Commit Graph

593 Commits

Author SHA1 Message Date
David Goldblatt
1d4a7666d5 HPA: Do deferred operations on background threads. 2021-07-12 17:59:18 -07:00
David Goldblatt
47d8a7e6b0 psset: Purge empty slabs first.
These are particularly good candidates for purging (listed in the diff).
2021-07-12 17:59:18 -07:00
David Goldblatt
41fd56605e HPA: Purge across retained extents.
This lets us cut down on the number of expensive system calls we perform.
2021-07-12 17:59:18 -07:00
David Goldblatt
de033f56c0 mpsc_queue: Add module.
This is a simple multi-producer, single-consumer queue.  The intended use case
is in the HPA, as we begin supporting hpdatas that move between hpa_shards.  We
take just a single CAS as the cost to send a message (or a batch of messages) in
the low-contention case, and lock-freedom lets us avoid some lock-ordering
issues.
2021-06-24 14:55:49 -07:00
David Goldblatt
0689448b1e Travis: Unbreak the builds.
In the hopes of future-proofing as much as possible, jump to the latest
distribution Travis supports.
2021-06-24 07:40:28 -07:00
David Goldblatt
36c6bfb963 SEC: Allow arbitrarily many shards, cached sizes. 2021-05-22 08:17:41 -07:00
Qi Wang
08089589f7 Fix an interaction between the oversize_threshold test and bgthds.
Also added the shared utility to check if background_thread is enabled.
2021-05-13 16:19:14 -07:00
David Goldblatt
5417938215 Red-black tree: add summarize/filter.
This allows tracking extra information in the nodes of an red-black tree to
filter searches in the tree to just those that match some property.
2021-05-12 11:14:23 -07:00
David Goldblatt
b2c08ef2e6 RB unit tests: don't test reentrantly.
The RB code doesn't do any allocation, and takes a little bit of time to run.
There's no sense in doing everything three times.
2021-05-12 11:14:23 -07:00
David Goldblatt
aea91b8c33 Clean up some minor data structure inconsistencies
Namely, unify the include guard styling with the majority of the project, and do
flat_bitmap -> fb, to match its naming convention.
2021-05-12 11:14:23 -07:00
David Goldblatt
12cd13cd41 Fix thread.name/prof_sys_thread_name interaction
When prof_sys_thread_name is true, we don't allow setting the thread name.
Teach the unit test this.
2021-03-31 14:45:12 -07:00
David Goldblatt
304cdbb132 Fix a prof_recent/prof_sys_thread_name interaction
When both of these are enabled, the output format changes slightly.  Teach the
unit test about the interaction.
2021-03-31 14:45:12 -07:00
Qi Wang
7c964b0352 Add rtree_write_range(): writing the same content to multiple leaf elements.
Apply to emap_(de)register_interior which became noticeable in perf profiles.
2021-03-29 17:19:53 -07:00
Qi Wang
4d8c22f9a5 Store edata->state in rtree leaf and make edata_t 128B aligned.
Verified that this doesn't result in any real increase of edata_t bytes
allocated.
2021-03-29 17:19:53 -07:00
Qi Wang
70d1541c5b Track extent is_head state in rtree leaf. 2021-03-29 17:19:53 -07:00
David Goldblatt
73ca4b8ef8 HPA: Use dirtiest-first purging.
This seems to be practically beneficial, despite some pathological corner cases.
2021-02-19 15:10:54 -08:00
David Goldblatt
d21d5b46b6 Edata: Move sn into its own field.
This lets the bins use a fragmentation avoidance policy that matches the HPA's
(without affecting the PAC).
2021-02-19 15:10:54 -08:00
David Goldblatt
fb327368db SEC: Expand option configurability.
This change pulls the SEC options into a struct, which simplifies their handling
across various modules (e.g. PA needs to forward on SEC options from the
malloc_conf string, but it doesn't really need to know their names).  While
we're here, make some of the fixed constants configurable, and unify naming from
the configuration options to the internals.
2021-02-19 15:10:54 -08:00
David Goldblatt
ce9386370a HPA: Implement batch allocation. 2021-02-19 15:10:54 -08:00
David Goldblatt
cdae6706a6 SEC: Use batch fills.
Currently, this doesn't help much, since no PAI implementation supports
flushing.  This will change in subsequent commits.
2021-02-19 15:10:54 -08:00
David Goldblatt
480f3b11cd Add a batch allocation interface to the PAI.
For now, no real allocator actually implements this interface; this will change
in subsequent diffs.
2021-02-19 15:10:54 -08:00
David Goldblatt
bf448d7a5a SEC: Reduce lock hold times.
Only flush a subset of extents during flushing, and drop the lock while doing
so.
2021-02-19 15:10:54 -08:00
David Goldblatt
f47b4c2cd8 PAI/SEC: Add a dalloc_batch function.
This lets the SEC flush all of its items in a single call, rather than flushing
everything at once.
2021-02-19 15:10:54 -08:00
Qi Wang
a11be50332 Implement opt.cache_oblivious.
Keep config.cache_oblivious for now to remain backward-compatible.
2021-02-11 11:32:01 -08:00
David Goldblatt
b3df80bc79 Pull HPA options into a containing struct.
Currently that just means max_alloc, but we're about to add more.  While we're
touching these lines anyways, tweak things to be more in line with testing.
2021-02-04 20:58:31 -08:00
David Goldblatt
bdb7307ff2 fxp: Add FXP_INIT_PERCENT
This lets us specify fxp values easily in source.
2021-02-04 20:58:31 -08:00
David Goldblatt
caef4c2868 FXP: add fxp_mul_frac.
This can multiply size_ts by a fraction without the risk of overflow.
2021-02-04 20:58:31 -08:00
David Goldblatt
dc886e5608 hpdata: Return the number of pages to be purged.
We'll use this in the next commit.
2021-02-04 20:58:31 -08:00
David Goldblatt
9fd9c876bb psset: keep aggregate stats.
This will let us quickly query these stats to make purging decisions quickly.
2021-02-04 20:58:31 -08:00
David Goldblatt
da63f23e68 HPA: Track pending purges/hugifies in the psset.
This finishes the refactoring of the HPA/psset interactions the past few commits
have been building towards.

Rather than the HPA removing and then reinserting hpdatas, it simply begins
updates and ends them.  These updates can set flags on the hpdata that prevent
it from being returned for certain types of requests.  For example, it can call
hpdata_alloc_allowed_set(hpdata, false) during an update, at which point the
given hpdata will no longer be returned for psset_pick_alloc requests.

This has various of benefits:
- It maintains stats correctness during purges and hugifies.
- It allows simpler and more explicit concurrency control for the various
  special cases (e.g. allocations are disallowed during purge, but not during
  hugify).
- It lets allocations and deallocations avoid disturbing the purging and
  hugification orderings.  If an hpdata "loses its place" in one of the queues
  just do to an alloc / dalloc, it can result in pathological edge cases where
  very hot, very full hugepages never get hugified  (and cold extents on the
  same hugepage as hot ones never get purged).

The key benefit though is that tracking hpdatas to be purged / hugified in a
principled way will let us do delayed purging and hugification.  Eventually this
will let us move these operations to background threads, but in the short term
the benefit is that it will let us have global purging policies (e.g. purge when
the entire arena has too many dirty pages, rather than any particular hugepage).
2021-02-04 20:58:31 -08:00
David Goldblatt
bf64557ed6 Move empty slab tracking to the psset.
We're moving towards a world in which purging decisions are less rigidly
enforced at a single-hugepage level.  In that world, it makes sense to keep
around some hpdatas which are not completely purged, in which case we'll need to
track them.
2021-02-04 20:58:31 -08:00
David Goldblatt
99fc0717e6 psset: Reconceptualize insertion/removal.
Really, this isn't a functional change, just a naming change.  We start thinking
of pageslabs as being always in the psset.  What we used to think of as removal
is now thought of as being in the psset, but in the process of being updated
(and therefore, unavalable for serving new allocations).

This is in preparation of subsequent changes to support deferred purging;
allocations will still be in the psset for the purposes of choosing when to
purge, but not for purposes of allocation/deallocation.
2021-02-04 20:58:31 -08:00
David Goldblatt
68a1666e91 hpdata: Rename "dirty" to "touched".
This matches the usage in the rest of the codebase.
2021-02-04 20:58:31 -08:00
David Goldblatt
be0d7a53f3 HPA: Don't track inactive pages.
This is really only useful for human consumption.  Correspondingly, emit it only
in the human-readable stats, and let everybody else compute from the hugepage
size and nactive.
2021-02-04 20:58:31 -08:00
David Goldblatt
55e0f60ca1 psset stats: Simplify handling.
We can treat the huge and nonhuge cases uniformly using huge state as an array
index.
2021-02-04 20:58:31 -08:00
David Goldblatt
30b9e8162b HPA: Generalize purging.
Previously, we would purge a hugepage only when it's completely empty.  With
this change, we can purge even when only partially empty.  Although the
heuristic here is still fairly primitive, this infrastructure can scale to
become more advanced.
2021-02-04 20:58:31 -08:00
David Goldblatt
70692cfb13 hpdata: Add state changing helpers.
We're about to allow hugepage subextent purging; get as much of our metadata
handling ready as possible.
2021-02-04 20:58:31 -08:00
David Goldblatt
9b75808be1 flat bitmap: Add a bitwise and/or/not.
We're about to need them.
2021-02-04 20:58:31 -08:00
David Goldblatt
c259323ab3 Use ticker_geom_t for arena tcache decay. 2021-02-04 14:10:43 -08:00
David Goldblatt
8edfc5b170 Add ticker_geom_t.
This lets a single ticker object drive events across a large number of different
tick streams while sharing state.
2021-02-04 14:10:43 -08:00
David Goldblatt
2fcbd18115 Cache bin: Don't reverse flush order.
The items we pick to flush matter a lot, but the order in which they get flushed
doesn't; just use forward scans.  This simplifies the accessing code, both in
terms of the C and the generated assembly (i.e. this speeds up the flush
pathways).
2021-02-04 14:10:43 -08:00
David Goldblatt
229994a204 Tcache flush: keep common path state in registers.
By carefully force-inlining the division constants and the operation sum count,
we can eliminate redundant operations in the arena-level dalloc function.  Do
so.
2021-02-04 14:10:43 -08:00
Azat Khuzhin
a943172b73 Add runtime detection for MADV_DONTNEED zeroes pages (mostly for qemu)
qemu does not support this, yet [1], and you can get very tricky assert
if you will run program with jemalloc in use under qemu:

    <jemalloc>: ../contrib/jemalloc/src/extent.c:1195: Failed assertion: "p[i] == 0"

  [1]: https://patchwork.kernel.org/patch/10576637/

Here is a simple example that shows the problem [2]:

    // Gist to check possible issues with MADV_DONTNEED
    // For example it does not supported by qemu user
    // There is a patch for this [1], but it hasn't been applied.
    //   [1]: https://lists.gnu.org/archive/html/qemu-devel/2018-08/msg05422.html

    #include <sys/mman.h>
    #include <stdio.h>
    #include <stddef.h>
    #include <assert.h>
    #include <string.h>

    int main(int argc, char **argv)
    {
        void *addr = mmap(NULL, 1<<16, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
        if (addr == MAP_FAILED) {
            perror("mmap");
            return 1;
        }
        memset(addr, 'A', 1<<16);

        if (!madvise(addr, 1<<16, MADV_DONTNEED)) {
            puts("MADV_DONTNEED does not return error. Check memory.");
            for (int i = 0; i < 1<<16; ++i) {
                assert(((unsigned char *)addr)[i] == 0);
            }
        } else {
            perror("madvise");
        }

        if (munmap(addr, 1<<16)) {
            perror("munmap");
            return 1;
        }

        return 0;
    }

  ### unpatched qemu

      $ qemu-x86_64-static /tmp/test-MADV_DONTNEED
      MADV_DONTNEED does not return error. Check memory.
      test-MADV_DONTNEED: /tmp/test-MADV_DONTNEED.c:19: main: Assertion `((unsigned char *)addr)[i] == 0' failed.
      qemu: uncaught target signal 6 (Aborted) - core dumped
      Aborted (core dumped)

  ### patched qemu (by returning ENOSYS error)

      $ qemu-x86_64 /tmp/test-MADV_DONTNEED
      madvise: Success

  ### patch for qemu to return ENOSYS

      diff --git a/linux-user/syscall.c b/linux-user/syscall.c
      index 897d20c076..5540792e0e 100644
      --- a/linux-user/syscall.c
      +++ b/linux-user/syscall.c
      @@ -11775,7 +11775,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
                  turns private file-backed mappings into anonymous mappings.
                  This will break MADV_DONTNEED.
                  This is a hint, so ignoring and returning success is ok.  */
      -        return 0;
      +        return ENOSYS;
       #endif
       #ifdef TARGET_NR_fcntl64
           case TARGET_NR_fcntl64:

  [2]: https://gist.github.com/azat/12ba2c825b710653ece34dba7f926ece

v2:
- review fixes
- add opt_dont_trust_madvise
v3:
- review fixes
- rename opt_dont_trust_madvise to opt_trust_madvise
2021-01-20 20:08:30 -08:00
David Goldblatt
a011c4c22d cache_bin: Separate out local and remote accesses.
This fixes an incorrect debug-mode assert:
- T1 starts an arena stats update and reads stack_head from another thread's
  cache bin, when that cache bin has 1 item in it.
- T2 allocates from that cache bin.  The cache_bin's stack_head now points to a
  NULL pointer, since the cache bin is empty.
- T1 Re-reads the cache_bin's stack_head to perform an assertion check (since it
  previously saw that the bin was empty, whatever stack_head points to should be
  non-NULL).
2021-01-08 14:18:08 -08:00
Yinan Zhang
4352cbc21c Add alignment tests for prof stats 2021-01-07 20:39:49 -08:00
Yinan Zhang
54f3351f1f Add mallctl for prof stats fetching 2021-01-07 20:39:49 -08:00
Yinan Zhang
40fa4d29d3 Track per size class internal fragmentation 2021-01-07 20:39:49 -08:00
David Goldblatt
83cad746ae prof_log: cassert(config_prof) in public functions
This lets the compiler infer that the code is dead in builds where profiling is
enabled, saving on space there.
2021-01-04 14:55:49 -08:00
Yinan Zhang
4557c0a67d Enable ctl on partial mib and partial name 2020-12-18 10:39:58 -08:00
Yinan Zhang
006dd0414e Add partial name-to-mib functionality 2020-12-18 10:39:58 -08:00
Yinan Zhang
f2e1a5be77 Do not fail on partial ctl path for ctl_nametomib()
We do not fail on partial ctl path when the given `mib` array is
shorter than the given name, and we should keep the behavior the
same in the reverse case, which I feel is also the more natural way.
2020-12-18 10:39:58 -08:00
David Goldblatt
1e3b8636ff HPA: Remove unused malloc_conf options. 2020-12-08 12:10:48 -08:00
David Goldblatt
f51948d9e1 psset unit test: fix a bug.
The next commit adds assertions that reveal a bug in the test code
(double-free).  Fix it.
2020-12-07 06:21:08 -08:00
David Goldblatt
54c94c1679 flat bitmap: add scount / ucount functions.
These can compute the number or set or unset bits in a subrange of the bitmap.
2020-12-07 06:21:08 -08:00
David Goldblatt
734e72ce8f bit_util: Guarantee popcount's presence.
Implement popcount generically, so that we can rely on it being present.
2020-12-07 06:21:08 -08:00
David Goldblatt
d9f7e6c668 hpdata: Add a test.
We're about to make the functionality here more complicated; testing hpdata
directly (rather than relying on user's tests) will make debugging easier.
2020-12-07 06:21:08 -08:00
David Goldblatt
f7cf23aa4d psset: Relegate alloc/dalloc to test code.
This is no longer part of the "core" functionality; we only need the stub
implementations as an end-to-end test of hpdata + psset interactions when
metadata is being modified.  Treat them accordingly.
2020-12-07 06:21:08 -08:00
David Goldblatt
ca30b5db2b Introduce hpdata_t.
Using an edata_t both for hugepages and the allocations within those hugepages
was convenient at first, but has outlived its usefulness.  Representing
hugepages explicitly, with their own data structure, will make future
development easier.
2020-12-07 06:21:08 -08:00
David Goldblatt
4a15008cfb HPA unit test: skip if unsupported.
Previously, we replicated the logic in hpa_supported in the test as well.
2020-12-07 06:21:08 -08:00
David Goldblatt
43af63fff4 HPA: Manage whole hugepages at a time.
This redesigns the HPA implementation to allow us to manage hugepages all at
once, locally, without relying on a global fallback.
2020-12-07 06:21:08 -08:00
David Goldblatt
c1b2a77933 psset: Move in stats.
A later change will benefit from having these functions pulled into a
psset-module set of functions.
2020-12-07 06:21:08 -08:00
David Goldblatt
d0a991d47b psset: Add insert/remove functions.
These will allow us to (for instance) move pageslabs from a psset dedicated to
not-yet-hugeified pages to one dedicated to hugeified ones.
2020-12-07 06:21:08 -08:00
David Goldblatt
ecd39418ac Add fxp: A fixed-point math library.
This will be used in the next commit to allow non-integer values for
narenas_ratio.
2020-12-04 23:48:19 -08:00
Yinan Zhang
d96e4525ad Route batch allocation of small batch size to tcache 2020-11-16 20:58:01 -08:00
Yinan Zhang
ac480136d7 Split out locality checking in batch allocation tests 2020-11-16 20:58:01 -08:00
Yinan Zhang
be5e49f4fa Add a batch mode for cache_bin_alloc() 2020-11-16 20:58:01 -08:00
Yinan Zhang
4a65f34930 Fix a cache bin test 2020-11-16 20:58:01 -08:00
Yinan Zhang
9545c2cd36 Add sample interval to prof last-N dump 2020-11-13 15:33:27 -08:00
David Goldblatt
cf2549a149 Add a per-arena oversize_threshold.
This can let manual arenas trade off memory and CPU the way auto arenas do.
2020-11-13 13:45:35 -08:00
David Goldblatt
4ca3d91e96 Rename geom_grow -> exp_grow.
This was promised in the review of the introduction of geom_grow, but would have
been painful to do there because of the series that introduced it.  Now that
those are comitted, renaming is easier.
2020-11-13 13:42:33 -08:00
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
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
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
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
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
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
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
Qi Wang
3de19ba401 Eagerly detect double free and sized dealloc bugs for large sizes. 2020-10-15 10:03:16 -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
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
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
Yinan Zhang
09eda2c9b6 Add unit tests for usize in prof recent records 2020-09-09 13:31:35 -07:00
David Goldblatt
131b1b5338 Rename ecache_grow -> geom_grow.
We're about to start using it outside of the ecaches, in the HPA central
allocator.
2020-08-19 16:53:21 -07:00
David Goldblatt
b399463fba flat_bitmap unit test: Silence a warning. 2020-08-17 12:50:27 -07:00
David Goldblatt
7b187360e9 IO: Support 0-padding for unsigned numbers. 2020-08-13 10:03:15 -07:00
David Goldblatt
eaed1e39be Add sized-delete size-checking functionality.
The existing checks are good at finding such issues (on tcache flush), but not
so good at pinpointing them.  Debug mode can find them, but sometimes debug mode
slows down a program so much that hard-to-hit bugs can take a long time to
crash.

This commit adds functionality to keep programs mostly on their fast paths,
while also checking every sized delete argument they get.
2020-08-05 19:34:05 -07:00
Yinan Zhang
f6cf5eb388 Add mallctl for batch allocation API 2020-07-31 09:16:50 -07:00
Yinan Zhang
978f830ee3 Add batch allocation API 2020-07-31 09:16:50 -07:00
David Goldblatt
ddb8dc4ad0 FB: Add range iteration support. 2020-07-30 15:25:23 -07:00
David Goldblatt
ceee823519 Add flat_bitmap.
The flat_bitmap module offers an extended API, at the cost of decreased
performance in the case of very large bitmaps.
2020-07-30 15:25:23 -07:00
David Goldblatt
efeab1f498 bitset test: Pull NBITS_TAB into its own file. 2020-07-30 15:25:23 -07:00
David Goldblatt
22da836094 bit_util: Add fls_ functions; "find last set".
These simplify a lot of the bit_util module, which had grown bits and pieces of
this functionality across a variety of places over the years.

While we're here, kill off BIT_UTIL_INLINE and don't do reentrancy testing for
bit_util.
2020-07-30 15:25:23 -07:00
David Goldblatt
1ed0288d9c bit_util: Change ffs functions indexing.
Making these 0-based instead of 1-based makes calling code simpler and will be
more consistent with functions introduced in subsequent diffs.
2020-07-30 15:25:23 -07:00
David Goldblatt
471eb5913c PAC: Move in decay rate setting. 2020-07-09 13:41:04 -07:00