evans
2014-09-29 01:31:39 -07:00
parent f97e5ac4ec
commit 0c5dd03e88
3 changed files with 232 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

144
include/jemalloc/internal/arena.h
View File

@@ -1,30 +1,8 @@
/******************************************************************************/
#ifdef JEMALLOC_H_TYPES
/*
* RUN_MAX_OVRHD indicates maximum desired run header overhead. Runs are sized
* as small as possible such that this setting is still honored, without
* violating other constraints. The goal is to make runs as small as possible
* without exceeding a per run external fragmentation threshold.
*
* We use binary fixed point math for overhead computations, where the binary
* point is implicitly RUN_BFP bits to the left.
*
* Note that it is possible to set RUN_MAX_OVRHD low enough that it cannot be
* honored for some/all object sizes, since when heap profiling is enabled
* there is one pointer of header overhead per object (plus a constant). This
* constraint is relaxed (ignored) for runs that are so small that the
* per-region overhead is greater than:
*
* (RUN_MAX_OVRHD / (reg_interval << (3+RUN_BFP))
*/
#define RUN_BFP 12
/* \/ Implicit binary fixed point. */
#define RUN_MAX_OVRHD 0x0000003dU
#define RUN_MAX_OVRHD_RELAX 0x00001800U
/* Maximum number of regions in one run. */
#define LG_RUN_MAXREGS 11
#define LG_RUN_MAXREGS (LG_PAGE - LG_TINY_MIN)
#define RUN_MAXREGS (1U << LG_RUN_MAXREGS)
/*
@@ -43,10 +21,10 @@
*/
#define LG_DIRTY_MULT_DEFAULT 3
typedef struct arena_run_s arena_run_t;
typedef struct arena_chunk_map_bits_s arena_chunk_map_bits_t;
typedef struct arena_chunk_map_misc_s arena_chunk_map_misc_t;
typedef struct arena_chunk_s arena_chunk_t;
typedef struct arena_run_s arena_run_t;
typedef struct arena_bin_info_s arena_bin_info_t;
typedef struct arena_bin_s arena_bin_t;
typedef struct arena_s arena_t;
@@ -55,6 +33,20 @@ typedef struct arena_s arena_t;
/******************************************************************************/
#ifdef JEMALLOC_H_STRUCTS
struct arena_run_s {
/* Bin this run is associated with. */
arena_bin_t *bin;
/* Index of next region that has never been allocated, or nregs. */
uint32_t nextind;
/* Number of free regions in run. */
unsigned nfree;
/* Per region allocated/deallocated bitmap. */
bitmap_t bitmap[BITMAP_GROUPS_MAX];
};
/* Each element of the chunk map corresponds to one page within the chunk. */
struct arena_chunk_map_bits_s {
/*
@@ -130,15 +122,6 @@ struct arena_chunk_map_bits_s {
* chunk header in order to improve cache locality.
*/
struct arena_chunk_map_misc_s {
#ifndef JEMALLOC_PROF
/*
* Overlay prof_tctx in order to allow it to be referenced by dead code.
* Such antics aren't warranted for per arena data structures, but
* chunk map overhead accounts for a percentage of memory, rather than
* being just a fixed cost.
*/
union {
#endif
/*
* Linkage for run trees. There are two disjoint uses:
*
@@ -146,16 +129,18 @@ struct arena_chunk_map_misc_s {
* 2) arena_run_t conceptually uses this linkage for in-use non-full
* runs, rather than directly embedding linkage.
*/
rb_node(arena_chunk_map_misc_t) rb_link;
rb_node(arena_chunk_map_misc_t) rb_link;
/* Profile counters, used for large object runs. */
prof_tctx_t *prof_tctx;
#ifndef JEMALLOC_PROF
}; /* union { ... }; */
#endif
union {
/* Linkage for list of dirty runs. */
ql_elm(arena_chunk_map_misc_t) dr_link;
/* Linkage for list of dirty runs. */
ql_elm(arena_chunk_map_misc_t) dr_link;
/* Profile counters, used for large object runs. */
prof_tctx_t *prof_tctx;
/* Small region run metadata. */
arena_run_t run;
};
};
typedef rb_tree(arena_chunk_map_misc_t) arena_avail_tree_t;
typedef rb_tree(arena_chunk_map_misc_t) arena_run_tree_t;
@@ -175,17 +160,6 @@ struct arena_chunk_s {
arena_chunk_map_bits_t map_bits[1]; /* Dynamically sized. */
};
struct arena_run_s {
/* Bin this run is associated with. */
arena_bin_t *bin;
/* Index of next region that has never been allocated, or nregs. */
uint32_t nextind;
/* Number of free regions in run. */
unsigned nfree;
};
/*
* Read-only information associated with each element of arena_t's bins array
* is stored separately, partly to reduce memory usage (only one copy, rather
@@ -194,10 +168,7 @@ struct arena_run_s {
* Each run has the following layout:
*
* /--------------------\
* | arena_run_t header |
* | ... |
* bitmap_offset | bitmap |
* | ... |
* | pad? |
* |--------------------|
* | redzone |
* reg0_offset | region 0 |
@@ -238,12 +209,6 @@ struct arena_bin_info_s {
/* Total number of regions in a run for this bin's size class. */
uint32_t nregs;
/*
* Offset of first bitmap_t element in a run header for this bin's size
* class.
*/
uint32_t bitmap_offset;
/*
* Metadata used to manipulate bitmaps for runs associated with this
* bin.
@@ -451,6 +416,9 @@ arena_chunk_map_bits_t *arena_bitselm_get(arena_chunk_t *chunk,
size_t pageind);
arena_chunk_map_misc_t *arena_miscelm_get(arena_chunk_t *chunk,
size_t pageind);
size_t arena_miscelm_to_pageind(arena_chunk_map_misc_t *miscelm);
void *arena_miscelm_to_rpages(arena_chunk_map_misc_t *miscelm);
arena_chunk_map_misc_t *arena_run_to_miscelm(arena_run_t *run);
size_t *arena_mapbitsp_get(arena_chunk_t *chunk, size_t pageind);
size_t arena_mapbitsp_read(size_t *mapbitsp);
size_t arena_mapbits_get(arena_chunk_t *chunk, size_t pageind);
@@ -659,6 +627,40 @@ arena_miscelm_get(arena_chunk_t *chunk, size_t pageind)
(uintptr_t)map_misc_offset) + pageind-map_bias);
}
JEMALLOC_ALWAYS_INLINE size_t
arena_miscelm_to_pageind(arena_chunk_map_misc_t *miscelm)
{
arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(miscelm);
size_t pageind = ((uintptr_t)miscelm - ((uintptr_t)chunk +
map_misc_offset)) / sizeof(arena_chunk_map_misc_t) + map_bias;
assert(pageind >= map_bias);
assert(pageind < chunk_npages);
return (pageind);
}
JEMALLOC_ALWAYS_INLINE void *
arena_miscelm_to_rpages(arena_chunk_map_misc_t *miscelm)
{
arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(miscelm);
size_t pageind = arena_miscelm_to_pageind(miscelm);
return ((void *)((uintptr_t)chunk + (pageind << LG_PAGE)));
}
JEMALLOC_ALWAYS_INLINE arena_chunk_map_misc_t *
arena_run_to_miscelm(arena_run_t *run)
{
arena_chunk_map_misc_t *miscelm = (arena_chunk_map_misc_t
*)((uintptr_t)run - offsetof(arena_chunk_map_misc_t, run));
assert(arena_miscelm_to_pageind(miscelm) >= map_bias);
assert(arena_miscelm_to_pageind(miscelm) < chunk_npages);
return (miscelm);
}
JEMALLOC_ALWAYS_INLINE size_t *
arena_mapbitsp_get(arena_chunk_t *chunk, size_t pageind)
{
@@ -903,10 +905,13 @@ arena_ptr_small_binind_get(const void *ptr, size_t mapbits)
arena_t *arena;
size_t pageind;
size_t actual_mapbits;
size_t rpages_ind;
arena_run_t *run;
arena_bin_t *bin;
size_t actual_binind;
arena_bin_info_t *bin_info;
arena_chunk_map_misc_t *miscelm;
void *rpages;
assert(binind != BININD_INVALID);
assert(binind < NBINS);
@@ -917,13 +922,16 @@ arena_ptr_small_binind_get(const void *ptr, size_t mapbits)
assert(mapbits == actual_mapbits);
assert(arena_mapbits_large_get(chunk, pageind) == 0);
assert(arena_mapbits_allocated_get(chunk, pageind) != 0);
run = (arena_run_t *)((uintptr_t)chunk + (uintptr_t)((pageind -
(actual_mapbits >> LG_PAGE)) << LG_PAGE));
rpages_ind = pageind - arena_mapbits_small_runind_get(chunk,
pageind);
miscelm = arena_miscelm_get(chunk, rpages_ind);
run = &miscelm->run;
bin = run->bin;
actual_binind = bin - arena->bins;
assert(binind == actual_binind);
bin_info = &arena_bin_info[actual_binind];
assert(((uintptr_t)ptr - ((uintptr_t)run +
rpages = arena_miscelm_to_rpages(miscelm);
assert(((uintptr_t)ptr - ((uintptr_t)rpages +
(uintptr_t)bin_info->reg0_offset)) % bin_info->reg_interval
== 0);
}
@@ -946,19 +954,21 @@ arena_run_regind(arena_run_t *run, arena_bin_info_t *bin_info, const void *ptr)
{
unsigned shift, diff, regind;
size_t interval;
arena_chunk_map_misc_t *miscelm = arena_run_to_miscelm(run);
void *rpages = arena_miscelm_to_rpages(miscelm);
/*
* Freeing a pointer lower than region zero can cause assertion
* failure.
*/
assert((uintptr_t)ptr >= (uintptr_t)run +
assert((uintptr_t)ptr >= (uintptr_t)rpages +
(uintptr_t)bin_info->reg0_offset);
/*
* Avoid doing division with a variable divisor if possible. Using
* actual division here can reduce allocator throughput by over 20%!
*/
diff = (unsigned)((uintptr_t)ptr - (uintptr_t)run -
diff = (unsigned)((uintptr_t)ptr - (uintptr_t)rpages -
bin_info->reg0_offset);
/* Rescale (factor powers of 2 out of the numerator and denominator). */

3
include/jemalloc/internal/private_symbols.txt
View File

@@ -42,6 +42,8 @@ arena_mapbitsp_read
arena_mapbitsp_write
arena_maxclass
arena_miscelm_get
arena_miscelm_to_pageind
arena_miscelm_to_rpages
arena_new
arena_palloc
arena_postfork_child
@@ -61,6 +63,7 @@ arena_ralloc_junk_large
arena_ralloc_no_move
arena_redzone_corruption
arena_run_regind
arena_run_to_miscelm
arena_salloc
arena_sdalloc
arena_stats_merge