project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
2,956 Commits 2 Branches 0 Tags 13 MiB
c8209150f9
Commit Graph

1582 Commits

Author SHA1 Message Date
David Goldblatt
7503b5b33a Stats, CTL: Expose new tcache settings. 2020-05-16 13:34:23 -07:00
David Goldblatt
ee72bf1cfd Tcache: Add tcache gc delay option. 
This can reduce flushing frequency for small size classes.
 2020-05-16 13:34:23 -07:00
David Goldblatt
d338dd45d7 Tcache: Make incremental gc bytes configurable. 2020-05-16 13:34:23 -07:00
David Goldblatt
ec0b579563 Tcache: Privatize opt_lg_tcache_max default. 2020-05-16 13:34:23 -07:00
David Goldblatt
181093173d Tcache: make slot sizing configurable. 2020-05-16 13:34:23 -07:00
David Goldblatt
b58dea8d1b Cache bin: expose ncached_max publicly. 2020-05-16 13:34:23 -07:00
David Goldblatt
634afc4124 Tcache: Make size computation configurable. 2020-05-16 13:34:23 -07:00
David Goldblatt
eda9c2858f Edata: zero stack edatas before initializing. 
This avoids some UB. No compilers take advantage of it for now, but no sense in
tempting fate.
 2020-05-14 10:30:20 -07:00
David Goldblatt
5dead37a9d Allow narenas:default. 
This can be useful when you know you want to override some lower-priority
configuration setting with its default value, but don't know what that value
would be.
 2020-05-14 10:30:08 -07:00
Yinan Zhang
75dae934a1 Always initialize TE counters in TSD init 2020-05-12 09:16:16 -07:00
Yinan Zhang
b06dfb9ccc Push event handlers to constituent modules 2020-05-12 09:16:16 -07:00
Yinan Zhang
381c97caa4 Treat postponed prof sample event as new event 2020-05-12 09:16:16 -07:00
Yinan Zhang
abd4674931 Extract out per event postponed wait time fetching 2020-05-12 09:16:16 -07:00
Yinan Zhang
f72014d097 Only compute thread event threshold once per trigger 2020-05-12 09:16:16 -07:00
Yinan Zhang
7324c4f85f Break down event init and handler functions 2020-05-12 09:16:16 -07:00
Yinan Zhang
6de77799de Move thread event wait time update to local 2020-05-12 09:16:16 -07:00
Yinan Zhang
733ae918f0 Extract out per event new wait time fetching 2020-05-12 09:16:16 -07:00
Yinan Zhang
1e2524e15a Do not reset sample wait time when re-initing tdata 2020-05-12 09:16:16 -07:00
Yinan Zhang
fc052ff728 Migrate counter to use locked int 2020-05-12 08:23:15 -07:00
Yinan Zhang
f533ab6da6 Add forking handling for stats 2020-05-11 15:35:06 -07:00
Yinan Zhang
508303077b Add forking handling for prof idump counter 2020-05-11 15:35:06 -07:00
Yinan Zhang
4d970f8bfc Add forking handling for counter module 2020-05-11 15:35:06 -07:00
Yinan Zhang
2097e1945b Unify write callback signature 2020-05-11 14:51:24 -07:00
Yinan Zhang
8be5584494 Initialize prof idump counter once rather than once per arena 2020-05-11 12:24:56 -07:00
Yinan Zhang
e10e5059e8 Make prof_idump_accum() non-inline 2020-05-11 12:24:56 -07:00
Yinan Zhang
039bfd4e30 Do not rollback prof idump counter in arena_prof_promote() 2020-05-11 12:24:56 -07:00
Yinan Zhang
0295aa38a2 Deduplicate entries in witness error message 2020-05-11 12:04:02 -07:00
David Goldblatt
f1f8a75496 Let opt.zero propagate to core allocation. 
I.e. set dopts->zero early on if opt.zero is true, rather than leaving it set by
the entry-point function (malloc, calloc, etc.) and then memsetting.  This
avoids situations where we zero once in the large-alloc pathway and then again
via memset.
 2020-05-04 12:36:45 -07:00
David Goldblatt
46471ea327 SC: Name the max lookup constant. 2020-05-04 12:27:07 -07:00
David Goldblatt
cd29ebefd0 Tcache: treat small and large cache bins uniformly 2020-04-14 15:20:19 -07:00
David Goldblatt
a13fbad374 Tcache: split up fast and slow path data. 2020-04-14 15:20:19 -07:00
David Goldblatt
7099c66205 Arena: fill in terms of cache_bins. 2020-04-14 15:20:19 -07:00
David Goldblatt
40e7aed59e TSD: Move in some of the tcache fields. 
We had put these in the tcache for cache optimization reasons.  After the
previous diff, these no longer apply.
 2020-04-14 15:20:19 -07:00
David Goldblatt
3589571bfd SC: use SC_LG_NGROUP instead of its value. 
This magic constant introduces inconsistencies.  We should be able to change its
value solely by adjusting the definition in the header.
 2020-04-13 10:01:30 -07:00
David Goldblatt
79ae7f9211 Rtree: Remove the per-field accessors. 
We instead split things into "edata" and "metadata".
 2020-04-10 13:12:47 -07:00
David Goldblatt
bb6a418523 Emap: Drop szind/slab splitting parameters. 
After the previous diff, these are constants.
 2020-04-10 13:12:47 -07:00
David Goldblatt
50289750b3 Extent: Remove szind/slab knowledge. 2020-04-10 13:12:47 -07:00
David Goldblatt
dc26b30094 Rtree: Clean up compact/non-compact split. 2020-04-10 13:12:47 -07:00
David Goldblatt
93b99dd140 Extent: Stop passing an edata_cache everywhere. 
We already pass the pa_shard_t around everywhere; we can just use that.
 2020-04-10 13:12:47 -07:00
David Goldblatt
a4759a1911 Ehooks: avoid touching arena_emap_global in tests. 
That breaks our ability to test custom emaps in isolation.
 2020-04-10 13:12:47 -07:00
David Goldblatt
11c47cb133 Extent: Take "bool zero" over "bool *zero". 2020-04-10 13:12:47 -07:00
David Goldblatt
1a1124462e PA: Take zero as a bool rather than as a bool *. 
Now that we've moved junking to a higher level of the allocation stack, we don't
care about this performance optimization (which only occurred in debug modes).
 2020-04-10 13:12:47 -07:00
David Goldblatt
294b276fc7 PA: Parameterize emap. Move emap_global to arena. 
This lets us test the PA module without interfering with the global emap used by
the real allocator (the one not under test).
 2020-04-10 13:12:47 -07:00
David Goldblatt
f730577277 Eset: Parameterize last globals accesses. 
I.e. opt_retain and maps_coalesce.
 2020-04-10 13:12:47 -07:00
David Goldblatt
7bb6e2dc0d Eset: take opt_lg_max_active_fit as a parameter. 
This breaks its dependence on the global.
 2020-04-10 13:12:47 -07:00
David Goldblatt
883ab327cc Emap: Move out last edata state touching. 2020-04-10 13:12:47 -07:00
David Goldblatt
0c96a2f03b Emap: Move out remaining edata modifications. 2020-04-10 13:12:47 -07:00
David Goldblatt
dfef0df71a Emap: Move edata modification out of emap_remap. 2020-04-10 13:12:47 -07:00
David Goldblatt
12eb888e54 Edata: Add a ranged bit. 
We steal the dumpable bit, which we ended up not needing.
 2020-04-10 13:12:47 -07:00
David Goldblatt
bd4fdf295e Rtree: Pull leaf contents into their own struct. 2020-04-10 13:12:47 -07:00
David Goldblatt
faec7219b2 PA: Move in decay initialization. 2020-04-10 13:12:47 -07:00
David Goldblatt
45671e4a27 PA: Move in retain growth limit setting. 2020-04-10 13:12:47 -07:00
David Goldblatt
daefde88fe PA: Move in mutex stats reading. 2020-04-10 13:12:47 -07:00
David Goldblatt
07675840a5 PA: Move in some more internals accesses. 2020-04-10 13:12:47 -07:00
David Goldblatt
238f3c7430 PA: Move in full stats merging. 2020-04-10 13:12:47 -07:00
David Goldblatt
81c6027592 Arena stats: Give it its own "mapped". 
This distinguishes it from the PA mapped stat, which is now named "pa_mapped" to
avoid confusion. The (derived) arena stat includes base memory, and the PA stat
is no longer partially derived.
 2020-04-10 13:12:47 -07:00
David Goldblatt
506d907e40 PA: Move in basic stats merging. 2020-04-10 13:12:47 -07:00
David Goldblatt
f29f6090f5 PA: Add pa_extra.c and put PA forking there. 2020-04-10 13:12:47 -07:00
David Goldblatt
8164fad404 Stats: Fix edata_cache size merging. 
Previously, we assigned to the output rather than incrementing it.
 2020-04-10 13:12:47 -07:00
David Goldblatt
565045ef71 Arena: Make more derived stats non-atomic/locked. 2020-04-10 13:12:47 -07:00
David Goldblatt
d0c43217b5 Arena stats: Move retained to PA, use plain ints. 
Retained is a property of the allocated pages.  The derived fields no longer
require any locking; they're computed on demand.
 2020-04-10 13:12:47 -07:00
David Goldblatt
e2cf3fb1a3 PA: Move in all modifications of mapped. 2020-04-10 13:12:47 -07:00
David Goldblatt
436789ad96 PA: Make mapped stat atomic. 
We always have atomic_zu_t, and mapped/unmapped transitions are always expensive
enough that trying to piggyback on a lock is a waste of time.
 2020-04-10 13:12:47 -07:00
David Goldblatt
3c28aa6f17 PA: Move edata_avail stat in, make it non-atomic. 2020-04-10 13:12:47 -07:00
David Goldblatt
f6bfa3dcca Move extent stats to the PA module. 
While we're at it, make them non-atomic -- they are purely derived statistics
(and in fact aren't even in the arena_t or pa_shard_t).
 2020-04-10 13:12:47 -07:00
David Goldblatt
527dd4cdb8 PA: Move in nactive counter. 2020-04-10 13:12:47 -07:00
David Goldblatt
c075fd0bcb PA: Minor cleanups and comment fixes. 2020-04-10 13:12:47 -07:00
David Goldblatt
46a9d7fc0b PA: Move in rest of purging. 2020-04-10 13:12:47 -07:00
David Goldblatt
2d6eec7b5c PA: Move in decay-all pathway. 2020-04-10 13:12:47 -07:00
David Goldblatt
65698b7f2e PA: Remove public visibility of some internals. 2020-04-10 13:12:47 -07:00
David Goldblatt
f012c43be0 PA: Move in decay_to_limit 2020-04-10 13:12:47 -07:00
David Goldblatt
103f5feda5 Move bg thread activity check out of purging core. 2020-04-10 13:12:47 -07:00
David Goldblatt
3034f4a508 PA: Move in decay_stashed. 2020-04-10 13:12:47 -07:00
David Goldblatt
aef28b2f8f PA: Move in stash_decayed. 2020-04-10 13:12:47 -07:00
David Goldblatt
655a096343 Move bg inactivity check out of purge inner loop. 
I.e. do it once per call to arena_decay_stashed instead of once per muzzy purge.
 2020-04-10 13:12:47 -07:00
David Goldblatt
71fc0dc968 PA: Move in remaining page allocation functions. 2020-04-10 13:12:47 -07:00
David Goldblatt
74958567a4 PA: have expand take sizes instead of new usize. 
This avoids involving usize, which makes some of the stats modifications more
intuitively correct.
 2020-04-10 13:12:47 -07:00
David Goldblatt
5bcc2c2ab9 PA: Have expand take szind and slab. 
This isn't really necessary, but having a uniform API will help us later.
 2020-04-10 13:12:47 -07:00
David Goldblatt
0880c2ab97 PA: Have large expands use it. 2020-04-10 13:12:47 -07:00
David Goldblatt
7be3dea82c PA: Have slab allocations use it. 2020-04-10 13:12:47 -07:00
David Goldblatt
9f93625c14 PA: Move in arena large allocation functionality. 2020-04-10 13:12:47 -07:00
David Goldblatt
7624043a41 PA: Add ehook-getting support. 2020-04-10 13:12:47 -07:00
David Goldblatt
eba35e2e48 Remove extent knowledge of arena. 2020-04-10 13:12:47 -07:00
David Goldblatt
e77f47a85a Move arena decay getters to PA. 2020-04-10 13:12:47 -07:00
David Goldblatt
f77cec311e Decay: Take current time as an argument. 
This better facilitates testing.
 2020-04-10 13:12:47 -07:00
David Goldblatt
d1d7e1076b Decay: move in some background_thread accesses. 2020-04-10 13:12:47 -07:00
David Goldblatt
8f2193dc8d Decay: Move in arena decay functions. 2020-04-10 13:12:47 -07:00
David Goldblatt
4d090d23f1 Decay: Introduce a stub .c file. 2020-04-10 13:12:47 -07:00
David Goldblatt
7b62885476 Introduce decay module and put decay objects in PA 2020-04-10 13:12:47 -07:00
David Goldblatt
3192d6b77d Extents: Have extent_dalloc_gap take ehooks. 
We're almost to the point where the extent code doesn't know about arenas at
all.  In that world, we shouldn't pull them out of the arena.
 2020-04-10 13:12:47 -07:00
David Goldblatt
22a0a7b93a Move arena_decay_extent to extent module. 2020-04-10 13:12:47 -07:00
David Goldblatt
70d12ffa05 PA: Move mapped into pa stats. 2020-04-10 13:12:47 -07:00
David Goldblatt
ce8c0d6c09 PA: Move in arena extent_sn counter. 
Just another step towards making PA self-contained.
 2020-04-10 13:12:47 -07:00
David Goldblatt
1ada4aef84 PA: Get rid of arena_ind_get calls. 
This is another step on the path towards breaking the extent reliance on the
arena module.
 2020-04-10 13:12:47 -07:00
David Goldblatt
1ad368c8b7 PA: Move in decay stats. 2020-04-10 13:12:47 -07:00
David Goldblatt
356aaa7dc6 Introduce lockedint module. 
This pulls out the various abstractions where some stats counter is sometimes an
atomic, sometimes a plain variable, sometimes always protected by a lock,
sometimes protected by reads but not writes, etc.  With this change, these cases
are treated consistently, and access patterns tagged.

In the process, we fix a few missed-update bugs (where one caller assumes
"protected-by-a-lock" semantics and another does not).
 2020-04-10 13:12:47 -07:00
David Goldblatt
acd0bf6a26 PA: move in ecache_grow. 2020-04-10 13:12:47 -07:00
David Goldblatt
32cb7c2f0b PA: Add a stats type. 2020-04-10 13:12:47 -07:00
David Goldblatt
688fb3eb89 PA: Move in the arena edata_cache. 2020-04-10 13:12:47 -07:00
David Goldblatt
8433ad84ea PA: move in shard initialization. 2020-04-10 13:12:47 -07:00
David Goldblatt
a24faed569 PA: Move in the ecache_t objects. 2020-04-10 13:12:47 -07:00
David Goldblatt
585f925055 Move cache index randomization out of extent. 
This is logically at a higher level of the stack; extent should just allocate
things at the page-level; it shouldn't care exactly why the callers wants a
given number of pages.
 2020-04-10 13:12:47 -07:00
David Goldblatt
12be9f5727 Add a stub PA module -- a page allocator. 2020-04-10 13:12:47 -07:00
Yinan Zhang
c4e9ea8cc6 Get rid of locks in prof recent test 2020-04-07 17:22:24 -07:00
Yinan Zhang
2deabac079 Get rid of custom iterator for last-N records 2020-04-07 17:22:24 -07:00
Yinan Zhang
a5ddfa7d91 Use ql for prof last-N list 2020-04-07 17:22:24 -07:00
Yinan Zhang
f9aad7a49b Add piping API to buffered writer 2020-04-01 09:41:20 -07:00
Yinan Zhang
09cd79495f Encapsulate buffer allocation failure in buffered writer 2020-04-01 09:41:20 -07:00
Yinan Zhang
a166c20818 Make prof_tctx_t pointer a true prof atomic fence 2020-03-31 17:43:42 -07:00
David T. Goldblatt
d936b46d3a Add malloc_conf_2_conf_harder 
This comes in handy when you're just a user of a canary system who wants to
change settings set by the configuration system itself.
 2020-03-31 06:25:08 -07:00
Yinan Zhang
2256ef8961 Add option to fetch system thread name on each prof sample 2020-03-24 21:39:57 -07:00
Yinan Zhang
b30a5c2f90 Reorganize cpp APIs and suppress unused function warnings 2020-03-13 12:16:09 -07:00
David Goldblatt
2e5899c129 Stats: Fix tcache_bytes reporting. 
Previously, large allocations in tcaches would have their sizes reduced during
stats estimation.  Added a test, which fails before this change but passes now.

This fixes a bug introduced in 5934846612, which
was itself fixing a bug introduced in 9c0549007d.
 2020-03-13 07:53:34 -07:00
Yinan Zhang
a5780598b3 Remove thread_event_rollback() 2020-03-12 13:55:00 -07:00
Yinan Zhang
ba783b3a0f Remove prof -> thread_event dependency 2020-03-12 13:55:00 -07:00
Yinan Zhang
441d88d1c7 Rewrite profiling thread event 2020-03-12 13:55:00 -07:00
David Goldblatt
0dcd576600 Edata cache: atomic fetch-add -> load-store. 
The modifications to count are protected by a mutex; there's no need to use the
more costly version.
 2020-03-12 11:58:09 -07:00
David Goldblatt
99b1291d17 Edata cache: add edata_cache_small_t. 
This can be used to amortize the synchronization costs of edata_cache accesses.
 2020-03-12 11:58:09 -07:00
David Goldblatt
d701a085c2 Fast path: allow low-water mark changes. 
This lets us put more allocations on an "almost as fast" path after a flush.
This results in around a 4% reduction in malloc cycles in prod workloads
(corresponding to about a 0.1% reduction in overall cycles).
 2020-03-12 11:54:19 -07:00
David Goldblatt
397da03865 Cache bin: rewrite to track more state. 
With this, we track all of the empty, full, and low water states together.  This
simplifies a lot of the tracking logic, since we now don't need the
cache_bin_info_t for state queries (except for some debugging).
 2020-03-12 11:54:19 -07:00
David Goldblatt
fef0b1ffe4 Cache bin: Remove last internals accesses. 2020-03-12 11:54:19 -07:00
David Goldblatt
0a2fcfac01 Tcache: Hold cache bin allocation explicitly. 2020-03-12 11:54:19 -07:00
David Goldblatt
d498a4bb08 Cache bin: Add an emptiness assertion. 2020-03-12 11:54:19 -07:00
David Goldblatt
6a7aa46ef7 Cache bin: Add a debug method for init checking. 2020-03-12 11:54:19 -07:00
David Goldblatt
7f5ebd211c Cache bin: set low-water internally. 2020-03-12 11:54:19 -07:00
David Goldblatt
60113dfe3b Cache bin: Move in initialization code. 2020-03-12 11:54:19 -07:00
David Goldblatt
44529da852 Cache-bin: Make flush modifications internal 
I.e. the tcache code just calls a cache-bin function to finish flush (and move
pointers around, etc.).  It doesn't directly access the cache-bin's owned memory
any more.
 2020-03-12 11:54:19 -07:00
David Goldblatt
ff6acc6ed5 Cache bin: simplify names and argument ordering. 
We always start with the cache bin, then its info (if necessary).
 2020-03-12 11:54:19 -07:00
David Goldblatt
e1dcc557d6 Cache bin: Only take the relevant cache_bin_info_t 
Previously, we took an array of cache_bin_info_ts and an index, and dereferenced
ourselves.  But infos for other cache_bins aren't relevant to any particular
cache bin, so that should be the caller's job.
 2020-03-12 11:54:19 -07:00
David Goldblatt
1b00d808d7 cache_bin: Don't let arena see empty position. 2020-03-12 11:54:19 -07:00
David Goldblatt
d303f30796 cache_bin nflush -> n. 
We're going to use it on the fill pathway as well.
 2020-03-12 11:54:19 -07:00
David Goldblatt
74d36d78ef Cache bin: Make ncached_max a query on the info_t. 2020-03-12 11:54:19 -07:00
David Goldblatt
b66c0973cc cache_bin: Don't allow direct internals access. 2020-03-12 11:54:19 -07:00
David Goldblatt
909c501b07 Cache_bin: Shouldn't know about tcache. 
Instead, have it take the cache_bin_info_ts to use by pointer.  While we're
here, add a src file for the cache bin.
 2020-03-12 11:54:19 -07:00
David Goldblatt
79f1ee2fc0 Move junking out of arena/tcache code. 
This is debug only and we keep it off the fast path.  Moving it here simplifies
the internal logic.

This never tries to junk on regions that were shrunk via xallocx.  I think this
is fine for two reasons:
- The shrunk-with-xallocx case is rare.
- We don't always do that anyway before this diff (it depends on the opt
  settings and extent hooks in effect).
 2020-03-12 11:54:19 -07:00
David Goldblatt
22657a5e65 Extents: Silence the "potentially unused" warning. 2020-03-12 11:54:19 -07:00
Yinan Zhang
305b1f6d96 Correction on geometric sampling 2020-03-04 13:55:21 -08:00
David T. Goldblatt
6c3491ad31 Tcache: Unify bin flush logic. 
The small and large pathways share most of their logic, even if some of the
individual operations are different.  We pull out the common logic into a
force-inlined function, and then specialize twice, once for each value of
"small".
 2020-02-25 10:21:03 -08:00
David T. Goldblatt
162c2bcf31 Background thread: take base as a parameter. 2020-02-18 11:22:09 -08:00
David T. Goldblatt
29436fa056 Break prof and tcache knowledge of b0. 2020-02-18 11:22:09 -08:00
David T. Goldblatt
a0c1f4ac57 Rtree: take the base allocator as a parameter. 
This facilitates better testing by avoiding mixing of the "real" base with the
base used by the rtree under test.
 2020-02-18 11:22:09 -08:00
David T. Goldblatt
7013716aaa Emap: Take (and propagate) a zeroed parameter. 
Rtree needs this, and we should really treat them similarly.
 2020-02-18 11:22:09 -08:00
David T. Goldblatt
182192f83c Base: Pull into a single header. 2020-02-18 11:22:09 -08:00
David Goldblatt
7e6c8a7286 Emap: Standardize naming. 
Namespace everything under emap_, always specify what it is we're looking up
(emap_lookup -> emap_edata_lookup), and use "ctx" over "info".
 2020-02-17 10:50:51 -08:00
David Goldblatt
ac50c1e44b Emap: Remove direct access to emap internals. 
In the process, we do a few local cleanups and optimizations.  In particular,
the size safety check on tcache flush no longer does a redundant load.
 2020-02-17 10:50:51 -08:00
David Goldblatt
06e42090f7 Make jemalloc.c use the emap interface. 
While we're here, we'll also clean up some style nits.
 2020-02-17 10:50:51 -08:00
David Goldblatt
f7d9c6c42d Emap: Move in alloc_ctx lookup functionality. 2020-02-17 10:50:51 -08:00
David Goldblatt
65a54d7714 Emap: Move in szind and slab modifications. 2020-02-17 10:50:51 -08:00
David Goldblatt
9b5d105fc3 Emap: Move in iealloc. 
This is logically scoped to the emap.
 2020-02-17 10:50:51 -08:00
David Goldblatt
1d449bd9a6 Emap: Internal rtree context setting. 
The only time sharing an rtree context saves across extent operations isn't a
no-op is when tsd is unavailable.  But this happens only in situations like
thread death or initialization, and we don't care about shaving off every
possible cycle in such scenarios.
 2020-02-17 10:50:51 -08:00