1 Following are change highlights associated with official releases. Important
2 bug fixes are all mentioned, but some internal enhancements are omitted here for
3 brevity. Much more detail can be found in the git revision history:
5 https://github.com/jemalloc/jemalloc
7 * 4.3.1 (November 7, 2016)
10 - Fix a severe virtual memory leak. This regression was first released in
11 4.3.0. (@interwq, @jasone)
12 - Refactor atomic and prng APIs to restore support for 32-bit platforms that
13 use pre-C11 toolchains, e.g. FreeBSD's mips. (@jasone)
15 * 4.3.0 (November 4, 2016)
17 This is the first release that passes the test suite for multiple Windows
18 configurations, thanks in large part to @glandium setting up continuous
19 integration via AppVeyor (and Travis CI for Linux and OS X).
22 - Add "J" (JSON) support to malloc_stats_print(). (@jasone)
23 - Add Cray compiler support. (@ronawho)
26 - Add/use adaptive spinning for bootstrapping and radix tree node
27 initialization. (@jasone)
30 - Fix large allocation to search starting in the optimal size class heap,
31 which can substantially reduce virtual memory churn and fragmentation. This
32 regression was first released in 4.0.0. (@mjp41, @jasone)
33 - Fix stats.arenas.<i>.nthreads accounting. (@interwq)
34 - Fix and simplify decay-based purging. (@jasone)
35 - Make DSS (sbrk(2)-related) operations lockless, which resolves potential
36 deadlocks during thread exit. (@jasone)
37 - Fix over-sized allocation of radix tree leaf nodes. (@mjp41, @ogaun,
39 - Fix over-sized allocation of arena_t (plus associated stats) data
40 structures. (@jasone, @interwq)
41 - Fix EXTRA_CFLAGS to not affect configuration. (@jasone)
42 - Fix a Valgrind integration bug. (@ronawho)
43 - Disallow 0x5a junk filling when running in Valgrind. (@jasone)
44 - Fix a file descriptor leak on Linux. This regression was first released in
45 4.2.0. (@vsarunas, @jasone)
46 - Fix static linking of jemalloc with glibc. (@djwatson)
47 - Use syscall(2) rather than {open,read,close}(2) during boot on Linux. This
48 works around other libraries' system call wrappers performing reentrant
49 allocation. (@kspinka, @Whissi, @jasone)
50 - Fix OS X default zone replacement to work with OS X 10.12. (@glandium,
52 - Fix cached memory management to avoid needless commit/decommit operations
53 during purging, which resolves permanent virtual memory map fragmentation
54 issues on Windows. (@mjp41, @jasone)
55 - Fix TSD fetches to avoid (recursive) allocation. This is relevant to
56 non-TLS and Windows configurations. (@jasone)
57 - Fix malloc_conf overriding to work on Windows. (@jasone)
58 - Forcibly disable lazy-lock on Windows (was forcibly *enabled*). (@jasone)
60 * 4.2.1 (June 8, 2016)
63 - Fix bootstrapping issues for configurations that require allocation during
64 tsd initialization (e.g. --disable-tls). (@cferris1000, @jasone)
65 - Fix gettimeofday() version of nstime_update(). (@ronawho)
66 - Fix Valgrind regressions in calloc() and chunk_alloc_wrapper(). (@ronawho)
67 - Fix potential VM map fragmentation regression. (@jasone)
68 - Fix opt_zero-triggered in-place huge reallocation zeroing. (@jasone)
69 - Fix heap profiling context leaks in reallocation edge cases. (@jasone)
71 * 4.2.0 (May 12, 2016)
74 - Add the arena.<i>.reset mallctl, which makes it possible to discard all of
75 an arena's allocations in a single operation. (@jasone)
76 - Add the stats.retained and stats.arenas.<i>.retained statistics. (@jasone)
77 - Add the --with-version configure option. (@jasone)
78 - Support --with-lg-page values larger than actual page size. (@jasone)
81 - Use pairing heaps rather than red-black trees for various hot data
82 structures. (@djwatson, @jasone)
83 - Streamline fast paths of rtree operations. (@jasone)
84 - Optimize the fast paths of calloc() and [m,d,sd]allocx(). (@jasone)
85 - Decommit unused virtual memory if the OS does not overcommit. (@jasone)
86 - Specify MAP_NORESERVE on Linux if [heuristic] overcommit is active, in order
87 to avoid unfortunate interactions during fork(2). (@jasone)
90 - Fix chunk accounting related to triggering gdump profiles. (@jasone)
91 - Link against librt for clock_gettime(2) if glibc < 2.17. (@jasone)
92 - Scale leak report summary according to sampling probability. (@jasone)
96 This bugfix release resolves a variety of mostly minor issues, though the
97 bitmap fix is critical for 64-bit Windows.
100 - Fix the linear scan version of bitmap_sfu() to shift by the proper amount
101 even when sizeof(long) is not the same as sizeof(void *), as on 64-bit
103 - Fix hashing functions to avoid unaligned memory accesses (and resulting
104 crashes). This is relevant at least to some ARM-based platforms.
106 - Fix fork()-related lock rank ordering reversals. These reversals were
107 unlikely to cause deadlocks in practice except when heap profiling was
108 enabled and active. (@jasone)
109 - Fix various chunk leaks in OOM code paths. (@jasone)
110 - Fix malloc_stats_print() to print opt.narenas correctly. (@jasone)
111 - Fix MSVC-specific build/test issues. (@rustyx, @yuslepukhin)
112 - Fix a variety of test failures that were due to test fragility rather than
115 * 4.1.0 (February 28, 2016)
117 This release is primarily about optimizations, but it also incorporates a lot
118 of portability-motivated refactoring and enhancements. Many people worked on
119 this release, to an extent that even with the omission here of minor changes
120 (see git revision history), and of the people who reported and diagnosed
121 issues, so much of the work was contributed that starting with this release,
122 changes are annotated with author credits to help reflect the collaborative
126 - Implement decay-based unused dirty page purging, a major optimization with
127 mallctl API impact. This is an alternative to the existing ratio-based
128 unused dirty page purging, and is intended to eventually become the sole
129 purging mechanism. New mallctls:
133 + arena.<i>.decay_time
135 + stats.arenas.<i>.decay_time
137 - Add --with-malloc-conf, which makes it possible to embed a default
138 options string during configuration. This was motivated by the desire to
139 specify --with-malloc-conf=purge:decay , since the default must remain
140 purge:ratio until the 5.0.0 release. (@jasone)
141 - Add MS Visual Studio 2015 support. (@rustyx, @yuslepukhin)
142 - Make *allocx() size class overflow behavior defined. The maximum
143 size class is now less than PTRDIFF_MAX to protect applications against
144 numerical overflow, and all allocation functions are guaranteed to indicate
145 errors rather than potentially crashing if the request size exceeds the
146 maximum size class. (@jasone)
148 + Add raw heap profile support. (@jasone)
149 + Add --retain and --exclude for backtrace symbol filtering. (@jasone)
152 - Optimize the fast path to combine various bootstrapping and configuration
153 checks and execute more streamlined code in the common case. (@interwq)
154 - Use linear scan for small bitmaps (used for small object tracking). In
155 addition to speeding up bitmap operations on 64-bit systems, this reduces
156 allocator metadata overhead by approximately 0.2%. (@djwatson)
157 - Separate arena_avail trees, which substantially speeds up run tree
158 operations. (@djwatson)
159 - Use memoization (boot-time-computed table) for run quantization. Separate
160 arena_avail trees reduced the importance of this optimization. (@jasone)
161 - Attempt mmap-based in-place huge reallocation. This can dramatically speed
162 up incremental huge reallocation. (@jasone)
164 Incompatible changes:
165 - Make opt.narenas unsigned rather than size_t. (@jasone)
168 - Fix stats.cactive accounting regression. (@rustyx, @jasone)
169 - Handle unaligned keys in hash(). This caused problems for some ARM systems.
170 (@jasone, @cferris1000)
171 - Refactor arenas array. In addition to fixing a fork-related deadlock, this
172 makes arena lookups faster and simpler. (@jasone)
173 - Move retained memory allocation out of the default chunk allocation
174 function, to a location that gets executed even if the application installs
175 a custom chunk allocation function. This resolves a virtual memory leak.
177 - Fix a potential tsd cleanup leak. (@cferris1000, @jasone)
178 - Fix run quantization. In practice this bug had no impact unless
179 applications requested memory with alignment exceeding one page.
181 - Fix LinuxThreads-specific bootstrapping deadlock. (Cosmin Paraschiv)
183 + Don't discard curl options if timeout is not defined. (@djwatson)
184 + Detect failed profile fetches. (@djwatson)
185 - Fix stats.arenas.<i>.{dss,lg_dirty_mult,decay_time,pactive,pdirty} for
186 --disable-stats case. (@jasone)
188 * 4.0.4 (October 24, 2015)
190 This bugfix release fixes another xallocx() regression. No other regressions
191 have come to light in over a month, so this is likely a good starting point
192 for people who prefer to wait for "dot one" releases with all the major issues
196 - Fix xallocx(..., MALLOCX_ZERO to zero the last full trailing page of large
197 allocations that have been randomly assigned an offset of 0 when
198 --enable-cache-oblivious configure option is enabled.
200 * 4.0.3 (September 24, 2015)
202 This bugfix release continues the trend of xallocx() and heap profiling fixes.
205 - Fix xallocx(..., MALLOCX_ZERO) to zero all trailing bytes of large
206 allocations when --enable-cache-oblivious configure option is enabled.
207 - Fix xallocx(..., MALLOCX_ZERO) to zero trailing bytes of huge allocations
208 when resizing from/to a size class that is not a multiple of the chunk size.
209 - Fix prof_tctx_dump_iter() to filter out nodes that were created after heap
210 profile dumping started.
211 - Work around a potentially bad thread-specific data initialization
212 interaction with NPTL (glibc's pthreads implementation).
214 * 4.0.2 (September 21, 2015)
216 This bugfix release addresses a few bugs specific to heap profiling.
219 - Fix ixallocx_prof_sample() to never modify nor create sampled small
220 allocations. xallocx() is in general incapable of moving small allocations,
221 so this fix removes buggy code without loss of generality.
222 - Fix irallocx_prof_sample() to always allocate large regions, even when
223 alignment is non-zero.
224 - Fix prof_alloc_rollback() to read tdata from thread-specific data rather
225 than dereferencing a potentially invalid tctx.
227 * 4.0.1 (September 15, 2015)
229 This is a bugfix release that is somewhat high risk due to the amount of
230 refactoring required to address deep xallocx() problems. As a side effect of
231 these fixes, xallocx() now tries harder to partially fulfill requests for
232 optional extra space. Note that a couple of minor heap profiling
233 optimizations are included, but these are better thought of as performance
234 fixes that were integral to disovering most of the other bugs.
237 - Avoid a chunk metadata read in arena_prof_tctx_set(), since it is in the
238 fast path when heap profiling is enabled. Additionally, split a special
239 case out into arena_prof_tctx_reset(), which also avoids chunk metadata
241 - Optimize irallocx_prof() to optimistically update the sampler state. The
242 prior implementation appears to have been a holdover from when
243 rallocx()/xallocx() functionality was combined as rallocm().
246 - Fix TLS configuration such that it is enabled by default for platforms on
247 which it works correctly.
248 - Fix arenas_cache_cleanup() and arena_get_hard() to handle
249 allocation/deallocation within the application's thread-specific data
250 cleanup functions even after arenas_cache is torn down.
251 - Fix xallocx() bugs related to size+extra exceeding HUGE_MAXCLASS.
252 - Fix chunk purge hook calls for in-place huge shrinking reallocation to
253 specify the old chunk size rather than the new chunk size. This bug caused
254 no correctness issues for the default chunk purge function, but was
255 visible to custom functions set via the "arena.<i>.chunk_hooks" mallctl.
256 - Fix heap profiling bugs:
257 + Fix heap profiling to distinguish among otherwise identical sample sites
258 with interposed resets (triggered via the "prof.reset" mallctl). This bug
259 could cause data structure corruption that would most likely result in a
261 + Fix irealloc_prof() to prof_alloc_rollback() on OOM.
262 + Make one call to prof_active_get_unlocked() per allocation event, and use
263 the result throughout the relevant functions that handle an allocation
264 event. Also add a missing check in prof_realloc(). These fixes protect
265 allocation events against concurrent prof_active changes.
266 + Fix ixallocx_prof() to pass usize_max and zero to ixallocx_prof_sample()
267 in the correct order.
268 + Fix prof_realloc() to call prof_free_sampled_object() after calling
269 prof_malloc_sample_object(). Prior to this fix, if tctx and old_tctx were
270 the same, the tctx could have been prematurely destroyed.
271 - Fix portability bugs:
272 + Don't bitshift by negative amounts when encoding/decoding run sizes in
273 chunk header maps. This affected systems with page sizes greater than 8
275 + Rename index_t to szind_t to avoid an existing type on Solaris.
276 + Add JEMALLOC_CXX_THROW to the memalign() function prototype, in order to
277 match glibc and avoid compilation errors when including both
278 jemalloc/jemalloc.h and malloc.h in C++ code.
279 + Don't assume that /bin/sh is appropriate when running size_classes.sh
280 during configuration.
281 + Consider __sparcv9 a synonym for __sparc64__ when defining LG_QUANTUM.
282 + Link tests to librt if it contains clock_gettime(2).
284 * 4.0.0 (August 17, 2015)
286 This version contains many speed and space optimizations, both minor and
287 major. The major themes are generalization, unification, and simplification.
288 Although many of these optimizations cause no visible behavior change, their
289 cumulative effect is substantial.
292 - Normalize size class spacing to be consistent across the complete size
293 range. By default there are four size classes per size doubling, but this
294 is now configurable via the --with-lg-size-class-group option. Also add the
295 --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and
296 --with-lg-tiny-min options, which can be used to tweak page and size class
298 + Worst case performance for incrementally growing/shrinking reallocation
299 is improved because there are far fewer size classes, and therefore
300 copying happens less often.
301 + Internal fragmentation is limited to 20% for all but the smallest size
302 classes (those less than four times the quantum). (1B + 4 KiB)
303 and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation.
304 + Chunk fragmentation tends to be lower because there are fewer distinct run
306 - Add support for explicit tcaches. The "tcache.create", "tcache.flush", and
307 "tcache.destroy" mallctls control tcache lifetime and flushing, and the
308 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API
309 control which tcache is used for each operation.
310 - Implement per thread heap profiling, as well as the ability to
311 enable/disable heap profiling on a per thread basis. Add the "prof.reset",
312 "prof.lg_sample", "thread.prof.name", "thread.prof.active",
313 "opt.prof_thread_active_init", "prof.thread_active_init", and
314 "thread.prof.active" mallctls.
315 - Add support for per arena application-specified chunk allocators, configured
316 via the "arena.<i>.chunk_hooks" mallctl.
317 - Refactor huge allocation to be managed by arenas, so that arenas now
318 function as general purpose independent allocators. This is important in
319 the context of user-specified chunk allocators, aside from the scalability
320 benefits. Related new statistics:
321 + The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc",
322 "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests"
323 mallctls provide high level per arena huge allocation statistics.
324 + The "arenas.nhchunks", "arenas.hchunk.<i>.size",
325 "stats.arenas.<i>.hchunks.<j>.nmalloc",
326 "stats.arenas.<i>.hchunks.<j>.ndalloc",
327 "stats.arenas.<i>.hchunks.<j>.nrequests", and
328 "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class
330 - Add the 'util' column to malloc_stats_print() output, which reports the
331 proportion of available regions that are currently in use for each small
333 - Add "alloc" and "free" modes for for junk filling (see the "opt.junk"
334 mallctl), so that it is possible to separately enable junk filling for
335 allocation versus deallocation.
336 - Add the jemalloc-config script, which provides information about how
337 jemalloc was configured, and how to integrate it into application builds.
338 - Add metadata statistics, which are accessible via the "stats.metadata",
339 "stats.arenas.<i>.metadata.mapped", and
340 "stats.arenas.<i>.metadata.allocated" mallctls.
341 - Add the "stats.resident" mallctl, which reports the upper limit of
342 physically resident memory mapped by the allocator.
343 - Add per arena control over unused dirty page purging, via the
344 "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and
345 "stats.arenas.<i>.lg_dirty_mult" mallctls.
346 - Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump
347 feature on/off during program execution.
348 - Add sdallocx(), which implements sized deallocation. The primary
349 optimization over dallocx() is the removal of a metadata read, which often
350 suffers an L1 cache miss.
351 - Add missing header includes in jemalloc/jemalloc.h, so that applications
352 only have to #include <jemalloc/jemalloc.h>.
353 - Add support for additional platforms:
362 - Maintain dirty runs in per arena LRUs rather than in per arena trees of
363 dirty-run-containing chunks. In practice this change significantly reduces
364 dirty page purging volume.
365 - Integrate whole chunks into the unused dirty page purging machinery. This
366 reduces the cost of repeated huge allocation/deallocation, because it
367 effectively introduces a cache of chunks.
368 - Split the arena chunk map into two separate arrays, in order to increase
369 cache locality for the frequently accessed bits.
370 - Move small run metadata out of runs, into arena chunk headers. This reduces
371 run fragmentation, smaller runs reduce external fragmentation for small size
372 classes, and packed (less uniformly aligned) metadata layout improves CPU
373 cache set distribution.
374 - Randomly distribute large allocation base pointer alignment relative to page
375 boundaries in order to more uniformly utilize CPU cache sets. This can be
376 disabled via the --disable-cache-oblivious configure option, and queried via
377 the "config.cache_oblivious" mallctl.
378 - Micro-optimize the fast paths for the public API functions.
379 - Refactor thread-specific data to reside in a single structure. This assures
380 that only a single TLS read is necessary per call into the public API.
381 - Implement in-place huge allocation growing and shrinking.
382 - Refactor rtree (radix tree for chunk lookups) to be lock-free, and make
383 additional optimizations that reduce maximum lookup depth to one or two
384 levels. This resolves what was a concurrency bottleneck for per arena huge
385 allocation, because a global data structure is critical for determining
386 which arenas own which huge allocations.
388 Incompatible changes:
389 - Replace --enable-cc-silence with --disable-cc-silence to suppress spurious
391 - Assure that the constness of malloc_usable_size()'s return type matches that
392 of the system implementation.
393 - Change the heap profile dump format to support per thread heap profiling,
394 rename pprof to jeprof, and enhance it with the --thread=<n> option. As a
395 result, the bundled jeprof must now be used rather than the upstream
397 - Disable "opt.prof_final" by default, in order to avoid atexit(3), which can
398 internally deadlock on some platforms.
399 - Change the "arenas.nlruns" mallctl type from size_t to unsigned.
400 - Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with
401 "stats.arenas.<i>.bins.<j>.curregs".
402 - Ignore MALLOC_CONF in set{uid,gid,cap} binaries.
403 - Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the
404 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage.
407 - Remove the *allocm() API, which is superseded by the *allocx() API.
408 - Remove the --enable-dss options, and make dss non-optional on all platforms
409 which support sbrk(2).
410 - Remove the "arenas.purge" mallctl, which was obsoleted by the
411 "arena.<i>.purge" mallctl in 3.1.0.
412 - Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically
413 detects whether it is running inside Valgrind.
414 - Remove the "stats.huge.allocated", "stats.huge.nmalloc", and
415 "stats.huge.ndalloc" mallctls.
416 - Remove the --enable-mremap option.
417 - Remove the "stats.chunks.current", "stats.chunks.total", and
418 "stats.chunks.high" mallctls.
421 - Fix the cactive statistic to decrease (rather than increase) when active
422 memory decreases. This regression was first released in 3.5.0.
423 - Fix OOM handling in memalign() and valloc(). A variant of this bug existed
424 in all releases since 2.0.0, which introduced these functions.
425 - Fix an OOM-related regression in arena_tcache_fill_small(), which could
426 cause cache corruption on OOM. This regression was present in all releases
427 from 2.2.0 through 3.6.0.
428 - Fix size class overflow handling for malloc(), posix_memalign(), memalign(),
429 calloc(), and realloc() when profiling is enabled.
430 - Fix the "arena.<i>.dss" mallctl to return an error if "primary" or
431 "secondary" precedence is specified, but sbrk(2) is not supported.
432 - Fix fallback lg_floor() implementations to handle extremely large inputs.
433 - Ensure the default purgeable zone is after the default zone on OS X.
434 - Fix latent bugs in atomic_*().
435 - Fix the "arena.<i>.dss" mallctl to handle read-only calls.
436 - Fix tls_model configuration to enable the initial-exec model when possible.
437 - Mark malloc_conf as a weak symbol so that the application can override it.
438 - Correctly detect glibc's adaptive pthread mutexes.
439 - Fix the --without-export configure option.
441 * 3.6.0 (March 31, 2014)
443 This version contains a critical bug fix for a regression present in 3.5.0 and
447 - Fix a regression in arena_chunk_alloc() that caused crashes during
448 small/large allocation if chunk allocation failed. In the absence of this
449 bug, chunk allocation failure would result in allocation failure, e.g. NULL
450 return from malloc(). This regression was introduced in 3.5.0.
451 - Fix backtracing for gcc intrinsics-based backtracing by specifying
452 -fno-omit-frame-pointer to gcc. Note that the application (and all the
453 libraries it links to) must also be compiled with this option for
454 backtracing to be reliable.
455 - Use dss allocation precedence for huge allocations as well as small/large
457 - Fix test assertion failure message formatting. This bug did not manifest on
458 x86_64 systems because of implementation subtleties in va_list.
459 - Fix inconsequential test failures for hash and SFMT code.
462 - Support heap profiling on FreeBSD. This feature depends on the proc
463 filesystem being mounted during heap profile dumping.
465 * 3.5.1 (February 25, 2014)
467 This version primarily addresses minor bugs in test code.
470 - Configure Solaris/Illumos to use MADV_FREE.
471 - Fix junk filling for mremap(2)-based huge reallocation. This is only
472 relevant if configuring with the --enable-mremap option specified.
473 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the
475 - Add a configure test for SSE2 rather than assuming it is usable on i686
476 systems. This fixes test compilation errors, especially on 32-bit Linux
478 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit
480 - Fix/remove flawed alignment-related overflow tests.
481 - Prevent compiler optimizations that could change backtraces in the
482 prof_accum unit test.
484 * 3.5.0 (January 22, 2014)
486 This version focuses on refactoring and automated testing, though it also
487 includes some non-trivial heap profiling optimizations not mentioned below.
490 - Add the *allocx() API, which is a successor to the experimental *allocm()
491 API. The *allocx() functions are slightly simpler to use because they have
492 fewer parameters, they directly return the results of primary interest, and
493 mallocx()/rallocx() avoid the strict aliasing pitfall that
494 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is
495 slated for removal in the next non-bugfix release.
496 - Add support for LinuxThreads.
499 - Unless heap profiling is enabled, disable floating point code and don't link
500 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64
501 systems, makes it possible to completely disable floating point register
502 use. Some versions of glibc neglect to save/restore caller-saved floating
503 point registers during dynamic lazy symbol loading, and the symbol loading
504 code uses whatever malloc the application happens to have linked/loaded
505 with, the result being potential floating point register corruption.
506 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling
507 backtrace creation in imemalign(). This bug impacted posix_memalign() and
509 - Fix a file descriptor leak in a prof_dump_maps() error path.
510 - Fix prof_dump() to close the dump file descriptor for all relevant error
512 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for
513 allocation, not just deallocation.
514 - Fix a data race for large allocation stats counters.
515 - Fix a potential infinite loop during thread exit. This bug occurred on
516 Solaris, and could affect other platforms with similar pthreads TSD
518 - Don't junk-fill reallocations unless usable size changes. This fixes a
519 violation of the *allocx()/*allocm() semantics.
520 - Fix growing large reallocation to junk fill new space.
521 - Fix huge deallocation to junk fill when munmap is disabled.
522 - Change the default private namespace prefix from empty to je_, and change
523 --with-private-namespace-prefix so that it prepends an additional prefix
524 rather than replacing je_. This reduces the likelihood of applications
525 which statically link jemalloc experiencing symbol name collisions.
526 - Add missing private namespace mangling (relevant when
527 --with-private-namespace is specified).
528 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as
529 static even for debug builds.
530 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice
531 this error path is never executed.
532 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These
533 bugs had no impact except for malformed inputs.
534 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by
535 existing calls, so they had no impact.
537 * 3.4.1 (October 20, 2013)
540 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption
541 of internal data structures and subsequent crashes.
542 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of
543 uninitialized memory in:
544 + arena chunk headers
545 + internal zero-initialized data structures (relevant to tcache and prof
547 - Preserve errno during the first allocation. A readlink(2) call during
548 initialization fails unless /etc/malloc.conf exists, so errno was typically
549 set during the first allocation prior to this fix.
550 - Fix compilation warnings reported by gcc 4.8.1.
552 * 3.4.0 (June 2, 2013)
554 This version is essentially a small bugfix release, but the addition of
555 aarch64 support requires that the minor version be incremented.
558 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were
559 typically triggered by multiple threads concurrently deallocating huge
563 - Add support for the aarch64 architecture.
565 * 3.3.1 (March 6, 2013)
567 This version fixes bugs that are typically encountered only when utilizing
568 custom run-time options.
571 - Fix a locking order bug that could cause deadlock during fork if heap
572 profiling were enabled.
573 - Fix a chunk recycling bug that could cause the allocator to lose track of
574 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause
575 corruption if allocating via sbrk(2) (unlikely unless running with the
576 "dss:primary" option specified). This was completely harmless on Linux
577 unless using mlockall(2) (and unlikely even then, unless the
578 --disable-munmap configure option or the "dss:primary" option was
579 specified). This regression was introduced in 3.1.0 by the
580 mlockall(2)/madvise(2) interaction fix.
581 - Fix TLS-related memory corruption that could occur during thread exit if the
582 thread never allocated memory. Only the quarantine and prof facilities were
584 - Fix two quarantine bugs:
585 + Internal reallocation of the quarantined object array leaked the old
587 + Reallocation failure for internal reallocation of the quarantined object
588 array (very unlikely) resulted in memory corruption.
589 - Fix Valgrind integration to annotate all internally allocated memory in a
590 way that keeps Valgrind happy about internal data structure access.
591 - Fix building for s390 systems.
593 * 3.3.0 (January 23, 2013)
595 This version includes a few minor performance improvements in addition to the
596 listed new features and bug fixes.
599 - Add clipping support to lg_chunk option processing.
600 - Add the --enable-ivsalloc option.
601 - Add the --without-export option.
602 - Add the --disable-zone-allocator option.
605 - Fix "arenas.extend" mallctl to output the number of arenas.
606 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory
608 - Fix build break on FreeBSD related to alloca.h.
610 * 3.2.0 (November 9, 2012)
612 In addition to a couple of bug fixes, this version modifies page run
613 allocation and dirty page purging algorithms in order to better control
614 page-level virtual memory fragmentation.
616 Incompatible changes:
617 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
620 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
621 after primary dss allocation fails.
622 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced
623 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
625 * 3.1.0 (October 16, 2012)
628 - Auto-detect whether running inside Valgrind, thus removing the need to
629 manually specify MALLOC_CONF=valgrind:true.
630 - Add the "arenas.extend" mallctl, which allows applications to create
631 manually managed arenas.
632 - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
633 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
634 which provide control over dss/mmap precedence.
635 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
636 - Define LG_QUANTUM for hppa.
638 Incompatible changes:
639 - Disable tcache by default if running inside Valgrind, in order to avoid
640 making unallocated objects appear reachable to Valgrind.
641 - Drop const from malloc_usable_size() argument on Linux.
644 - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
645 - Remove const from __*_hook variable declarations, so that glibc can modify
646 them during process forking.
647 - Fix mlockall(2)/madvise(2) interaction.
648 - Fix fork(2)-related deadlocks.
649 - Fix error return value for "thread.tcache.enabled" mallctl.
651 * 3.0.0 (May 11, 2012)
653 Although this version adds some major new features, the primary focus is on
654 internal code cleanup that facilitates maintainability and portability, most
655 of which is not reflected in the ChangeLog. This is the first release to
656 incorporate substantial contributions from numerous other developers, and the
657 result is a more broadly useful allocator (see the git revision history for
658 contribution details). Note that the license has been unified, thanks to
659 Facebook granting a license under the same terms as the other copyright
660 holders (see COPYING).
663 - Implement Valgrind support, redzones, and quarantine.
664 - Add support for additional platforms:
668 + Windows (no support yet for replacing the system malloc)
669 - Add support for additional architectures:
673 - Add support for cross compiling.
674 - Add nallocm(), which rounds a request size up to the nearest size class
675 without actually allocating.
676 - Implement aligned_alloc() (blame C11).
677 - Add the "thread.tcache.enabled" mallctl.
678 - Add the "opt.prof_final" mallctl.
679 - Update pprof (from gperftools 2.0).
680 - Add the --with-mangling option.
681 - Add the --disable-experimental option.
682 - Add the --disable-munmap option, and make it the default on Linux.
683 - Add the --enable-mremap option, which disables use of mremap(2) by default.
685 Incompatible changes:
686 - Enable stats by default.
687 - Enable fill by default.
688 - Disable lazy locking by default.
689 - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
690 - Rename the "arenas.pagesize" mallctl to "arenas.page".
691 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
692 - Change the "opt.prof_accum" default from true to false.
695 - Remove the swap feature, including the "config.swap", "swap.avail",
696 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
697 - Remove highruns statistics, including the
698 "stats.arenas.<i>.bins.<j>.highruns" and
699 "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
700 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
701 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
702 "arenas.[tqcs]bins" mallctls.
703 - Remove the "arenas.chunksize" mallctl.
704 - Remove the "opt.lg_prof_tcmax" option.
705 - Remove the "opt.lg_prof_bt_max" option.
706 - Remove the "opt.lg_tcache_gc_sweep" option.
707 - Remove the --disable-tiny option, including the "config.tiny" mallctl.
708 - Remove the --enable-dynamic-page-shift configure option.
709 - Remove the --enable-sysv configure option.
712 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
713 invalid statistics and crashes.
714 - Work around TLS deallocation via free() on Linux. This bug could cause
715 write-after-free memory corruption.
716 - Fix a potential deadlock that could occur during interval- and
717 growth-triggered heap profile dumps.
718 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
719 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could
720 cause memory corruption and crashes with --enable-dss specified.
721 - Fix fork-related bugs that could cause deadlock in children between fork
723 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
724 - Fix realloc(p, 0) to act like free(p).
725 - Do not enforce minimum alignment in memalign().
726 - Check for NULL pointer in malloc_usable_size().
727 - Fix an off-by-one heap profile statistics bug that could be observed in
728 interval- and growth-triggered heap profiles.
729 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
731 - Fix bin->runcur management to fix a layout policy bug. This bug did not
733 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
734 initialized than necessary.
735 - Add missing "opt.lg_tcache_max" mallctl implementation.
736 - Use glibc allocator hooks to make mixed allocator usage less likely.
737 - Fix build issues for --disable-tcache.
738 - Don't mangle pthread_create() when --with-private-namespace is specified.
740 * 2.2.5 (November 14, 2011)
743 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that
744 could cause memory corruption and/or crashes.
745 - Fix huge_ralloc() to maintain chunk statistics.
746 - Fix malloc_stats_print(..., "a") output.
748 * 2.2.4 (November 5, 2011)
751 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as
752 well as for --disable-tls builds in earlier releases.
753 - Do not assume a 4 KiB page size in test/rallocm.c.
755 * 2.2.3 (August 31, 2011)
757 This version fixes numerous bugs related to heap profiling.
760 - Fix a prof-related race condition. This bug could cause memory corruption,
761 but only occurred in non-default configurations (prof_accum:false).
762 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
763 excluded from backtraces).
764 - Fix a prof-related bug in realloc() (only triggered by OOM errors).
765 - Fix prof-related bugs in allocm() and rallocm().
766 - Fix prof_tdata_cleanup() for --disable-tls builds.
767 - Fix a relative include path, to fix objdir builds.
769 * 2.2.2 (July 30, 2011)
772 - Fix a build error for --disable-tcache.
773 - Fix assertions in arena_purge() (for real this time).
774 - Add the --with-private-namespace option. This is a workaround for symbol
775 conflicts that can inadvertently arise when using static libraries.
777 * 2.2.1 (March 30, 2011)
780 - Implement atomic operations for x86/x64. This fixes compilation failures
781 for versions of gcc that are still in wide use.
782 - Fix an assertion in arena_purge().
784 * 2.2.0 (March 22, 2011)
786 This version incorporates several improvements to algorithms and data
787 structures that tend to reduce fragmentation and increase speed.
790 - Add the "stats.cactive" mallctl.
791 - Update pprof (from google-perftools 1.7).
792 - Improve backtracing-related configuration logic, and add the
793 --disable-prof-libgcc option.
796 - Change default symbol visibility from "internal", to "hidden", which
797 decreases the overhead of library-internal function calls.
798 - Fix symbol visibility so that it is also set on OS X.
799 - Fix a build dependency regression caused by the introduction of the .pic.o
800 suffix for PIC object files.
801 - Add missing checks for mutex initialization failures.
802 - Don't use libgcc-based backtracing except on x64, where it is known to work.
803 - Fix deadlocks on OS X that were due to memory allocation in
804 pthread_mutex_lock().
805 - Heap profiling-specific fixes:
806 + Fix memory corruption due to integer overflow in small region index
807 computation, when using a small enough sample interval that profiling
808 context pointers are stored in small run headers.
809 + Fix a bootstrap ordering bug that only occurred with TLS disabled.
810 + Fix a rallocm() rsize bug.
811 + Fix error detection bugs for aligned memory allocation.
813 * 2.1.3 (March 14, 2011)
816 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
818 - Fix a "thread.arena" mallctl bug.
819 - Fix a thread cache stats merging bug.
821 * 2.1.2 (March 2, 2011)
824 - Fix "thread.{de,}allocatedp" mallctl for OS X.
825 - Add missing jemalloc.a to build system.
827 * 2.1.1 (January 31, 2011)
830 - Fix aligned huge reallocation (affected allocm()).
831 - Fix the ALLOCM_LG_ALIGN macro definition.
832 - Fix a heap dumping deadlock.
833 - Fix a "thread.arena" mallctl bug.
835 * 2.1.0 (December 3, 2010)
837 This version incorporates some optimizations that can't quite be considered
841 - Use Linux's mremap(2) for huge object reallocation when possible.
842 - Avoid locking in mallctl*() when possible.
843 - Add the "thread.[de]allocatedp" mallctl's.
844 - Convert the manual page source from roff to DocBook, and generate both roff
848 - Fix a crash due to incorrect bootstrap ordering. This only impacted
849 --enable-debug --enable-dss configurations.
850 - Fix a minor statistics bug for mallctl("swap.avail", ...).
852 * 2.0.1 (October 29, 2010)
855 - Fix a race condition in heap profiling that could cause undefined behavior
856 if "opt.prof_accum" were disabled.
857 - Add missing mutex unlocks for some OOM error paths in the heap profiling
859 - Fix a compilation error for non-C99 builds.
861 * 2.0.0 (October 24, 2010)
863 This version focuses on the experimental *allocm() API, and on improved
864 run-time configuration/introspection. Nonetheless, numerous performance
865 improvements are also included.
868 - Implement the experimental {,r,s,d}allocm() API, which provides a superset
869 of the functionality available via malloc(), calloc(), posix_memalign(),
870 realloc(), malloc_usable_size(), and free(). These functions can be used to
871 allocate/reallocate aligned zeroed memory, ask for optional extra memory
872 during reallocation, prevent object movement during reallocation, etc.
873 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
874 more human-readable, and more flexible. For example:
877 MALLOC_CONF=abort:true,fill:true,stats_print:true
878 - Port to Apple OS X. Sponsored by Mozilla.
879 - Make it possible for the application to control thread-->arena mappings via
880 the "thread.arena" mallctl.
881 - Add compile-time support for all TLS-related functionality via pthreads TSD.
882 This is mainly of interest for OS X, which does not support TLS, but has a
883 TSD implementation with similar performance.
884 - Override memalign() and valloc() if they are provided by the system.
885 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
887 - Make cumulative heap profiling data optional, so that it is possible to
888 limit the amount of memory consumed by heap profiling data structures.
889 - Add per thread allocation counters that can be accessed via the
890 "thread.allocated" and "thread.deallocated" mallctls.
892 Incompatible changes:
893 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
894 - Increase default backtrace depth from 4 to 128 for heap profiling.
895 - Disable interval-based profile dumps by default.
898 - Remove bad assertions in fork handler functions. These assertions could
899 cause aborts for some combinations of configure settings.
900 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
901 - Fix leak context reporting. This bug tended to cause the number of contexts
902 to be underreported (though the reported number of objects and bytes were
904 - Fix a realloc() bug for large in-place growing reallocation. This bug could
905 cause memory corruption, but it was hard to trigger.
906 - Fix an allocation bug for small allocations that could be triggered if
907 multiple threads raced to create a new run of backing pages.
908 - Enhance the heap profiler to trigger samples based on usable size, rather
910 - Fix a heap profiling bug due to sometimes losing track of requested object
911 size for sampled objects.
913 * 1.0.3 (August 12, 2010)
916 - Fix the libunwind-based implementation of stack backtracing (used for heap
917 profiling). This bug could cause zero-length backtraces to be reported.
918 - Add a missing mutex unlock in library initialization code. If multiple
919 threads raced to initialize malloc, some of them could end up permanently
922 * 1.0.2 (May 11, 2010)
925 - Fix junk filling of large objects, which could cause memory corruption.
926 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
927 memory limits could cause swap file configuration to fail. Contributed by
930 * 1.0.1 (April 14, 2010)
933 - Fix compilation when --enable-fill is specified.
934 - Fix threads-related profiling bugs that affected accuracy and caused memory
935 to be leaked during thread exit.
936 - Fix dirty page purging race conditions that could cause crashes.
937 - Fix crash in tcache flushing code during thread destruction.
939 * 1.0.0 (April 11, 2010)
941 This release focuses on speed and run-time introspection. Numerous
942 algorithmic improvements make this release substantially faster than its
946 - Implement autoconf-based configuration system.
947 - Add mallctl*(), for the purposes of introspection and run-time
949 - Make it possible for the application to manually flush a thread's cache, via
950 the "tcache.flush" mallctl.
951 - Base maximum dirty page count on proportion of active memory.
952 - Compute various additional run-time statistics, including per size class
953 statistics for large objects.
954 - Expose malloc_stats_print(), which can be called repeatedly by the
956 - Simplify the malloc_message() signature to only take one string argument,
957 and incorporate an opaque data pointer argument for use by the application
958 in combination with malloc_stats_print().
959 - Add support for allocation backed by one or more swap files, and allow the
960 application to disable over-commit if swap files are in use.
961 - Implement allocation profiling and leak checking.
964 - Remove the dynamic arena rebalancing code, since thread-specific caching
968 - Modify chunk allocation to work when address space layout randomization
970 - Fix thread cleanup bugs related to TLS destruction.
971 - Handle 0-size allocation requests in posix_memalign().
972 - Fix a chunk leak. The leaked chunks were never touched, so this impacted
973 virtual memory usage, but not physical memory usage.
975 * linux_2008082[78]a (August 27/28, 2008)
977 These snapshot releases are the simple result of incorporating Linux-specific
978 support into the FreeBSD malloc sources.
980 --------------------------------------------------------------------------------
981 vim:filetype=text:textwidth=80