Commit Graph

2974 Commits

Author SHA1 Message Date
David Goldblatt
03a6047111 Edata cache small: rewrite.
In previous designs, this was intended to be a sort of cache that couldn't fail.
In the current design, we want to use it just as a contention reduction
mechanism.  Rewrite it with those goals in mind.
2020-11-05 12:34:43 -08:00
David Goldblatt
c9757d9e3b HPA: Don't disable shards that were never started. 2020-11-05 12:34:43 -08:00
David Goldblatt
1b3ee75667 Add experimental.thread.activity_callback.
This (experimental, undocumented) functionality can be used by users to track
various statistics of interest at a finer level of granularity than the thread.
2020-11-05 12:33:25 -08:00
David Carlier
27ef02ca9a Android build fix proposal.
These are detected at configure time while they are glibc
specifics. the bionic equivalent is not api compatible
and dlopen is restricted in this platform.
2020-11-02 13:38:44 -08:00
David Carlier
d2d941017b MADV_DO[NOT]DUMP support equivalence on FreeBSD. 2020-11-02 09:15:15 -08:00
David Goldblatt
180b843159 Appveyor: fix 404 errors.
It looks like the mirrors we were using no longer carry this package, but that
it is installed by default and so no longer needs a remote mirror.
2020-10-27 15:28:20 -07:00
DC
ef6d51ed44 DragonFlyBSD build support. 2020-10-27 12:35:19 -07:00
Qi Wang
bf72188f80 Allow opt.tcache_max to accept small size classes.
Previously all the small size classes were cached.  However this has downsides
-- particularly when page size is greater than 4K (e.g. iOS), which will result
in much higher SMALL_MAXCLASS.

This change allows tcache_max to be set to lower values, to better control
resources taken by tcache.
2020-10-24 20:43:44 -07:00
David Goldblatt
ea32060f9c SEC: Implement thread affinity.
For now, just have every thread pick a shard once and stick with it.
2020-10-23 11:14:34 -07:00
David Goldblatt
d16849c91d psset: Do first-fit based on slab age.
This functions more like the serial number strategy of the ecache and
hpa_central_t.  Longer-lived slabs are more likely to continue to live for
longer in the future.
2020-10-23 11:14:34 -07:00
David Goldblatt
634ec6f50a Edata: add an "age" field. 2020-10-23 11:14:34 -07:00
David Goldblatt
6599651aee PA: Use an SEC in fron of the HPA shard. 2020-10-23 11:14:34 -07:00
David Goldblatt
ea51e97bb8 Add SEC module: a small extent cache.
This can be used to take pressure off a more centralized, worse-sharded
allocator without requiring a full break of the arena abstraction.
2020-10-23 11:14:34 -07:00
David Goldblatt
1964b08394 HPA: Add stats for the hpa_shard. 2020-10-23 11:14:34 -07:00
David Goldblatt
534504d4a7 HPA: add size-exclusion functionality.
I.e. only allowing allocations under or over certain sizes.
2020-10-23 11:14:34 -07:00
David Goldblatt
484f04733e HPA: Add central mutex contention stats. 2020-10-23 11:14:34 -07:00
David Goldblatt
bf025d2ec8 HPA: Make slab sizes and maxes configurable.
This allows easy experimentation with them as tuning parameters.
2020-10-23 11:14:34 -07:00
David Goldblatt
1c7da33317 HPA: Tie components into a PAI implementation. 2020-10-23 11:14:34 -07:00
Qi Wang
c8209150f9 Switch from opt.lg_tcache_max to opt.tcache_max
Though for convenience, keep parsing lg_tcache_max.
2020-10-22 20:40:41 -07:00
Yinan Zhang
5ba861715a Add thread name in prof last-N records 2020-10-20 15:58:24 -07:00
David Goldblatt
4ef5b8b4df Add a logo to doc_internal.
This is the logo from the jemalloc development team's snazzy windbreakers.  We
don't actually use it in any documentation yet, but there's no reason we
couldn't.  In the meantime, it's probably best if it exists somewhere more
stable than various email inboxes.
2020-10-19 15:32:51 -07:00
Qi Wang
5e41ff9b74 Add a hard limit on tcache max size class.
For locality reasons, tcache bins are integrated in TSD.  Allowing all size
classes to be cached has little benefit, but takes up much thread local storage.
In addition, it complicates the layout which we try hard to optimize.
2020-10-16 13:49:51 -07:00
Qi Wang
3de19ba401 Eagerly detect double free and sized dealloc bugs for large sizes. 2020-10-15 10:03:16 -07:00
David Goldblatt
be9548f2be Tcaches: Fix a subtle race condition.
Without a lock held continuously between checking tcaches_past and incrementing
it, it's possible for two threads to go down manual creation path
simultaneously.  If the number of tcaches is one less than the maximum, it's
possible for both to create a tcache and increment tcaches_past, with the second
thread returning a value larger than TCACHES_MAX.
2020-10-13 15:06:16 -07:00
Qi Wang
a9aa6f6d0f Fix the alloc_ctx check in free_fastpath.
The sanity check requires a functional TSD, which free_fastpath only guarantees
after the threshold branch.  Move the check function to afterwards.
2020-10-12 19:02:27 -07:00
David Goldblatt
b971f7c4dd Add "default" option to slab sizes.
This comes in handy when overriding earlier settings to test alternate ones.  We
don't really include tests for this, but I claim that's OK here:
- It's fairly straightforward
- It's fairly hard to test well
- This entire code path is undocumented and mostly for our internal
  experimentation in the first place.
- I tested manually.
2020-10-07 12:54:29 -07:00
David Goldblatt
21b70cb540 Add hpa_central module
This will be the centralized component of the coming hugepage allocator; the
source of larger chunks of memory from which smaller ones can be obtained.
2020-10-05 19:55:57 -07:00
David Goldblatt
1ed7ec369f Emap: Add emap_assert_not_mapped.
The counterpart to emap_assert_mapped, it lets callers check that some edata is
not already in the emap.
2020-10-05 19:55:57 -07:00
David Goldblatt
2a6ba121b5 PRNG test: cleanups.
Since we no longer have both atomic and non-atomic variants, there's no reason
to try to test both.
2020-10-05 19:55:57 -07:00
David Goldblatt
9e6aa77ab9 PRNG: Remove atomic functionality.
These had no uses and complicated the API.  As a rule we now expect to only use
thread-local randomization for contention-reduction reasons, so we only pay the
API costs and never get the functionality benefits.
2020-10-05 19:55:57 -07:00
David Goldblatt
0513047170 PRNG: Allow a a range argument of 1.
This is convenient when the range argument itself is generated from some
computation whose value we don't know in advance.
2020-10-05 19:55:57 -07:00
David Goldblatt
bdb60a8053 Appveyor: don't update msys2 keyring.
This is no longer required, and the step now fails.
2020-10-05 19:54:21 -07:00
David Goldblatt
025d8c37c9 Add a script to check for clang-formattedness. 2020-10-02 14:49:56 -07:00
David Goldblatt
f6bbfc1e96 Add a .clang-format file. 2020-10-02 14:49:56 -07:00
David Goldblatt
259c5e3e8f psset: Add stats 2020-09-18 12:39:25 -07:00
David Goldblatt
018b162d67 Add psset: a set of pageslabs.
This introduces a new sort of edata_t; a pageslab, and a set to manage them.
This is part of a series of a commits to implement a hugepage allocator; the
pageset will be per-arena, and track small page allocations requests within a
larger extent allocated from a centralized hugepage allocator.
2020-09-18 12:39:25 -07:00
David Goldblatt
ed99d300b9 Flat bitmap: Add longest-range computation.
This will come in handy in the (upcoming) page-slab set assertions.
2020-09-18 12:39:25 -07:00
David Goldblatt
e034500698 Edata: rename "ranged" bit to "pai".
This better represents its intended purpose; the hugepage allocator design
evolved away from needing contiguity of hugepage virtual address space.
2020-09-18 12:39:25 -07:00
David Goldblatt
7ad2f78663 Avoid a -Wundef warning on LG_SLAB_MAXREGS. 2020-09-17 10:05:40 -07:00
David Goldblatt
40cf71a06d Remove --with-slab-maxregs options from INSTALL.md
The variable slab sizes feature is still experimental; we don't want people to
start using it willy-nilly, or document its existence as a guarantee.
2020-09-17 10:05:40 -07:00
ezeeyahoo
36ebb5abe3 CI support for PPC64LE architecture 2020-09-17 10:03:08 -07:00
Hao Liu
1541ffc765 configure: add --with-lg-slab-maxregs configure option.
Specify the maximum number of regions in a slab, which is
(<lg-page> - <lg-tiny-min>) by default. This increases the limit of slab sizes
specified by "slab_sizes" in malloc_conf. This should never be less than
the default value. The max value of this option is related to LG_BITMAP_MAXBITS
(see more in bitmap.h).

For example, on a 4k page size system, if we:
  1) configure jemalloc with with --with-lg-slab-maxregs=12.
  2) export MALLOC_CONF="slab_sizes:9-16:4"
The slab size of 16 bytes is set to 4 pages. Previously, the default
lg-slab-maxregs is 9 (i.e. 12 - 3). The max slab size of 16 bytes is 2 pages
(i.e. (1<<9) * 16 bytes). By increasing the value from 9 to 12, the max slab
size can be set by MALLOC_CONF is 16 pages (i.e. (1<<12) * 16 bytes).
2020-09-16 13:58:38 -07:00
David Goldblatt
d243b4ec48 Add PROFILING_INTERNALS.md
This documents and explains some of the logic behind the profiling
implementation.
2020-09-10 15:56:59 -07:00
Yinan Zhang
09eda2c9b6 Add unit tests for usize in prof recent records 2020-09-09 13:31:35 -07:00
Yinan Zhang
b549389e4a Correct usize in prof last-N record 2020-09-09 13:31:35 -07:00
Yinan Zhang
202f01d4f8 Fix szind computation in profiling 2020-08-27 15:52:25 -07:00
Yinan Zhang
866231fc61 Do not repeat reentrancy test in profiling 2020-08-25 16:49:32 -07:00
Yinan Zhang
20f2479ed7 Do not create size class tables for non-prof builds 2020-08-24 20:10:02 -07:00
Yinan Zhang
8efcdc3f98 Move unbias data to prof_data 2020-08-24 20:10:02 -07:00
David Goldblatt
5e90fd006e Geom_grow: Don't keep the mutex internal.
We're about to use it in ways that will have external synchronization.
2020-08-19 16:53:21 -07:00