Expand and restructure the rtree API such that all common operations can
be achieved with minimal work, regardless of whether the rtree leaf
fields are independent versus packed into a single atomic pointer.
This allows leaf elements to differ in size from internal node elements.
In principle it would be more correct to use a different type for each
level of the tree, but due to implementation details related to atomic
operations, we use casts anyway, thus counteracting the value of
additional type correctness. Furthermore, such a scheme would require
function code generation (via cpp macros), as well as either unwieldy
type names for leaves or type aliases, e.g.
typedef struct rtree_elm_d2_s rtree_leaf_elm_t;
This alternate strategy would be more correct, and with less code
duplication, but probably not worth the complexity.
Rather than storing usize only for large (and prof-promoted)
allocations, store the size class index for allocations that reside
within the extent, such that the size class index is valid for all
extents that contain extant allocations, and invalid otherwise (mainly
to make debugging simpler).
Split decay-based purging into two phases, the first of which uses lazy
purging to convert dirty pages to "muzzy", and the second of which uses
forced purging, decommit, or unmapping to convert pages to clean or
destroy them altogether. Not all operating systems support lazy
purging, yet the application may provide extent hooks that implement
lazy purging, so care must be taken to dynamically omit the first phase
when necessary.
The mallctl interfaces change as follows:
- opt.decay_time --> opt.{dirty,muzzy}_decay_time
- arena.<i>.decay_time --> arena.<i>.{dirty,muzzy}_decay_time
- arenas.decay_time --> arenas.{dirty,muzzy}_decay_time
- stats.arenas.<i>.pdirty --> stats.arenas.<i>.p{dirty,muzzy}
- stats.arenas.<i>.{npurge,nmadvise,purged} -->
stats.arenas.<i>.{dirty,muzzy}_{npurge,nmadvise,purged}
This resolves#521.
Refactor most of the decay-related functions to take as parameters the
decay_t and associated extents_t structures to operate on. This
prepares for supporting both lazy and forced purging on different decay
schedules.
These were all size_ts, so we have atomics support for them on all platforms, so
the conversion is straightforward.
Left non-atomic is curlextents, which AFAICT is not used atomically anywhere.
I expect this to be the trickiest conversion we will see, since we want atomics
on 64-bit platforms, but are also always able to piggyback on some sort of
external synchronization on non-64 bit platforms.
In the process, I changed the implementation of rtree_elm_acquire so that it
won't even try to CAS if its initial read (getting the extent + lock bit)
indicates that the CAS is doomed to fail. This can significantly improve
performance under contention.
The new feature, opt.percpu_arena, determines thread-arena association
dynamically based CPU id. Three modes are supported: "percpu", "phycpu"
and disabled.
"percpu" uses the current core id (with help from sched_getcpu())
directly as the arena index, while "phycpu" will assign threads on the
same physical CPU to the same arena. In other words, "percpu" means # of
arenas == # of CPUs, while "phycpu" has # of arenas == 1/2 * (# of
CPUs). Note that no runtime check on whether hyper threading is enabled
is added yet.
When enabled, threads will be migrated between arenas when a CPU change
is detected. In the current design, to reduce overhead from reading CPU
id, each arena tracks the thread accessed most recently. When a new
thread comes in, we will read CPU id and update arena if necessary.
When witness is enabled, lock rank order needs to be preserved during
prefork, not only for each arena, but also across arenas. This change
breaks arena_prefork into further stages to ensure valid rank order
across arenas. Also changed test/unit/fork to use a manual arena to
catch this case.
In the process, we can do some strength reduction, changing the fetch-adds and
fetch-subs to be simple loads followed by stores, since the modifications all
occur while holding the mutex.
The C11 atomics backport removed this #define, which degraded atomic 64-bit
reads to require a lock even on platforms that support them. This commit fixes
that.
This fixes tcache_flush for manual tcaches, which wasn't able to find
the correct arena it associated with. Also changed the decay test to
cover this case (by using manually created arenas).
These functions select the easiest-to-remove element in the heap, which
is either the most recently inserted aux list element or the root. If
no calls are made to first() or remove_first(), the behavior (and time
complexity) is the same as for a LIFO queue.
Rather than purging uncoalesced extents, perform just enough incremental
coalescing to purge only fully coalesced extents. In the absence of
cached extent reuse, the immediate versus delayed incremental purging
algorithms result in the same purge order.
This resolves#655.
In the C11 atomics backport, we couldn't use not_reached() in
atomic_enum_to_builtin (in atomic_gcc_atomic.h), since atomic.h was hermetic and
assert.h wasn't; there was a dependency issue. assert.h is hermetic now, so we
can include it.
This is the first header refactoring diff, #533. It splits the assert and util
components into separate, hermetic, header files. In the process, it splits out
two of the large sub-components of util (the stdio.h replacement, and bit
manipulation routines) into their own components (malloc_io.h and bit_util.h).
This is mostly to break up cyclic dependencies, but it also breaks off a good
chunk of the catch-all-ness of util, which is nice.
Convert the nrequests field to be partially derived, and the curlextents
to be fully derived, in order to reduce the number of stats updates
needed during common operations.
This change affects ndalloc stats during arena reset, because it is no
longer possible to cancel out ndalloc effects (curlextents would become
negative).
This introduces a backport of C11 atomics. It has four implementations; ranked
in order of preference, they are:
- GCC/Clang __atomic builtins
- GCC/Clang __sync builtins
- MSVC _Interlocked builtins
- C11 atomics, from <stdatomic.h>
The primary advantages are:
- Close adherence to the standard API gives us a defined memory model.
- Type safety: atomic objects are now separate types from non-atomic ones, so
that it's impossible to mix up atomic and non-atomic updates (which is
undefined behavior that compilers are starting to take advantage of).
- Efficiency: we can specify ordering for operations, avoiding fences and
atomic operations on strongly ordered architectures (example:
`atomic_write_u32(ptr, val);` involves a CAS loop, whereas
`atomic_store(ptr, val, ATOMIC_RELEASE);` is a plain store.
This diff leaves in the current atomics API (implementing them in terms of the
backport). This lets us transition uses over piecemeal.
Testing:
This is by nature hard to test. I've manually tested the first three options on
Linux on gcc by futzing with the #defines manually, on freebsd with gcc and
clang, on MSVC, and on OS X with clang. All of these were x86 machines though,
and we don't have any test infrastructure set up for non-x86 platforms.
In the long term, we'll transition to C99-style inline semantics. In the
short-term, this will allow both styles to coexist without breaking one another.
Remove obsolete unit test scaffolding for extent quantization. Remove
redundant assertions. Add an assertion to
extents_first_best_fit_locked() that should help prevent aligned
allocation regressions.
We don't touch witness at all when config_debug == false. Let's only pay the
memory cost in malloc_mutex_s when needed. Note that when !config_debug, we keep
the field in a union so that we don't have to do #ifdefs in multiple places.
Extent splitting and coalescing is a major component of large allocation
overhead, and disabling coalescing of cached extents provides a simple
and effective hysteresis mechanism. Once two-phase purging is
implemented, it will probably make sense to leave coalescing disabled
for the first phase, but coalesce during the second phase.
This avoids a gcc diagnostic note:
note: The ABI for passing parameters with 64-byte alignment has
changed in GCC 4.6
This note related to the cacheline alignment of rtree_ctx_t, which was
introduced by 4a346f5593 (Replace rtree
path cache with LRU cache.).
Fix rtree_subkey() to use uintptr_t rather than unsigned for key
bitmasking. This regression was introduced by
4a346f5593 (Replace rtree path cache with
LRU cache.).
Rather than dynamically building a table to aid per level computations,
define a constant table at compile time. Omit both high and low
insignificant bits. Use one to three tree levels, depending on the
number of significant bits.
Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather
than a single-path element lookup cache. The replacement is logically
much simpler, as well as slightly faster in the fast path case and less
prone to degraded performance during non-trivial sequences of lookups.
Refactor arena and extent locking protocols such that arena and
extent locks are never held when calling into the extent_*_wrapper()
API. This requires extra care during purging since the arena lock no
longer protects the inner purging logic. It also requires extra care to
protect extents from being merged with adjacent extents.
Convert extent_t's 'active' flag to an enumerated 'state', so that
retained extents are explicitly marked as such, rather than depending on
ring linkage state.
Refactor the extent collections (and their synchronization) for cached
and retained extents into extents_t. Incorporate LRU functionality to
support purging. Incorporate page count accounting, which replaces
arena->ndirty and arena->stats.retained.
Assert that no core locks are held when entering any internal
[de]allocation functions. This is in addition to existing assertions
that no locks are held when entering external [de]allocation functions.
Audit and document synchronization protocols for all arena_t fields.
This fixes a potential deadlock due to recursive allocation during
gdump, in a similar fashion to b49c649bc1
(Fix lock order reversal during gdump.), but with a necessarily much
broader code impact.
Synchronize tcaches with tcaches_mtx rather than ctl_mtx. Add missing
synchronization for tcache flushing. This bug was introduced by
1cb181ed63 (Implement explicit tcache
support.), which was first released in 4.0.0.
The SDK jemalloc is built against might be not be the latest for various
reasons, but the resulting binary ought to work on newer versions of
OSX.
In order to ensure this, we need the fullest definitions possible, so
copy what we need from the latest version of malloc/malloc.h available
on opensource.apple.com.
Mostly revert the prof_realloc() changes in
498856f44a (Move slabs out of chunks.) so
that prof_free_sampled_object() is called when appropriate. Leave the
prof_tctx_[re]set() optimization in place, but add an assertion to
verify that all eight cases are correctly handled. Add a comment to
make clear the code ordering, so that the regression originally fixed by
ea8d97b897 (Fix
prof_{malloc,free}_sample_object() call order in prof_realloc().) is not
repeated.
This resolves#499.
This is part of a broader change to make header files better represent the
dependencies between one another (see
https://github.com/jemalloc/jemalloc/issues/533). It breaks up component headers
into smaller parts that can be made to have a simpler dependency graph.
For the autogenerated headers (smoothstep.h and size_classes.h), no splitting
was necessary, so I didn't add support to emit multiple headers.
Currently, jemalloc detects sparc64 targets by checking whether
__sparc64__ is defined. However, this definition is used on BSD
targets only. Linux targets define both __sparc__ and __arch64__
for sparc64. Since this also works on BSD, rather use __sparc__
and __arch64__ instead of __sparc64__ to detect sparc64 targets.
Refactor ctl_stats_t to be a demand-zeroed non-growing data structure.
To keep the size from being onerous (~60 MiB) on 32-bit systems, convert
the arenas field to contain pointers rather than directly embedded
ctl_arena_stats_t elements.
Add the MALLCTL_ARENAS_ALL cpp macro as a fixed index for use
in accessing the arena.<i>.{purge,decay,dss} and stats.arenas.<i>.*
mallctls, and deprecate access via the arenas.narenas index (to be
removed in 6.0.0).
Add/rename related mallctls:
- Add stats.arenas.<i>.base .
- Rename stats.arenas.<i>.metadata to stats.arenas.<i>.internal .
- Add stats.arenas.<i>.resident .
Modify the arenas.extend mallctl to take an optional (extent_hooks_t *)
argument so that it is possible for all base allocations to be serviced
by the specified extent hooks.
This resolves#463.
Split purging into lazy and forced variants. Use the forced variant for
zeroing dss.
Add support for NULL function pointers as an opt-out mechanism for the
dalloc, commit, decommit, purge_lazy, purge_forced, split, and merge
fields of extent_hooks_t.
Add short-circuiting checks in large_ralloc_no_move_{shrink,expand}() so
that no attempt is made if splitting/merging is not supported.
This resolves#268.
If virtual memory is retained, allocate extents such that their sizes
form an exponentially growing series. This limits the number of
disjoint virtual memory ranges so that extent merging can be effective
even if multiple arenas' extent allocation requests are highly
interleaved.
This resolves#462.
Add the --with-lg-hugepage configure option, but automatically configure
LG_HUGEPAGE even if it isn't specified.
Add the pages_[no]huge() functions, which toggle huge page state via
madvise(..., MADV_[NO]HUGEPAGE) calls.
Rewrite arena_slab_regind() to provide sufficient constant data for
the compiler to perform division strength reduction. This replaces
more general manual strength reduction that was implemented before
arena_bin_info was compile-time-constant. It would be possible to
slightly improve on the compiler-generated division code by taking
advantage of range limits that the compiler doesn't know about.
Adds cpp bindings for jemalloc, along with necessary autoconf settings.
This is mostly to add sized deallocation support, which can't be added
from C directly. Sized deallocation is ~10% microbench improvement.
* Import ax_cxx_compile_stdcxx.m4 from the autoconf repo, seems like the
easiest way to get c++14 detection.
* Adds various other changes, like CXXFLAGS, to configure.ac.
* Adds new rules to Makefile.in for src/jemalloc-cpp.cpp, and a basic
unittest.
* Both new and delete are overridden, to ensure jemalloc is used for
both.
* TODO future enhancement of avoiding extra PLT thunks for new and
delete - sdallocx and malloc are publicly exported jemalloc symbols,
using an alias would link them directly. Unfortunately, was having
trouble getting it to play nice with jemalloc's namespace support.
Testing:
Tested gcc 4.8, gcc 5, gcc 5.2, clang 4.0. Only gcc >= 5 has sized
deallocation support, verified that the rest build correctly.
Tested mac osx and Centos.
Tested --with-jemalloc-prefix and --without-export.
This resolves#202.
Some versions of Android provide a pthreads library without providing
pthread_atfork(), so in practice a separate feature test is necessary
for the latter.
Add feature tests for the MADV_FREE and MADV_DONTNEED flags to
madvise(2), so that MADV_FREE is detected and used for Linux kernel
versions 4.5 and newer. Refactor pages_purge() so that on systems which
support both flags, MADV_FREE is preferred over MADV_DONTNEED.
This resolves#387.
Add extent serial numbers and use them where appropriate as a sort key
that is higher priority than address, so that the allocation policy
prefers older extents.
This resolves#147.
Add an "over-size" extent heap in which to store extents which exceed
the maximum size class (plus cache-oblivious padding, if enabled).
Remove psz2ind_clamp() and use psz2ind() instead so that trying to
allocate the maximum size class can in principle succeed. In practice,
this allows assertions to hold so that OOM errors can be successfully
generated.
Fix extent_alloc_cache[_locked]() to support decommitted allocation, and
use this ability in arena_stash_dirty(), so that decommitted extents are
not needlessly committed during purging. In practice this does not
happen on any currently supported systems, because both extent merging
and decommit must be implemented; all supported systems implement one
xor the other.
rtree_node_init spinlocks the node, allocates, and then sets the node.
This is under heavy contention at the top of the tree if many threads
start to allocate at the same time.
Instead, take a per-rtree sleeping mutex to reduce spinning. Tested
both pthreads and osx OSSpinLock, and both reduce spinning adequately
Previous benchmark time:
./ttest1 500 100
~15s
New benchmark time:
./ttest1 500 100
.57s
Fix zone_force_unlock() to reinitialize, rather than unlocking mutexes,
since OS X 10.12 cannot tolerate a child unlocking mutexes that were
locked by its parent.
Refactor; this was a side effect of experimenting with zone
{de,re}registration during fork(2).
The raw clock variant is slow (even relative to plain CLOCK_MONOTONIC),
whereas the coarse clock variant is faster than CLOCK_MONOTONIC, but
still has resolution (~1ms) that is adequate for our purposes.
This resolves#479.
glibc defines its malloc implementation with several weak and strong
symbols:
strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc)
strong_alias (__libc_free, __cfree) weak_alias (__libc_free, cfree)
strong_alias (__libc_free, __free) strong_alias (__libc_free, free)
strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc)
The issue is not with the weak symbols, but that other parts of glibc
depend on __libc_malloc explicitly. Defining them in terms of jemalloc
API's allows the linker to drop glibc's malloc.o completely from the link,
and static linking no longer results in symbol collisions.
Another wrinkle: jemalloc during initialization calls sysconf to
get the number of CPU's. GLIBC allocates for the first time before
setting up isspace (and other related) tables, which are used by
sysconf. Instead, use the pthread API to get the number of
CPUs with GLIBC, which seems to work.
This resolves#442.
Rather than relying on two's complement negation for alignment mask
generation, use bitwise not and addition. This dodges warnings from
MSVC, and should be strength-reduced by compiler optimization anyway.
Rather than protecting dss operations with a mutex, use atomic
operations. This has negligible impact on synchronization overhead
during typical dss allocation, but is a substantial improvement for
extent_in_dss() and the newly added extent_dss_mergeable(), which can be
called multiple times during extent deallocations.
This change also has the advantage of avoiding tsd in deallocation paths
associated with purging, which resolves potential deadlocks during
thread exit due to attempted tsd resurrection.
This resolves#425.
Add spin_t and spin_{init,adaptive}(), which provide a simple
abstraction for adaptive spinning.
Adaptively spin during busy waits in bootstrapping and rtree node
initialization.
Simplify decay-based purging attempts to only be triggered when the
epoch is advanced, rather than every time purgeable memory increases.
In a correctly functioning system (not previously the case; see below),
this only causes a behavior difference if during subsequent purge
attempts the least recently used (LRU) purgeable memory extent is
initially too large to be purged, but that memory is reused between
attempts and one or more of the next LRU purgeable memory extents are
small enough to be purged. In practice this is an arbitrary behavior
change that is within the set of acceptable behaviors.
As for the purging fix, assure that arena->decay.ndirty is recorded
*after* the epoch advance and associated purging occurs. Prior to this
fix, it was possible for purging during epoch advance to cause a
substantially underrepresentative (arena->ndirty - arena->decay.ndirty),
i.e. the number of dirty pages attributed to the current epoch was too
low, and a series of unintended purges could result. This fix is also
relevant in the context of the simplification described above, but the
bug's impact would be limited to over-purging at epoch advances.
Instead, move the epoch backward in time. Additionally, add
nstime_monotonic() and use it in debug builds to assert that time only
goes backward if nstime_update() is using a non-monotonic time source.
Add missing #include <time.h>. The critical time facilities appear to
have been transitively included via unistd.h and sys/time.h, but in
principle this omission was capable of having caused
clock_gettime(CLOCK_MONOTONIC, ...) to have been overlooked in favor of
gettimeofday(), which in turn could cause spurious non-monotonic time
updates.
Refactor nstime_get() out of nstime_update() and add configure tests for
all variants.
Add CLOCK_MONOTONIC_RAW support (Linux-specific) and
mach_absolute_time() support (OS X-specific).
Do not fall back to clock_gettime(CLOCK_REALTIME, ...). This was a
fragile Linux-specific workaround, which we're unlikely to use at all
now that clock_gettime(CLOCK_MONOTONIC_RAW, ...) is supported, and if we
have no choice besides non-monotonic clocks, gettimeofday() is only
incrementally worse.
Avoid calling s2u() on raw extent sizes in extent_recycle().
Clamp psz2ind() (implemented as psz2ind_clamp()) when inserting/removing
into/from size-segregated extent heaps.
GCC 4.9.3 cross-compiled for sparc64 defines __sparc_v9__, not
__sparc64__ nor __sparcv9. This prevents LG_QUANTUM from being defined
properly. Adding this new value to the check solves the issue.
Add a configure check for __builtin_unreachable instead of basing its
availability on the __GNUC__ version. On OS X using gcc (a real gcc, not the
bundled version that's just a gcc front-end) leads to a linker assertion:
https://github.com/jemalloc/jemalloc/issues/266
It turns out that this is caused by a gcc bug resulting from the use of
__builtin_unreachable():
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57438
To work around this bug, check that __builtin_unreachable() actually works at
configure time, and if it doesn't use abort() instead. The check is based on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57438#c21.
With this `make check` passes with a homebrew installed gcc-5 and gcc-6.
Some bug (either in the red-black tree code, or in the pgi compiler) seems to
cause red-black trees to become unbalanced. This issue seems to go away if we
don't use compact red-black trees. Since red-black trees don't seem to be used
much anymore, I opted for what seems to be an easy fix here instead of digging
in and trying to find the root cause of the bug.
Some context in case it's helpful:
I experienced a ton of segfaults while using pgi as Chapel's target compiler
with jemalloc 4.0.4. The little bit of debugging I did pointed me somewhere
deep in red-black tree manipulation, but I didn't get a chance to investigate
further. It looks like 4.2.0 replaced most uses of red-black trees with
pairing-heaps, which seems to avoid whatever bug I was hitting.
However, `make check_unit` was still failing on the rb test, so I figured the
core issue was just being masked. Here's the `make check_unit` failure:
```sh
=== test/unit/rb ===
test_rb_empty: pass
tree_recurse:test/unit/rb.c:90: Failed assertion: (((_Bool) (((uintptr_t) (left_node)->link.rbn_right_red) & ((size_t)1)))) == (false) --> true != false: Node should be black
test_rb_random:test/unit/rb.c:274: Failed assertion: (imbalances) == (0) --> 1 != 0: Tree is unbalanced
tree_recurse:test/unit/rb.c:90: Failed assertion: (((_Bool) (((uintptr_t) (left_node)->link.rbn_right_red) & ((size_t)1)))) == (false) --> true != false: Node should be black
test_rb_random:test/unit/rb.c:274: Failed assertion: (imbalances) == (0) --> 1 != 0: Tree is unbalanced
node_remove:test/unit/rb.c:190: Failed assertion: (imbalances) == (0) --> 2 != 0: Tree is unbalanced
<jemalloc>: test/unit/rb.c:43: Failed assertion: "pathp[-1].cmp < 0"
test/test.sh: line 22: 12926 Aborted
Test harness error
```
While starting to debug I saw the RB_COMPACT option and decided to check if
turning that off resolved the bug. It seems to have fixed it (`make check_unit`
passes and the segfaults under Chapel are gone) so it seems like on okay
work-around. I'd imagine this has performance implications for red-black trees
under pgi, but if they're not going to be used much anymore it's probably not a
big deal.
Revert 245ae6036c (Support --with-lg-page
values larger than actual page size.), because it could cause VM map
fragmentation if the kernel grows mmap()ed memory downward.
This resolves#391.
rtree-based extent lookups remain more expensive than chunk-based run
lookups, but with this optimization the fast path slowdown is ~3 CPU
cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles
prior. The path caching speedup tends to degrade gracefully unless
allocated memory is spread far apart (as is the case when using a
mixture of sbrk() and mmap()).
In the case where prof_alloc_prep() is called with an over-estimate of
allocation size, and sampling doesn't end up being triggered, the tctx
must be discarded.
When an allocation is large enough to trigger multiple dumps, use
modular math rather than subtraction to reset the interval counter.
Prior to this change, it was possible for a single allocation to cause
many subsequent allocations to all trigger profile dumps.
When updating usable size for a sampled object, try to cancel out
the difference between LARGE_MINCLASS and usable size from the interval
counter.
Look up chunk metadata via the radix tree, rather than using
CHUNK_ADDR2BASE().
Propagate pointer's containing extent.
Minimize extent lookups by doing a single lookup (e.g. in free()) and
propagating the pointer's extent into nearly all the functions that may
need it.
This makes it possible to acquire short-term "ownership" of rtree
elements so that it is possible to read an extent pointer *and* read the
extent's contents with a guarantee that the element will not be modified
until the ownership is released. This is intended as a mechanism for
resolving rtree read/write races rather than as a way to lock extents.
Use pszind_t size classes rather than szind_t size classes, and always
reserve space for NPSIZES elements. This removes unused heaps that are
not multiples of the page size, and adds (currently) unused heaps for
all huge size classes, with the immediate benefit that the size of
arena_t allocations is constant (no longer dependent on chunk size).
These compute size classes and indices similarly to size2index(),
index2size() and s2u(), respectively, but using the subset of size
classes that are multiples of the page size. Note that pszind_t and
szind_t are not interchangeable.
Short-circuit commonly called witness functions so that they only
execute in debug builds, and remove equivalent guards from mutex
functions. This avoids pointless code execution in
witness_assert_lockless(), which is typically called twice per
allocation/deallocation function invocation.
Inline commonly called witness functions so that optimized builds can
completely remove calls as dead code.
b2c0d6322d (Add witness, a simple online
locking validator.) caused a broad propagation of tsd throughout the
internal API, but tsd_fetch() was designed to fail prior to tsd
bootstrapping. Fix this by splitting tsd_t into non-nullable tsd_t and
nullable tsdn_t, and modifying all internal APIs that do not critically
rely on tsd to take nullable pointers. Furthermore, add the
tsd_booted_get() function so that tsdn_fetch() can probe whether tsd
bootstrapping is complete and return NULL if not. All dangerous
conversions of nullable pointers are tsdn_tsd() calls that assert-fail
on invalid conversion.
This is a broader application of optimizations to malloc() and free() in
f4a0f32d34 (Fast-path improvement:
reduce # of branches and unnecessary operations.).
This resolves#321.
If the OS overcommits:
- Commit all mappings in pages_map() regardless of whether the caller
requested committed memory.
- Linux-specific: Specify MAP_NORESERVE to avoid
unfortunate interactions with heuristic overcommit mode during
fork(2).
This resolves#193.
Split arena_choose() into arena_[i]choose() and use arena_ichoose() for
arena lookup during internal allocation. This fixes huge_palloc() so
that it always succeeds during extent node allocation.
This regression was introduced by
66cd953514 (Do not allocate metadata via
non-auto arenas, nor tcaches.).
Change test-related mangling to simplify symbol filtering.
The following commands can be used to detect missing/obsolete symbol
mangling, with the caveat that the full set of symbols is based on the
union of symbols generated by all configurations, some of which are
platform-specific:
./autogen.sh --enable-debug --enable-prof --enable-lazy-lock
make all tests
nm -a lib/libjemalloc.a src/*.jet.o \
|grep " [TDBCR] " \
|awk '{print $3}' \
|sed -e 's/^\(je_\|jet_\(n_\)\?\)\([a-zA-Z0-9_]*\)/\3/g' \
|LC_COLLATE=C sort -u \
|grep -v \
-e '^\(malloc\|calloc\|posix_memalign\|aligned_alloc\|realloc\|free\)$' \
-e '^\(m\|r\|x\|s\|d\|sd\|n\)allocx$' \
-e '^mallctl\(\|nametomib\|bymib\)$' \
-e '^malloc_\(stats_print\|usable_size\|message\)$' \
-e '^\(memalign\|valloc\)$' \
-e '^__\(malloc\|memalign\|realloc\|free\)_hook$' \
-e '^pthread_create$' \
> /tmp/private_symbols.txt
Fix a compilation error that occurs if Valgrind is not enabled. This
regression was caused by b2c0d6322d (Add
witness, a simple online locking validator.).
During over-allocation in preparation for creating aligned mappings,
allocate one more page than necessary if PAGE is the actual page size,
so that trimming still succeeds even if the system returns a mapping
that has less than PAGE alignment. This allows compiling with e.g. 64
KiB "pages" on systems that actually use 4 KiB pages.
Note that for e.g. --with-lg-page=21, it is also necessary to increase
the chunk size (e.g. --with-malloc-conf=lg_chunk:22) so that there are
at least two "pages" per chunk. In practice this isn't a particularly
compelling configuration because so much (unusable) virtual memory is
dedicated to chunk headers.
Refactor ph to support configurable comparison functions. Use a cpp
macro code generation form equivalent to the rb macros so that pairing
heaps can be used for both run heaps and chunk heaps.
Remove per node parent pointers, and instead use leftmost siblings' prev
pointers to track parents.
Fix multi-pass sibling merging to iterate over intermediate results
using a FIFO, rather than a LIFO. Use this fixed sibling merging
implementation for both merge phases of the auxiliary twopass algorithm
(first merging the aux list, then replacing the root with its merged
children). This fixes both degenerate merge behavior and the potential
for deep recursion.
This regression was introduced by
6bafa6678f (Pairing heap).
This resolves#371.
Fix bitmap_sfu() to shift by LG_BITMAP_GROUP_NBITS rather than
hard-coded 6 when using linear (non-USE_TREE) bitmap search. In
practice this affects only 64-bit systems for which sizeof(long) is not
8 (i.e. Windows), since USE_TREE is defined for 32-bit systems.
This regression was caused by b8823ab026
(Use linear scan for small bitmaps).
This resolves#368.
Move chunk_dalloc_arena()'s implementation into chunk_dalloc_wrapper(),
so that if the dalloc hook fails, proper decommit/purge/retain cascading
occurs. This fixes three potential chunk leaks on OOM paths, one during
dss-based chunk allocation, one during chunk header commit (currently
relevant only on Windows), and one during rtree write (e.g. if rtree
node allocation fails).
Merge chunk_purge_arena() into chunk_purge_default() (refactor, no
change to functionality).
The arenas_extend() function was renamed to arenas_init() in commit
8bb3198f72, but its function declaration
was not removed from jemalloc_internal.h.in.
Use pairing heap instead of red black tree in arena runs_avail. The
extra links are unioned with the bitmap_t, so this change doesn't use
any extra memory.
Canaries show this change to be a 1% cpu win, and 2% latency win. In
particular, large free()s, and small bin frees are now O(1) (barring
coalescing).
I also tested changing bin->runs to be a pairing heap, but saw a much
smaller win, and it would mean increasing the size of arena_run_s by two
pointers, so I left that as an rb-tree for now.
Initial implementation of a twopass pairing heap with aux list.
Research papers linked in comments.
Where search/nsearch/last aren't needed, this gives much faster first(),
delete(), and insert(). Insert is O(1), and first/delete don't have to
walk the whole tree.
Also tested rb_old with parent pointers - it was better than the current
rb.h for memory loads, but still much worse than a pairing heap.
An array-based heap would be much faster if everything fits in memory,
but on a cold cache it has many more memory loads for most operations.
Add a cast to avoid comparing a ssize_t value to a uint64_t value that
is always larger than a 32-bit ssize_t. This silences an innocuous
compiler warning from e.g. gcc 4.2.1 about the comparison always having
the same result.
Add missing stats.arenas.<i>.{dss,lg_dirty_mult,decay_time}
initialization.
Fix stats.arenas.<i>.{pactive,pdirty} to read under the protection of
the arena mutex.
Fix stats.cactive accounting to always increase/decrease by multiples of
the chunk size, even for huge size classes that are not multiples of the
chunk size, e.g. {2.5, 3, 3.5, 5, 7} MiB with 2 MiB chunk size. This
regression was introduced by 155bfa7da1
(Normalize size classes.) and first released in 4.0.0.
This resolves#336.
For small bitmaps, a linear scan of the bitmap is slightly faster than
a tree search - bitmap_t is more compact, and there are fewer writes
since we don't have to propogate state transitions up the tree.
On x86_64 with the current settings, I'm seeing ~.5%-1% CPU improvement
in production canaries with this change.
The old tree code is left since 32bit sizes are much larger (and ffsl
smaller), and maybe the run sizes will change in the future.
This resolves#339.
Refactor the arenas array, which contains pointers to all extant arenas,
such that it starts out as a sparse array of maximum size, and use
double-checked atomics-based reads as the basis for fast and simple
arena_get(). Additionally, reduce arenas_lock's role such that it only
protects against arena initalization races. These changes remove the
possibility for arena lookups to trigger locking, which resolves at
least one known (fork-related) deadlock.
This resolves#315.
Attempt mmap-based in-place huge reallocation by plumbing new_addr into
chunk_alloc_mmap(). This can dramatically speed up incremental huge
reallocation.
This resolves#335.
Separate run trees by index, replacing the previous quantize logic.
Quantization by index is now performed only on insertion / removal from
the tree, and not on node comparison, saving some cpu. This also means
we don't have to dereference the miscelm* pointers, saving half of the
memory loads from miscelms/mapbits that have fallen out of cache. A
linear scan of the indicies appears to be fast enough.
The only cost of this is an extra tree array in each arena.
Use a single uint64_t in nstime_t to store nanoseconds rather than using
struct timespec. This reduces fragility around conversions between long
and uint64_t, especially missing casts that only cause problems on
32-bit platforms.
This is an alternative to the existing ratio-based unused dirty page
purging, and is intended to eventually become the sole purging
mechanism.
Add mallctls:
- opt.purge
- opt.decay_time
- arena.<i>.decay
- arena.<i>.decay_time
- arenas.decay_time
- stats.arenas.<i>.decay_time
This resolves#325.
- Combine multiple runtime branches into a single malloc_slow check.
- Avoid calling arena_choose / size2index / index2size on fast path.
- A few micro optimizations.
ex_destroy iterates over the tree using post-order traversal so nodes
can be removed and processed by the callback function without paying the
cost to rebalance the tree. The destruction process cannot be stopped
once started.
Fix xallocx(..., MALLOCX_ZERO to zero the last full trailing page of
large allocations that have been randomly assigned an offset of 0 when
--enable-cache-oblivious configure option is enabled. This addresses a
special case missed in d260f442ce (Fix
xallocx(..., MALLOCX_ZERO) bugs.).
Don't assume Bourne shell is in /bin/sh when running size_classes.sh .
Consider __sparcv9 a synonym for __sparc64__ when defining LG_QUANTUM.
This resolves#275.
Add arena_prof_tctx_reset() and use it instead of arena_prof_tctx_set()
when resetting the tctx pointer during reallocation, which happens
whenever an originally sampled reallocated object is not sampled during
reallocation.
This regression was introduced by
594c759f37 (Optimize
arena_prof_tctx_set().)
Fix prof_realloc() to call prof_free_sampled_object() after calling
prof_malloc_sample_object(). Prior to this fix, if tctx and old_tctx
were the same, the tctx could have been prematurely destroyed.
Make one call to prof_active_get_unlocked() per allocation event, and
use the result throughout the relevant functions that handle an
allocation event. Also add a missing check in prof_realloc(). These
fixes protect allocation events against concurrent prof_active changes.
Fix heap profiling to distinguish among otherwise identical sample sites
with interposed resets (triggered via the "prof.reset" mallctl). This
bug could cause data structure corruption that would most likely result
in a segfault.
This didn't cause bad code generation in the one case spot-checked (gcc
4.8.1), but had the potential to to so. This bug was introduced by
594c759f37 (Optimize
arena_prof_tctx_set().).
Don't bitshift by negative amounts when encoding/decoding run sizes in
chunk header maps. This affected systems with page sizes greater than 8
KiB.
Reported by Ingvar Hagelund <ingvar@redpill-linpro.com>.
Only set the unzeroed flag when initializing the entire mapbits entry,
rather than mutating just the unzeroed bit. This simplifies the
possible mapbits state transitions.
Cascade from decommit to purge when purging unused dirty pages, so that
it is possible to decommit cleaned memory rather than just purging. For
non-Windows debug builds, decommit runs rather than purging them, since
this causes access of deallocated runs to segfault.
This resolves#251.
In builds with profiling disabled (default), the opt_prof_prefix array
has a one byte length as a micro-optimization. This will cause the usage
of write in the unused profiling code to be statically detected as a
buffer overflow by Bionic's _FORTIFY_SOURCE implementation as it tries
to detect read overflows in addition to write overflows.
This works around the problem by informing the compiler that
not_reached() means code in unreachable in release builds.
- Decorate public function with __declspec(allocator) and __declspec(restrict), just like MSVC 1900
- Support JEMALLOC_HAS_RESTRICT by defining the restrict keyword
- Move __declspec(nothrow) between 'void' and '*' so it compiles once more
Add the "arena.<i>.chunk_hooks" mallctl, which replaces and expands on
the "arena.<i>.chunk.{alloc,dalloc,purge}" mallctls. The chunk hooks
allow control over chunk allocation/deallocation, decommit/commit,
purging, and splitting/merging, such that the application can rely on
jemalloc's internal chunk caching and retaining functionality, yet
implement a variety of chunk management mechanisms and policies.
Merge the chunks_[sz]ad_{mmap,dss} red-black trees into
chunks_[sz]ad_retained. This slightly reduces how hard jemalloc tries
to honor the dss precedence setting; prior to this change the precedence
setting was also consulted when recycling chunks.
Fix chunk purging. Don't purge chunks in arena_purge_stashed(); instead
deallocate them in arena_unstash_purged(), so that the dirty memory
linkage remains valid until after the last time it is used.
This resolves#176 and #201.
huge_ralloc() passes a size that may not be precisely a size class, so
make huge_palloc() handle the more general case of a size input rather
than usize.
This regression appears to have been introduced by the addition of
in-place huge reallocation; as such it was never incorporated into a
release.
Take large_pad into account when determining whether an aligned
allocation can be satisfied by a large size class.
This regression was introduced by
8a03cf039c (Implement cache index
randomization for large allocations.).
Create and use FMT* macros that are equivalent to the PRI* macros that
inttypes.h defines. This allows uniform use of the Unix-specific format
specifiers, e.g. "%zu", as well as avoiding Windows-specific definitions
of e.g. PRIu64.
Add ffs()/ffsl() support for compiling with gcc.
Extract compatibility definitions of ENOENT, EINVAL, EAGAIN, EPERM,
ENOMEM, and ENORANGE into include/msvc_compat/windows_extra.h and
use the file for tests as well as for core jemalloc code.
Replace JEMALLOC_ATTR(format(printf, ...). with
JEMALLOC_FORMAT_PRINTF(), so that configuration feature tests can
omit the attribute if it would cause extraneous compilation warnings.
This effectively reverts 97c04a9383 (Use
first-fit rather than first-best-fit run/chunk allocation.). In some
pathological cases, first-fit search dominates allocation time, and it
also tends not to converge as readily on a steady state of memory
layout, since precise allocation order has a bigger effect than for
first-best-fit.
Conditionally define ENOENT, EINVAL, etc. (was unconditional).
Add/use PRIzu, PRIzd, and PRIzx for use in malloc_printf() calls. gcc issued
(harmless) warnings since e.g. "%zu" should be "%Iu" on Windows, and the
alternative to this workaround would have been to disable the function
attributes which cause gcc to look for type mismatches in formatted printing
function calls.
- Set opt_lg_chunk based on run-time OS setting
- Verify LG_PAGE is compatible with run-time OS setting
- When targeting Windows Vista or newer, use SRWLOCK instead of CRITICAL_SECTION
- When targeting Windows Vista or newer, statically initialize init_lock
Fix size class overflow handling for malloc(), posix_memalign(),
memalign(), calloc(), and realloc() when profiling is enabled.
Remove an assertion that erroneously caused arena_sdalloc() to fail when
profiling was enabled.
This resolves#232.
Now that small allocation runs have fewer regions due to run metadata
residing in chunk headers, an explicit minimum tcache count is needed to
make sure that tcache adequately amortizes synchronization overhead.
Take into account large_pad when computing whether to pass the
deallocation request to tcache_dalloc_large(), so that the largest
cacheable size makes it back to tcache. This regression was introduced
by 8a03cf039c (Implement cache index
randomization for large allocations.).
Extract szad size quantization into {extent,run}_quantize(), and .
quantize szad run sizes to the union of valid small region run sizes and
large run sizes.
Refactor iteration in arena_run_first_fit() to use
run_quantize{,_first,_next(), and add support for padded large runs.
For large allocations that have no specified alignment constraints,
compute a pseudo-random offset from the beginning of the first backing
page that is a multiple of the cache line size. Under typical
configurations with 4-KiB pages and 64-byte cache lines this results in
a uniform distribution among 64 page boundary offsets.
Add the --disable-cache-oblivious option, primarily intended for
performance testing.
This resolves#13.
However, unlike before it was removed do not force --enable-ivsalloc
when Darwin zone allocator integration is enabled, since the zone
allocator code uses ivsalloc() regardless of whether
malloc_usable_size() and sallocx() do.
This resolves#211.
Add mallctls:
- arenas.lg_dirty_mult is initialized via opt.lg_dirty_mult, and can be
modified to change the initial lg_dirty_mult setting for newly created
arenas.
- arena.<i>.lg_dirty_mult controls an individual arena's dirty page
purging threshold, and synchronously triggers any purging that may be
necessary to maintain the constraint.
- arena.<i>.chunk.purge allows the per arena dirty page purging function
to be replaced.
This resolves#93.
Remove the prof_tctx_state_destroying transitory state and instead add
the tctx_uid field, so that the tuple <thr_uid, tctx_uid> uniquely
identifies a tctx. This assures that tctx's are well ordered even when
more than two with the same thr_uid coexist. A previous attempted fix
based on prof_tctx_state_destroying was only sufficient for protecting
against two coexisting tctx's, but it also introduced a new dumping
race.
These regressions were introduced by
602c8e0971 (Implement per thread heap
profiling.) and 764b00023f (Fix a heap
profiling regression.).
Add the prof_tctx_state_destroying transitionary state to fix a race
between a thread destroying a tctx and another thread creating a new
equivalent tctx.
This regression was introduced by
602c8e0971 (Implement per thread heap
profiling.).
This tends to more effectively pack active memory toward low addresses.
However, additional tree searches are required in many cases, so whether
this change stands the test of time will depend on real-world
benchmarks.
Recent changes have improved huge allocation scalability, which removes
upward pressure to set the chunk size so large that huge allocations are
rare. Smaller chunks are more likely to completely drain, so set the
default to the smallest size that doesn't leave excessive unusable
trailing space in chunk headers.
TlsGetValue has a semantic difference with pthread_getspecific, in that it
can return a non-error NULL value, so it always sets the LastError.
But allocator callers may not be expecting calling e.g. free() to change
the value of the last error, so preserve it.
Rename "dirty chunks" to "cached chunks", in order to avoid overloading
the term "dirty".
Fix the regression caused by 339c2b23b2
(Fix chunk_unmap() to propagate dirty state.), and actually address what
that change attempted, which is to only purge chunks once, and propagate
whether zeroed pages resulted into chunk_record().
Fix chunk_unmap() to propagate whether a chunk is dirty, and modify
dirty chunk purging to record this information so it can be passed to
chunk_unmap(). Since the broken version of chunk_unmap() claimed that
all chunks were clean, this resulted in potential memory corruption for
purging implementations that do not zero (e.g. MADV_FREE).
This regression was introduced by
ee41ad409a (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.
This regression was introduced by
88fef7ceda (Refactor huge_*() calls into
arena internals.), and went undetected because of the --enable-debug
regression.
This regression was introduced by
88fef7ceda (Refactor huge_*() calls into
arena internals.), and went undetected because of the --enable-debug
regression.
Although exceedingly unlikely, it appears that writes to the prof_tctx
field of arena_chunk_map_misc_t could be reordered such that a stale
value could be read during deallocation, with profiler metadata
corruption and invalid pointer dereferences being the most likely
effects.
Migrate all centralized data structures related to huge allocations and
recyclable chunks into arena_t, so that each arena can manage huge
allocations and recyclable virtual memory completely independently of
other arenas.
Add chunk node caching to arenas, in order to avoid contention on the
base allocator.
Use chunks_rtree to look up huge allocations rather than a red-black
tree. Maintain a per arena unsorted list of huge allocations (which
will be needed to enumerate huge allocations during arena reset).
Remove the --enable-ivsalloc option, make ivsalloc() always available,
and use it for size queries if --enable-debug is enabled. The only
practical implications to this removal are that 1) ivsalloc() is now
always available during live debugging (and the underlying radix tree is
available during core-based debugging), and 2) size query validation can
no longer be enabled independent of --enable-debug.
Remove the stats.chunks.{current,total,high} mallctls, and replace their
underlying statistics with simpler atomically updated counters used
exclusively for gdump triggering. These statistics are no longer very
useful because each arena manages chunks independently, and per arena
statistics provide similar information.
Simplify chunk synchronization code, now that base chunk allocation
cannot cause recursive lock acquisition.
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.
Fix arena_get() to refresh the cache as needed in the (!init_if_missing
&& refresh_if_missing) case.
This flaw was introduced by the initial arena_get() implementation,
which was part of 8bb3198f72 (Refactor/fix
arenas manipulation.).
Recent huge allocation refactoring associates huge allocations with
arenas, but it remains necessary to quickly look up huge allocation
metadata during reallocation/deallocation. A global radix tree remains
a good solution to this problem, but locking would have become the
primary bottleneck after (upcoming) migration of chunk management from
global to per arena data structures.
This lock-free implementation uses double-checked reads to traverse the
tree, so that in the steady state, each read or write requires only a
single atomic operation.
This implementation also assures that no more than two tree levels
actually exist, through a combination of careful virtual memory
allocation which makes large sparse nodes cheap, and skipping the root
node on x64 (possible because the top 16 bits are all 0 in practice).
Refactor base_alloc() to guarantee that allocations are carved from
demand-zeroed virtual memory. This supports sparse data structures such
as multi-page radix tree nodes.
Enhance base_alloc() to keep track of fragments which were too small to
support previous allocation requests, and try to consume them during
subsequent requests. This becomes important when request sizes commonly
approach or exceed the chunk size (as could radix tree node
allocations).
This feature makes it possible to toggle the gdump feature on/off during
program execution, whereas the the opt.prof_dump mallctl value can only
be set during program startup.
This resolves#72.
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.
Refactor bootstrapping to delay tsd initialization, primarily to support
integration with FreeBSD's libc.
Refactor a0*() for internal-only use, and add the
bootstrap_{malloc,calloc,free}() API for use by FreeBSD's libc. This
separation limits use of the a0*() functions to metadata allocation,
which doesn't require malloc/calloc/free API compatibility.
This resolves#170.
In addition to true/false, opt.junk can now be either "alloc" or "free",
giving applications the possibility of junking memory only on allocation
or deallocation.
This resolves#172.
This provides in-place expansion of huge allocations when the end of the
allocation is at the end of the sbrk heap. There's already the ability
to extend in-place via recycled chunks but this handles the initial
growth of the heap via repeated vector / string reallocations.
A possible future extension could allow realloc to go from the following:
| huge allocation | recycled chunks |
^ dss_end
To a larger allocation built from recycled *and* new chunks:
| huge allocation |
^ dss_end
Doing that would involve teaching the chunk recycling code to request
new chunks to satisfy the request. The chunk_dss code wouldn't require
any further changes.
#include <stdlib.h>
int main(void) {
size_t chunk = 4 * 1024 * 1024;
void *ptr = NULL;
for (size_t size = chunk; size < chunk * 128; size *= 2) {
ptr = realloc(ptr, size);
if (!ptr) return 1;
}
}
dss:secondary: 0.083s
dss:primary: 0.083s
After:
dss:secondary: 0.083s
dss:primary: 0.003s
The dss heap grows in the upwards direction, so the oldest chunks are at
the low addresses and they are used first. Linux prefers to grow the
mmap heap downwards, so the trick will not work in the *current* mmap
chunk allocator as a huge allocation will only be at the top of the heap
in a contrived case.
Fix quarantine to actually update tsd when expanding, and to avoid
double initialization (leaking the first quarantine) due to recursive
initialization.
This resolves#161.
* use sized deallocation in iralloct_realign
* iralloc and ixalloc always need the old size, so pass it in from the
caller where it's often already calculated
Add per size class huge allocation statistics, and normalize various
stats:
- Change the arenas.nlruns type from size_t to unsigned.
- Add the arenas.nhchunks and arenas.hchunks.<i>.size mallctl's.
- Replace the stats.arenas.<i>.bins.<j>.allocated mallctl with
stats.arenas.<i>.bins.<j>.curregs .
- Add the stats.arenas.<i>.hchunks.<j>.nmalloc,
stats.arenas.<i>.hchunks.<j>.ndalloc,
stats.arenas.<i>.hchunks.<j>.nrequests, and
stats.arenas.<i>.hchunks.<j>.curhchunks mallctl's.
Fix a prof_tctx_t/prof_tdata_t cleanup race by storing a copy of thr_uid
in prof_tctx_t, so that the associated tdata need not be present during
tctx teardown.
Remove code in arena_dalloc_bin_run() that preserved the "clean" state
of trailing clean pages by splitting them into a separate run during
deallocation. This was a useful mechanism for reducing dirty page
churn when bin runs comprised many pages, but bin runs are now quite
small.
Remove the nextind field from arena_run_t now that it is no longer
needed, and change arena_run_t's bin field (arena_bin_t *) to binind
(index_t). These two changes remove 8 bytes of chunk header overhead
per page, which saves 1/512 of all arena chunk memory.
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.
This avoids grabbing the base mutex, as a step towards fine-grained
locking for huge allocations. The thread cache also provides a tiny
(~3%) improvement for serial huge allocations.
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.
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).
This adds support for expanding huge allocations in-place by requesting
memory at a specific address from the chunk allocator.
It's currently only implemented for the chunk recycling path, although
in theory it could also be done by optimistically allocating new chunks.
On Linux, it could attempt an in-place mremap. However, that won't work
in practice since the heap is grown downwards and memory is not unmapped
(in a normal build, at least).
Repeated vector reallocation micro-benchmark:
#include <string.h>
#include <stdlib.h>
int main(void) {
for (size_t i = 0; i < 100; i++) {
void *ptr = NULL;
size_t old_size = 0;
for (size_t size = 4; size < (1 << 30); size *= 2) {
ptr = realloc(ptr, size);
if (!ptr) return 1;
memset(ptr + old_size, 0xff, size - old_size);
old_size = size;
}
free(ptr);
}
}
The glibc allocator fails to do any in-place reallocations on this
benchmark once it passes the M_MMAP_THRESHOLD (default 128k) but it
elides the cost of copies via mremap, which is currently not something
that jemalloc can use.
With this improvement, jemalloc still fails to do any in-place huge
reallocations for the first outer loop, but then succeeds 100% of the
time for the remaining 99 iterations. The time spent doing allocations
and copies drops down to under 5%, with nearly all of it spent doing
purging + faulting (when huge pages are disabled) and the array memset.
An improved mremap API (MREMAP_RETAIN - #138) would be far more general
but this is a portable optimization and would still be useful on Linux
for xallocx.
Numbers with transparent huge pages enabled:
glibc (copies elided via MREMAP_MAYMOVE): 8.471s
jemalloc: 17.816s
jemalloc + no-op madvise: 13.236s
jemalloc + this commit: 6.787s
jemalloc + this commit + no-op madvise: 6.144s
Numbers with transparent huge pages disabled:
glibc (copies elided via MREMAP_MAYMOVE): 15.403s
jemalloc: 39.456s
jemalloc + no-op madvise: 12.768s
jemalloc + this commit: 15.534s
jemalloc + this commit + no-op madvise: 6.354s
Closes#137
Don't disable tcache when lazy-lock is configured. There already exists
a mechanism to disable tcache, but doing so automatically due to
lazy-lock causes surprising performance behavior.
Fix tsd cleanup regressions that were introduced in
5460aa6f66 (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.
Implement/test/fix the opt.prof_thread_active_init,
prof.thread_active_init, and thread.prof.active mallctl's.
Test/fix the thread.prof.name mallctl.
Refactor opt_prof_active to be read-only and move mutable state into the
prof_active variable. Stop leaning on ctl-related locking for
protection.
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.
Fix a race that caused a non-critical assertion failure. To trigger the
race, a thread had to be part way through initializing a new sample,
such that it was discoverable by the dumping thread, but not yet linked
into its gctx by the time a later dump phase would normally have reset
its state to 'nominal'.
Additionally, lock access to the state field during modification to
transition to the dumping state. It's not apparent that this oversight
could have caused an actual problem due to outer locking that protects
the dumping machinery, but the added locking pedantically follows the
stated locking protocol for the state field.
* assertion failure
* malloc_init failure
* malloc not already initialized (in malloc_init)
* running in valgrind
* thread cache disabled at runtime
Clang and GCC already consider a comparison with NULL or -1 to be cold,
so many branches (out-of-memory) are already correctly considered as
cold and marking them is not important.
Fix a profile sampling race that was due to preparing to sample, yet
doing nothing to assure that the context remains valid until the stats
are updated.
These regressions were caused by
602c8e0971 (Implement per thread heap
profiling.), which did not make it into any releases prior to these
fixes.
Fix prof_tdata_get() to avoid dereferencing an invalid tdata pointer
(when it's PROF_TDATA_STATE_{REINCARNATED,PURGATORY}).
Fix prof_tdata_get() callers to check for invalid results besides NULL
(PROF_TDATA_STATE_{REINCARNATED,PURGATORY}).
These regressions were caused by
602c8e0971 (Implement per thread heap
profiling.), which did not make it into any releases prior to these
fixes.
This adds a new `sdallocx` function to the external API, allowing the
size to be passed by the caller. It avoids some extra reads in the
thread cache fast path. In the case where stats are enabled, this
avoids the work of calculating the size from the pointer.
An assertion validates the size that's passed in, so enabling debugging
will allow users of the API to debug cases where an incorrect size is
passed in.
The performance win for a contrived microbenchmark doing an allocation
and immediately freeing it is ~10%. It may have a different impact on a
real workload.
Closes#28
Optimize [nmd]alloc() fast paths such that the (flags == 0) case is
streamlined, flags decoding only happens to the minimum degree
necessary, and no conditionals are repeated.
__*_hook() is glibc, but on at least one glibc platform (homebrew),
the __GLIBC__ define isn't set correctly and we miss being able to
use these hooks.
Do a feature test for it during configuration so that we enable it
anywhere the hooks are actually available.
Rename data structures (prof_thr_cnt_t-->prof_tctx_t,
prof_ctx_t-->prof_gctx_t), and convert to storing a prof_tctx_t for
sampled objects.
Convert PROF_ALLOC_PREP() to prof_alloc_prep(), since precise backtrace
depth within jemalloc functions is no longer an issue (pprof prunes
irrelevant frames).
Implement mallctl's:
- prof.reset implements full sample data reset, and optional change of
sample interval.
- prof.lg_sample reads the current sample interval (opt.lg_prof_sample
was the permanent source of truth prior to prof.reset).
- thread.prof.name provides naming capability for threads within heap
profile dumps.
- thread.prof.active makes it possible to activate/deactivate heap
profiling for individual threads.
Modify the heap dump files to contain per thread heap profile data.
This change is incompatible with the existing pprof, which will require
enhancements to read and process the enriched data.
Treat prof_tdata_t's bt2cnt as a comprehensive map of the thread's
extant allocation samples (do not limit the total number of entries).
This helps prepare the way for per thread heap profiling.
Fix runs_dirty-based purging to also purge dirty pages in the spare
chunk.
Refactor runs_dirty manipulation into arena_dirty_{insert,remove}(), and
move the arena->ndirty accounting into those functions.
Remove the u.ql_link field from arena_chunk_map_t, and get rid of the
enclosing union for u.rb_link, since only rb_link remains.
Remove the ndirty field from arena_chunk_t.
Some platforms, such as Google's Portable Native Client, use Newlib and
thus lack access to madvise(2). In those instances, pages_purge() is
transformed into a no-op.
Some platforms (like those using Newlib) don't have ffs/ffsl. This
commit adds a check to configure.ac for __builtin_ffsl if ffsl isn't
found. __builtin_ffsl performs the same function as ffsl, and has the
added benefit of being available on any platform utilizing
Gcc-compatible compiler.
This change does not address the used of ffs in the MALLOCX_ARENA()
macro.
Add size class computation capability, currently used only as validation
of the size class lookup tables. Generalize the size class spacing used
for bins, for eventual use throughout the full range of allocation
sizes.
Refactor huge allocation to be managed by arenas (though the global
red-black tree of huge allocations remains for lookup during
deallocation). This is the logical conclusion of recent changes that 1)
made per arena dss precedence apply to huge allocation, and 2) made it
possible to replace the per arena chunk allocation/deallocation
functions.
Remove the top level huge stats, and replace them with per arena huge
stats.
Normalize function names and types to *dalloc* (some were *dealloc*).
Remove the --enable-mremap option. As jemalloc currently operates, this
is a performace regression for some applications, but planned work to
logarithmically space huge size classes should provide similar amortized
performance. The motivation for this change was that mremap-based huge
reallocation forced leaky abstractions that prevented refactoring.
Add new mallctl endpoints "arena<i>.chunk.alloc" and
"arena<i>.chunk.dealloc" to allow userspace to configure
jemalloc's chunk allocator and deallocator on a per-arena
basis.
Simplify backtracing to not ignore any frames, and compensate for this
in pprof in order to increase flexibility with respect to function-based
refactoring even in the presence of non-deterministic inlining. Modify
pprof to blacklist all jemalloc allocation entry points including
non-standard ones like mallocx(), and ignore all allocator-internal
frames. Prior to this change, pprof excluded the specifically
blacklisted functions from backtraces, but it left allocator-internal
frames intact.
Fix debug-only compilation failures introduced by changes to
prof_sample_accum_update() in:
6c39f9e059
refactor profiling. only use a bytes till next sample variable.
Forcefully disable tcache if running inside Valgrind, and remove
Valgrind calls in tcache-specific code.
Restructure Valgrind-related code to move most Valgrind calls out of the
fast path functions.
Take advantage of static knowledge to elide some branches in
JEMALLOC_VALGRIND_REALLOC().
Make dss non-optional on all platforms which support sbrk(2).
Fix the "arena.<i>.dss" mallctl to return an error if "primary" or
"secondary" precedence is specified, but sbrk(2) is not supported.
Make promotion of sampled small objects to large objects mandatory, so
that profiling metadata can always be stored in the chunk map, rather
than requiring one pointer per small region in each small-region page
run. In practice the non-prof-promote code was only useful when using
jemalloc to track all objects and report them as leaks at program exit.
However, Valgrind is at least as good a tool for this particular use
case.
Furthermore, the non-prof-promote code is getting in the way of
some optimizations that will make heap profiling much cheaper for the
predominant use case (sampling a small representative proportion of all
allocations).
When you call free() we load chunk->arena even though that
data isn't used on the tcache hot path.
In profiling some FB applications, I found that ~30% of the
dTLB misses in the free() function come from this line. With
4 MB chunks, the arena_chunk_t->map is ~ 32 KB (1024 pages
in the chunk, 4 8 byte pointers in arena_chunk_map_t). This
means there's only a 1/8 chance of the page containing
chunk->arena also comtaining the map bits.
The hash code, which has MurmurHash3 at its core, generates different
output depending on system endianness, so adapt the expected output on
big-endian systems. MurmurHash3 code also makes the assumption that
unaligned access is okay (not true on all systems), but jemalloc only
hashes data structures that have sufficient alignment to dodge this
limitation.
Add a cpp #define that removes 'restrict' keyword usage unless the
compiler definitely supports C99. As written, 'restrict' is only
enabled if the compiler supports the -std=gnu99 option (e.g. gcc and
llvm).
Reported by Tobias Hieta.
Avoid copying "jeprof" to a 1-byte buffer within prof_boot0() when heap
profiling is disabled. Although this is dead code under such
conditions, the compiler doesn't figure that part out.
Reported by Eduardo Silva.
