blob: fe62e52533090875487e880f5c3cf4f9fc316a41 [file] [log] [blame]
Jason Evans379f8472010-10-24 16:18:29 -07001Following are change highlights associated with official releases. Important
Jason Evansbe09b812015-07-07 09:33:22 -07002bug fixes are all mentioned, but some internal enhancements are omitted here for
3brevity. Much more detail can be found in the git revision history:
Jason Evans379f8472010-10-24 16:18:29 -07004
Jason Evansb9ec5c92014-02-25 16:43:51 -08005 https://github.com/jemalloc/jemalloc
6
Jason Evans54673fd2015-02-23 22:28:43 -08007* 4.0.0 (XXX) See https://github.com/jemalloc/jemalloc/milestones/4.0.0 for
8 remaining work.
9
10 This version contains many speed and space optimizations, both minor and
11 major. The major themes are generalization, unification, and simplification.
12 Although many of these optimizations cause no visible behavior change, their
13 cumulative effect is substantial.
14
15 New features:
16 - Normalize size class spacing to be consistent across the complete size
17 range. By default there are four size classes per size doubling, but this
18 is now configurable via the --with-lg-size-class-group option. Also add the
19 --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and
20 --with-lg-tiny-min options, which can be used to tweak page and size class
21 settings. Impacts:
22 + Worst case performance for incrementally growing/shrinking reallocation
23 is improved because there are far fewer size classes, and therefore
24 copying happens less often.
25 + Internal fragmentation is limited to 20% for all but the smallest size
26 classes (those less than four times the quantum). (1B + 4 KiB)
27 and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation.
28 + Chunk fragmentation tends to be lower because there are fewer distinct run
29 sizes to pack.
30 - Add support for explicit tcaches. The "tcache.create", "tcache.flush", and
31 "tcache.destroy" mallctls control tcache lifetime and flushing, and the
32 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API
33 control which tcache is used for each operation.
34 - Implement per thread heap profiling, as well as the ability to
35 enable/disable heap profiling on a per thread basis. Add the "prof.reset",
36 "prof.lg_sample", "thread.prof.name", "thread.prof.active",
37 "opt.prof_thread_active_init", "prof.thread_active_init", and
38 "thread.prof.active" mallctls.
39 - Add support for per arena application-specified chunk allocators, configured
Jason Evans562d2662015-03-24 16:36:12 -070040 via the "arena<i>.chunk.alloc", "arena<i>.chunk.dalloc", and
41 "arena.<i>.chunk.purge" mallctls.
Jason Evans54673fd2015-02-23 22:28:43 -080042 - Refactor huge allocation to be managed by arenas, so that arenas now
43 function as general purpose independent allocators. This is important in
44 the context of user-specified chunk allocators, aside from the scalability
45 benefits. Related new statistics:
46 + The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc",
47 "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests"
48 mallctls provide high level per arena huge allocation statistics.
Qinfan Wu89750352015-04-21 16:57:42 -070049 + The "arenas.nhchunks", "arenas.hchunk.<i>.size",
Jason Evans54673fd2015-02-23 22:28:43 -080050 "stats.arenas.<i>.hchunks.<j>.nmalloc",
51 "stats.arenas.<i>.hchunks.<j>.ndalloc",
52 "stats.arenas.<i>.hchunks.<j>.nrequests", and
53 "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class
54 statistics.
55 - Add the 'util' column to malloc_stats_print() output, which reports the
56 proportion of available regions that are currently in use for each small
57 size class.
58 - Add "alloc" and "free" modes for for junk filling (see the "opt.junk"
59 mallctl), so that it is possible to separately enable junk filling for
60 allocation versus deallocation.
61 - Add the jemalloc-config script, which provides information about how
62 jemalloc was configured, and how to integrate it into application builds.
63 - Add metadata statistics, which are accessible via the "stats.metadata",
64 "stats.arenas.<i>.metadata.mapped", and
65 "stats.arenas.<i>.metadata.allocated" mallctls.
Jason Evans4acd75a2015-03-23 17:25:57 -070066 - Add the "stats.resident" mallctl, which reports the upper limit of
67 physically resident memory mapped by the allocator.
Jason Evans562d2662015-03-24 16:36:12 -070068 - Add per arena control over unused dirty page purging, via the
69 "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and
70 "stats.arenas.<i>.lg_dirty_mult" mallctls.
Jason Evans54673fd2015-02-23 22:28:43 -080071 - Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump
72 feature on/off during program execution.
73 - Add sdallocx(), which implements sized deallocation. The primary
74 optimization over dallocx() is the removal of a metadata read, which often
75 suffers an L1 cache miss.
76 - Add missing header includes in jemalloc/jemalloc.h, so that applications
77 only have to #include <jemalloc/jemalloc.h>.
78 - Add support for additional platforms:
79 + Bitrig
80 + Cygwin
81 + DragonFlyBSD
82 + iOS
83 + OpenBSD
84 + OpenRISC/or1k
85
86 Optimizations:
87 - Switch run and chunk allocation from first-best-fit (among best-fit
88 candidates, choose the lowest in memory) to first-fit (among all candidates,
89 choose the lowest in memory). This tends to reduce chunk and virtual memory
90 fragmentation, respectively.
91 - Maintain dirty runs in per arena LRUs rather than in per arena trees of
92 dirty-run-containing chunks. In practice this change significantly reduces
93 dirty page purging volume.
94 - Integrate whole chunks into the unused dirty page purging machinery. This
95 reduces the cost of repeated huge allocation/deallocation, because it
96 effectively introduces a cache of chunks.
97 - Split the arena chunk map into two separate arrays, in order to increase
98 cache locality for the frequently accessed bits.
99 - Move small run metadata out of runs, into arena chunk headers. This reduces
100 run fragmentation, smaller runs reduce external fragmentation for small size
101 classes, and packed (less uniformly aligned) metadata layout improves CPU
102 cache set distribution.
Jason Evans8a03cf02015-05-04 09:58:36 -0700103 - Randomly distribute large allocation base pointer alignment relative to page
104 boundaries in order to more uniformly utilize CPU cache sets. This can be
Jason Evansf2bc8522015-07-17 16:38:25 -0700105 disabled via the --disable-cache-oblivious configure option, and queried via
106 the "config.cache_oblivious" mallctl.
Jason Evans54673fd2015-02-23 22:28:43 -0800107 - Micro-optimize the fast paths for the public API functions.
108 - Refactor thread-specific data to reside in a single structure. This assures
109 that only a single TLS read is necessary per call into the public API.
110 - Implement in-place huge allocation growing and shrinking.
111 - Refactor rtree (radix tree for chunk lookups) to be lock-free, and make
112 additional optimizations that reduce maximum lookup depth to one or two
113 levels. This resolves what was a concurrency bottleneck for per arena huge
114 allocation, because a global data structure is critical for determining
115 which arenas own which huge allocations.
116
117 Incompatible changes:
118 - Replace --enable-cc-silence with --disable-cc-silence to suppress spurious
119 warnings by default.
120 - Assure that the constness of malloc_usable_size()'s return type matches that
121 of the system implementation.
122 - Change the heap profile dump format to support per thread heap profiling,
Jason Evans70417202015-05-01 12:31:12 -0700123 rename pprof to jeprof, and enhance it with the --thread=<n> option. As a
124 result, the bundled jeprof must now be used rather than the upstream
125 (gperftools) pprof.
Jason Evans54673fd2015-02-23 22:28:43 -0800126 - Disable "opt.prof_final" by default, in order to avoid atexit(3), which can
127 internally deadlock on some platforms.
128 - Change the "arenas.nlruns" mallctl type from size_t to unsigned.
129 - Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with
130 "stats.arenas.<i>.bins.<j>.curregs".
131 - Ignore MALLOC_CONF in set{uid,gid,cap} binaries.
132 - Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the
133 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage.
134
135 Removed features:
136 - Remove the *allocm() API, which is superseded by the *allocx() API.
137 - Remove the --enable-dss options, and make dss non-optional on all platforms
138 which support sbrk(2).
139 - Remove the "arenas.purge" mallctl, which was obsoleted by the
140 "arena.<i>.purge" mallctl in 3.1.0.
141 - Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically
142 detects whether it is running inside Valgrind.
143 - Remove the "stats.huge.allocated", "stats.huge.nmalloc", and
144 "stats.huge.ndalloc" mallctls.
145 - Remove the --enable-mremap option.
Jason Evans54673fd2015-02-23 22:28:43 -0800146 - Remove the "stats.chunks.current", "stats.chunks.total", and
147 "stats.chunks.high" mallctls.
148
149 Bug fixes:
150 - Fix the cactive statistic to decrease (rather than increase) when active
151 memory decreases. This regression was first released in 3.5.0.
152 - Fix OOM handling in memalign() and valloc(). A variant of this bug existed
153 in all releases since 2.0.0, which introduced these functions.
Jason Evans32dca112015-07-09 11:34:13 -0700154 - Fix an OOM-related regression in arena_tcache_fill_small(), which could
155 cause cache corruption on OOM. This regression was present in all releases
156 from 2.2.0 through 3.6.0.
Jason Evans241abc62015-06-23 18:47:07 -0700157 - Fix size class overflow handling for malloc(), posix_memalign(), memalign(),
158 calloc(), and realloc() when profiling is enabled.
Jason Evans54673fd2015-02-23 22:28:43 -0800159 - Fix the "arena.<i>.dss" mallctl to return an error if "primary" or
160 "secondary" precedence is specified, but sbrk(2) is not supported.
161 - Fix fallback lg_floor() implementations to handle extremely large inputs.
162 - Ensure the default purgeable zone is after the default zone on OS X.
163 - Fix latent bugs in atomic_*().
164 - Fix the "arena.<i>.dss" mallctl to handle read-only calls.
165 - Fix tls_model configuration to enable the initial-exec model when possible.
166 - Mark malloc_conf as a weak symbol so that the application can override it.
167 - Correctly detect glibc's adaptive pthread mutexes.
168 - Fix the --without-export configure option.
169
Jason Evansff536312014-03-31 09:23:10 -0700170* 3.6.0 (March 31, 2014)
171
172 This version contains a critical bug fix for a regression present in 3.5.0 and
173 3.5.1.
174
175 Bug fixes:
176 - Fix a regression in arena_chunk_alloc() that caused crashes during
177 small/large allocation if chunk allocation failed. In the absence of this
178 bug, chunk allocation failure would result in allocation failure, e.g. NULL
179 return from malloc(). This regression was introduced in 3.5.0.
180 - Fix backtracing for gcc intrinsics-based backtracing by specifying
181 -fno-omit-frame-pointer to gcc. Note that the application (and all the
182 libraries it links to) must also be compiled with this option for
183 backtracing to be reliable.
184 - Use dss allocation precedence for huge allocations as well as small/large
185 allocations.
Jason Evans54673fd2015-02-23 22:28:43 -0800186 - Fix test assertion failure message formatting. This bug did not manifest on
Jason Evansff536312014-03-31 09:23:10 -0700187 x86_64 systems because of implementation subtleties in va_list.
188 - Fix inconsequential test failures for hash and SFMT code.
189
190 New features:
191 - Support heap profiling on FreeBSD. This feature depends on the proc
192 filesystem being mounted during heap profile dumping.
193
Jason Evansb9ec5c92014-02-25 16:43:51 -0800194* 3.5.1 (February 25, 2014)
195
196 This version primarily addresses minor bugs in test code.
197
198 Bug fixes:
199 - Configure Solaris/Illumos to use MADV_FREE.
200 - Fix junk filling for mremap(2)-based huge reallocation. This is only
201 relevant if configuring with the --enable-mremap option specified.
202 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the
203 compiler.
204 - Add a configure test for SSE2 rather than assuming it is usable on i686
205 systems. This fixes test compilation errors, especially on 32-bit Linux
206 systems.
207 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit
208 test.
209 - Fix/remove flawed alignment-related overflow tests.
210 - Prevent compiler optimizations that could change backtraces in the
211 prof_accum unit test.
Jason Evans379f8472010-10-24 16:18:29 -0700212
Jason Evans798a4812014-01-22 11:09:50 -0800213* 3.5.0 (January 22, 2014)
Jason Evans86abd0d2013-11-30 15:25:42 -0800214
Jason Evans264dfd32014-01-17 17:01:23 -0800215 This version focuses on refactoring and automated testing, though it also
216 includes some non-trivial heap profiling optimizations not mentioned below.
217
218 New features:
219 - Add the *allocx() API, which is a successor to the experimental *allocm()
220 API. The *allocx() functions are slightly simpler to use because they have
221 fewer parameters, they directly return the results of primary interest, and
222 mallocx()/rallocx() avoid the strict aliasing pitfall that
Jason Evans9c8baec2014-01-22 13:08:47 -0800223 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is
Jason Evans264dfd32014-01-17 17:01:23 -0800224 slated for removal in the next non-bugfix release.
225 - Add support for LinuxThreads.
226
Jason Evans86abd0d2013-11-30 15:25:42 -0800227 Bug fixes:
Jason Evansd37d5ad2013-12-05 23:01:50 -0800228 - Unless heap profiling is enabled, disable floating point code and don't link
229 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64
230 systems, makes it possible to completely disable floating point register
231 use. Some versions of glibc neglect to save/restore caller-saved floating
232 point registers during dynamic lazy symbol loading, and the symbol loading
233 code uses whatever malloc the application happens to have linked/loaded
234 with, the result being potential floating point register corruption.
Jason Evans264dfd32014-01-17 17:01:23 -0800235 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling
236 backtrace creation in imemalign(). This bug impacted posix_memalign() and
237 aligned_alloc().
238 - Fix a file descriptor leak in a prof_dump_maps() error path.
239 - Fix prof_dump() to close the dump file descriptor for all relevant error
240 paths.
241 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for
242 allocation, not just deallocation.
243 - Fix a data race for large allocation stats counters.
244 - Fix a potential infinite loop during thread exit. This bug occurred on
245 Solaris, and could affect other platforms with similar pthreads TSD
246 implementations.
247 - Don't junk-fill reallocations unless usable size changes. This fixes a
248 violation of the *allocx()/*allocm() semantics.
249 - Fix growing large reallocation to junk fill new space.
250 - Fix huge deallocation to junk fill when munmap is disabled.
Jason Evans86abd0d2013-11-30 15:25:42 -0800251 - Change the default private namespace prefix from empty to je_, and change
252 --with-private-namespace-prefix so that it prepends an additional prefix
253 rather than replacing je_. This reduces the likelihood of applications
254 which statically link jemalloc experiencing symbol name collisions.
Jason Evans264dfd32014-01-17 17:01:23 -0800255 - Add missing private namespace mangling (relevant when
256 --with-private-namespace is specified).
257 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as
258 static even for debug builds.
259 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice
260 this error path is never executed.
261 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These
262 bugs had no impact except for malformed inputs.
263 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by
264 existing calls, so they had no impact.
Jason Evans86abd0d2013-11-30 15:25:42 -0800265
Jason Evans0f7ba3f2013-10-20 19:40:09 -0700266* 3.4.1 (October 20, 2013)
Jason Evansff08ef72013-10-19 21:41:10 -0700267
268 Bug fixes:
Jason Evans7b651802013-10-20 14:09:54 -0700269 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption
270 of internal data structures and subsequent crashes.
Jason Evansdda90f52013-10-19 23:48:40 -0700271 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of
272 uninitialized memory in:
273 + arena chunk headers
274 + internal zero-initialized data structures (relevant to tcache and prof
275 code)
Jason Evansff08ef72013-10-19 21:41:10 -0700276 - Preserve errno during the first allocation. A readlink(2) call during
277 initialization fails unless /etc/malloc.conf exists, so errno was typically
278 set during the first allocation prior to this fix.
279 - Fix compilation warnings reported by gcc 4.8.1.
280
Jason Evans765cc2b2013-06-02 20:58:00 -0700281* 3.4.0 (June 2, 2013)
282
283 This version is essentially a small bugfix release, but the addition of
284 aarch64 support requires that the minor version be incremented.
285
286 Bug fixes:
287 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were
288 typically triggered by multiple threads concurrently deallocating huge
289 objects.
290
291 New features:
292 - Add support for the aarch64 architecture.
293
Jason Evans22988352013-03-06 11:11:17 -0800294* 3.3.1 (March 6, 2013)
295
296 This version fixes bugs that are typically encountered only when utilizing
297 custom run-time options.
Jason Evansbbe29d32013-01-30 15:03:11 -0800298
299 Bug fixes:
Jason Evans88c222c2013-02-06 11:59:30 -0800300 - Fix a locking order bug that could cause deadlock during fork if heap
301 profiling were enabled.
Jason Evansa7a28c32013-01-31 16:53:58 -0800302 - Fix a chunk recycling bug that could cause the allocator to lose track of
Jason Evans765cc2b2013-06-02 20:58:00 -0700303 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause
Jason Evansa7a28c32013-01-31 16:53:58 -0800304 corruption if allocating via sbrk(2) (unlikely unless running with the
305 "dss:primary" option specified). This was completely harmless on Linux
306 unless using mlockall(2) (and unlikely even then, unless the
307 --disable-munmap configure option or the "dss:primary" option was
308 specified). This regression was introduced in 3.1.0 by the
309 mlockall(2)/madvise(2) interaction fix.
Jason Evansbbe29d32013-01-30 15:03:11 -0800310 - Fix TLS-related memory corruption that could occur during thread exit if the
311 thread never allocated memory. Only the quarantine and prof facilities were
312 susceptible.
Jason Evansd0e942e2013-01-31 14:42:41 -0800313 - Fix two quarantine bugs:
314 + Internal reallocation of the quarantined object array leaked the old
315 array.
316 + Reallocation failure for internal reallocation of the quarantined object
317 array (very unlikely) resulted in memory corruption.
Jason Evans06912752013-01-31 17:02:53 -0800318 - Fix Valgrind integration to annotate all internally allocated memory in a
319 way that keeps Valgrind happy about internal data structure access.
Jason Evans22988352013-03-06 11:11:17 -0800320 - Fix building for s390 systems.
Jason Evansbbe29d32013-01-30 15:03:11 -0800321
Jason Evansb5681fb2013-01-22 22:45:09 -0800322* 3.3.0 (January 23, 2013)
323
324 This version includes a few minor performance improvements in addition to the
325 listed new features and bug fixes.
326
327 New features:
328 - Add clipping support to lg_chunk option processing.
329 - Add the --enable-ivsalloc option.
330 - Add the --without-export option.
331 - Add the --disable-zone-allocator option.
Jason Evans6eb84fb2012-11-29 22:13:04 -0800332
333 Bug fixes:
334 - Fix "arenas.extend" mallctl to output the number of arenas.
Jason Evansa33488d2013-10-03 14:38:39 -0700335 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory
Jason Evans12711852012-12-12 10:12:18 -0800336 is undefined.
Jason Evansb5681fb2013-01-22 22:45:09 -0800337 - Fix build break on FreeBSD related to alloca.h.
Jason Evans6eb84fb2012-11-29 22:13:04 -0800338
Jason Evans556ddc72012-11-07 15:16:29 -0800339* 3.2.0 (November 9, 2012)
340
341 In addition to a couple of bug fixes, this version modifies page run
342 allocation and dirty page purging algorithms in order to better control
343 page-level virtual memory fragmentation.
Jason Evans12efefb2012-10-16 22:06:56 -0700344
Jason Evanse3d13062012-10-30 15:42:37 -0700345 Incompatible changes:
Jason Evans556ddc72012-11-07 15:16:29 -0800346 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
Jason Evanse3d13062012-10-30 15:42:37 -0700347
Jason Evans12efefb2012-10-16 22:06:56 -0700348 Bug fixes:
349 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
350 after primary dss allocation fails.
Jason Evanse3d13062012-10-30 15:42:37 -0700351 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced
352 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
Jason Evans12efefb2012-10-16 22:06:56 -0700353
Jason Evans2b592b02012-10-16 10:12:40 -0700354* 3.1.0 (October 16, 2012)
Jason Evans3860eac2012-05-15 13:53:21 -0700355
Jason Evans781fe752012-05-15 14:48:14 -0700356 New features:
357 - Auto-detect whether running inside Valgrind, thus removing the need to
358 manually specify MALLOC_CONF=valgrind:true.
Jason Evans2b592b02012-10-16 10:12:40 -0700359 - Add the "arenas.extend" mallctl, which allows applications to create
360 manually managed arenas.
361 - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
362 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
363 which provide control over dss/mmap precedence.
364 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
365 - Define LG_QUANTUM for hppa.
Jason Evans781fe752012-05-15 14:48:14 -0700366
Jason Evans174b70e2012-05-15 23:31:53 -0700367 Incompatible changes:
368 - Disable tcache by default if running inside Valgrind, in order to avoid
369 making unallocated objects appear reachable to Valgrind.
Jason Evans2b592b02012-10-16 10:12:40 -0700370 - Drop const from malloc_usable_size() argument on Linux.
Jason Evans174b70e2012-05-15 23:31:53 -0700371
Jason Evans3860eac2012-05-15 13:53:21 -0700372 Bug fixes:
373 - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
Jason Evans5c710ce2012-05-23 16:09:22 -0700374 - Remove const from __*_hook variable declarations, so that glibc can modify
375 them during process forking.
Jason Evans2b592b02012-10-16 10:12:40 -0700376 - Fix mlockall(2)/madvise(2) interaction.
377 - Fix fork(2)-related deadlocks.
378 - Fix error return value for "thread.tcache.enabled" mallctl.
Jason Evans3860eac2012-05-15 13:53:21 -0700379
Jason Evanscbb71ca2012-05-11 17:00:20 -0700380* 3.0.0 (May 11, 2012)
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700381
382 Although this version adds some major new features, the primary focus is on
383 internal code cleanup that facilitates maintainability and portability, most
384 of which is not reflected in the ChangeLog. This is the first release to
385 incorporate substantial contributions from numerous other developers, and the
386 result is a more broadly useful allocator (see the git revision history for
387 contribution details). Note that the license has been unified, thanks to
388 Facebook granting a license under the same terms as the other copyright
389 holders (see COPYING).
390
391 New features:
392 - Implement Valgrind support, redzones, and quarantine.
Jason Evans079687b2012-04-23 12:49:23 -0700393 - Add support for additional platforms:
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700394 + FreeBSD
395 + Mac OS X Lion
Jason Evans40f514f2012-04-22 16:21:06 -0700396 + MinGW
Jason Evanscbb71ca2012-05-11 17:00:20 -0700397 + Windows (no support yet for replacing the system malloc)
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700398 - Add support for additional architectures:
399 + MIPS
400 + SH4
401 + Tilera
402 - Add support for cross compiling.
403 - Add nallocm(), which rounds a request size up to the nearest size class
404 without actually allocating.
405 - Implement aligned_alloc() (blame C11).
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700406 - Add the "thread.tcache.enabled" mallctl.
Jason Evans0b25fe72012-04-17 16:39:33 -0700407 - Add the "opt.prof_final" mallctl.
Jason Evans25a000e2012-04-17 15:49:30 -0700408 - Update pprof (from gperftools 2.0).
Jason Evanscbb71ca2012-05-11 17:00:20 -0700409 - Add the --with-mangling option.
410 - Add the --disable-experimental option.
411 - Add the --disable-munmap option, and make it the default on Linux.
412 - Add the --enable-mremap option, which disables use of mremap(2) by default.
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700413
414 Incompatible changes:
415 - Enable stats by default.
416 - Enable fill by default.
417 - Disable lazy locking by default.
418 - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
419 - Rename the "arenas.pagesize" mallctl to "arenas.page".
Jason Evans0b25fe72012-04-17 16:39:33 -0700420 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
421 - Change the "opt.prof_accum" default from true to false.
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700422
423 Removed features:
424 - Remove the swap feature, including the "config.swap", "swap.avail",
425 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
426 - Remove highruns statistics, including the
427 "stats.arenas.<i>.bins.<j>.highruns" and
428 "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
429 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
430 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
431 "arenas.[tqcs]bins" mallctls.
432 - Remove the "arenas.chunksize" mallctl.
433 - Remove the "opt.lg_prof_tcmax" option.
434 - Remove the "opt.lg_prof_bt_max" option.
435 - Remove the "opt.lg_tcache_gc_sweep" option.
436 - Remove the --disable-tiny option, including the "config.tiny" mallctl.
437 - Remove the --enable-dynamic-page-shift configure option.
438 - Remove the --enable-sysv configure option.
439
440 Bug fixes:
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700441 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
442 invalid statistics and crashes.
Jason Evans079687b2012-04-23 12:49:23 -0700443 - Work around TLS deallocation via free() on Linux. This bug could cause
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700444 write-after-free memory corruption.
Jason Evans52386b22012-04-22 16:00:11 -0700445 - Fix a potential deadlock that could occur during interval- and
446 growth-triggered heap profile dumps.
Jason Evansd8ceef62012-05-10 20:59:39 -0700447 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
Jason Evans8f0e0eb2012-04-21 13:33:48 -0700448 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could
449 cause memory corruption and crashes with --enable-dss specified.
Jason Evans52386b22012-04-22 16:00:11 -0700450 - Fix fork-related bugs that could cause deadlock in children between fork
451 and exec.
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700452 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
453 - Fix realloc(p, 0) to act like free(p).
454 - Do not enforce minimum alignment in memalign().
455 - Check for NULL pointer in malloc_usable_size().
Jason Evans52386b22012-04-22 16:00:11 -0700456 - Fix an off-by-one heap profile statistics bug that could be observed in
457 interval- and growth-triggered heap profiles.
Jason Evans6b9ed672012-04-25 13:12:46 -0700458 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
459 is 0.
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700460 - Fix bin->runcur management to fix a layout policy bug. This bug did not
461 affect correctness.
462 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
463 initialized than necessary.
464 - Add missing "opt.lg_tcache_max" mallctl implementation.
465 - Use glibc allocator hooks to make mixed allocator usage less likely.
466 - Fix build issues for --disable-tcache.
Jason Evans918d6e22012-04-20 13:42:21 -0700467 - Don't mangle pthread_create() when --with-private-namespace is specified.
Jason Evans9ef7f5d2012-04-16 18:16:48 -0700468
Jason Evansb3bd8852011-11-14 17:12:45 -0800469* 2.2.5 (November 14, 2011)
470
471 Bug fixes:
472 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that
473 could cause memory corruption and/or crashes.
474 - Fix huge_ralloc() to maintain chunk statistics.
475 - Fix malloc_stats_print(..., "a") output.
476
Jason Evansca9ee1a2011-11-05 21:46:23 -0700477* 2.2.4 (November 5, 2011)
478
479 Bug fixes:
480 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as
481 well as for --disable-tls builds in earlier releases.
482 - Do not assume a 4 KiB page size in test/rallocm.c.
483
Jason Evansc67e4fd2011-08-31 15:19:13 -0700484* 2.2.3 (August 31, 2011)
485
486 This version fixes numerous bugs related to heap profiling.
487
488 Bug fixes:
489 - Fix a prof-related race condition. This bug could cause memory corruption,
490 but only occurred in non-default configurations (prof_accum:false).
491 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
492 excluded from backtraces).
493 - Fix a prof-related bug in realloc() (only triggered by OOM errors).
494 - Fix prof-related bugs in allocm() and rallocm().
495 - Fix prof_tdata_cleanup() for --disable-tls builds.
496 - Fix a relative include path, to fix objdir builds.
497
Jason Evans4c484812011-07-30 16:59:13 -0700498* 2.2.2 (July 30, 2011)
499
500 Bug fixes:
501 - Fix a build error for --disable-tcache.
502 - Fix assertions in arena_purge() (for real this time).
503 - Add the --with-private-namespace option. This is a workaround for symbol
504 conflicts that can inadvertently arise when using static libraries.
505
Jason Evans7d9ebea2011-03-30 15:01:08 -0700506* 2.2.1 (March 30, 2011)
507
508 Bug fixes:
509 - Implement atomic operations for x86/x64. This fixes compilation failures
510 for versions of gcc that are still in wide use.
511 - Fix an assertion in arena_purge().
512
Jason Evans4bcd9872011-03-22 15:30:22 -0700513* 2.2.0 (March 22, 2011)
514
515 This version incorporates several improvements to algorithms and data
516 structures that tend to reduce fragmentation and increase speed.
517
518 New features:
519 - Add the "stats.cactive" mallctl.
520 - Update pprof (from google-perftools 1.7).
521 - Improve backtracing-related configuration logic, and add the
522 --disable-prof-libgcc option.
523
524 Bug fixes:
525 - Change default symbol visibility from "internal", to "hidden", which
526 decreases the overhead of library-internal function calls.
527 - Fix symbol visibility so that it is also set on OS X.
528 - Fix a build dependency regression caused by the introduction of the .pic.o
529 suffix for PIC object files.
530 - Add missing checks for mutex initialization failures.
531 - Don't use libgcc-based backtracing except on x64, where it is known to work.
532 - Fix deadlocks on OS X that were due to memory allocation in
533 pthread_mutex_lock().
534 - Heap profiling-specific fixes:
535 + Fix memory corruption due to integer overflow in small region index
536 computation, when using a small enough sample interval that profiling
537 context pointers are stored in small run headers.
538 + Fix a bootstrap ordering bug that only occurred with TLS disabled.
539 + Fix a rallocm() rsize bug.
540 + Fix error detection bugs for aligned memory allocation.
541
Jason Evans0e4d0d12011-03-14 16:41:03 -0700542* 2.1.3 (March 14, 2011)
543
544 Bug fixes:
545 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
546 for OS X in 2.1.2).
547 - Fix a "thread.arena" mallctl bug.
548 - Fix a thread cache stats merging bug.
549
je6e56e5e2011-03-02 11:23:41 -0800550* 2.1.2 (March 2, 2011)
551
552 Bug fixes:
553 - Fix "thread.{de,}allocatedp" mallctl for OS X.
554 - Add missing jemalloc.a to build system.
555
Jason Evans63692862011-02-07 22:48:35 -0800556* 2.1.1 (January 31, 2011)
Jason Evansada55b22011-01-31 20:08:56 -0800557
Jason Evans63692862011-02-07 22:48:35 -0800558 Bug fixes:
Jason Evansada55b22011-01-31 20:08:56 -0800559 - Fix aligned huge reallocation (affected allocm()).
560 - Fix the ALLOCM_LG_ALIGN macro definition.
561 - Fix a heap dumping deadlock.
562 - Fix a "thread.arena" mallctl bug.
563
Jason Evans63692862011-02-07 22:48:35 -0800564* 2.1.0 (December 3, 2010)
Jason Evans0e8d3d22010-12-03 17:02:16 -0800565
566 This version incorporates some optimizations that can't quite be considered
567 bug fixes.
568
569 New features:
570 - Use Linux's mremap(2) for huge object reallocation when possible.
571 - Avoid locking in mallctl*() when possible.
572 - Add the "thread.[de]allocatedp" mallctl's.
573 - Convert the manual page source from roff to DocBook, and generate both roff
574 and HTML manuals.
575
576 Bug fixes:
577 - Fix a crash due to incorrect bootstrap ordering. This only impacted
578 --enable-debug --enable-dss configurations.
579 - Fix a minor statistics bug for mallctl("swap.avail", ...).
580
Jason Evans63692862011-02-07 22:48:35 -0800581* 2.0.1 (October 29, 2010)
Jason Evans53806fe2010-10-29 20:16:39 -0700582
583 Bug fixes:
584 - Fix a race condition in heap profiling that could cause undefined behavior
Jason Evansada55b22011-01-31 20:08:56 -0800585 if "opt.prof_accum" were disabled.
Jason Evans53806fe2010-10-29 20:16:39 -0700586 - Add missing mutex unlocks for some OOM error paths in the heap profiling
587 code.
588 - Fix a compilation error for non-C99 builds.
589
Jason Evans63692862011-02-07 22:48:35 -0800590* 2.0.0 (October 24, 2010)
Jason Evans379f8472010-10-24 16:18:29 -0700591
592 This version focuses on the experimental *allocm() API, and on improved
593 run-time configuration/introspection. Nonetheless, numerous performance
594 improvements are also included.
595
596 New features:
Jason Evansb059a532010-10-24 16:54:40 -0700597 - Implement the experimental {,r,s,d}allocm() API, which provides a superset
598 of the functionality available via malloc(), calloc(), posix_memalign(),
599 realloc(), malloc_usable_size(), and free(). These functions can be used to
600 allocate/reallocate aligned zeroed memory, ask for optional extra memory
601 during reallocation, prevent object movement during reallocation, etc.
602 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
603 more human-readable, and more flexible. For example:
604 JEMALLOC_OPTIONS=AJP
605 is now:
606 MALLOC_CONF=abort:true,fill:true,stats_print:true
607 - Port to Apple OS X. Sponsored by Mozilla.
608 - Make it possible for the application to control thread-->arena mappings via
609 the "thread.arena" mallctl.
610 - Add compile-time support for all TLS-related functionality via pthreads TSD.
611 This is mainly of interest for OS X, which does not support TLS, but has a
612 TSD implementation with similar performance.
613 - Override memalign() and valloc() if they are provided by the system.
614 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
615 dirty unused pages.
616 - Make cumulative heap profiling data optional, so that it is possible to
617 limit the amount of memory consumed by heap profiling data structures.
618 - Add per thread allocation counters that can be accessed via the
619 "thread.allocated" and "thread.deallocated" mallctls.
Jason Evans379f8472010-10-24 16:18:29 -0700620
621 Incompatible changes:
Jason Evansb059a532010-10-24 16:54:40 -0700622 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
623 - Increase default backtrace depth from 4 to 128 for heap profiling.
624 - Disable interval-based profile dumps by default.
Jason Evans379f8472010-10-24 16:18:29 -0700625
626 Bug fixes:
627 - Remove bad assertions in fork handler functions. These assertions could
628 cause aborts for some combinations of configure settings.
629 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
630 - Fix leak context reporting. This bug tended to cause the number of contexts
631 to be underreported (though the reported number of objects and bytes were
632 correct).
633 - Fix a realloc() bug for large in-place growing reallocation. This bug could
634 cause memory corruption, but it was hard to trigger.
635 - Fix an allocation bug for small allocations that could be triggered if
636 multiple threads raced to create a new run of backing pages.
637 - Enhance the heap profiler to trigger samples based on usable size, rather
638 than request size.
639 - Fix a heap profiling bug due to sometimes losing track of requested object
640 size for sampled objects.
641
Jason Evans63692862011-02-07 22:48:35 -0800642* 1.0.3 (August 12, 2010)
Jason Evans379f8472010-10-24 16:18:29 -0700643
644 Bug fixes:
645 - Fix the libunwind-based implementation of stack backtracing (used for heap
646 profiling). This bug could cause zero-length backtraces to be reported.
647 - Add a missing mutex unlock in library initialization code. If multiple
648 threads raced to initialize malloc, some of them could end up permanently
649 blocked.
650
Jason Evans63692862011-02-07 22:48:35 -0800651* 1.0.2 (May 11, 2010)
Jason Evans379f8472010-10-24 16:18:29 -0700652
653 Bug fixes:
654 - Fix junk filling of large objects, which could cause memory corruption.
655 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
656 memory limits could cause swap file configuration to fail. Contributed by
657 Jordan DeLong.
658
Jason Evans63692862011-02-07 22:48:35 -0800659* 1.0.1 (April 14, 2010)
Jason Evans379f8472010-10-24 16:18:29 -0700660
661 Bug fixes:
662 - Fix compilation when --enable-fill is specified.
663 - Fix threads-related profiling bugs that affected accuracy and caused memory
664 to be leaked during thread exit.
665 - Fix dirty page purging race conditions that could cause crashes.
666 - Fix crash in tcache flushing code during thread destruction.
667
Jason Evans63692862011-02-07 22:48:35 -0800668* 1.0.0 (April 11, 2010)
Jason Evans379f8472010-10-24 16:18:29 -0700669
670 This release focuses on speed and run-time introspection. Numerous
671 algorithmic improvements make this release substantially faster than its
672 predecessors.
673
674 New features:
675 - Implement autoconf-based configuration system.
676 - Add mallctl*(), for the purposes of introspection and run-time
677 configuration.
678 - Make it possible for the application to manually flush a thread's cache, via
679 the "tcache.flush" mallctl.
680 - Base maximum dirty page count on proportion of active memory.
charsyamad6800f2015-07-04 01:06:06 +0900681 - Compute various additional run-time statistics, including per size class
Jason Evans379f8472010-10-24 16:18:29 -0700682 statistics for large objects.
683 - Expose malloc_stats_print(), which can be called repeatedly by the
684 application.
685 - Simplify the malloc_message() signature to only take one string argument,
686 and incorporate an opaque data pointer argument for use by the application
687 in combination with malloc_stats_print().
688 - Add support for allocation backed by one or more swap files, and allow the
689 application to disable over-commit if swap files are in use.
690 - Implement allocation profiling and leak checking.
691
692 Removed features:
693 - Remove the dynamic arena rebalancing code, since thread-specific caching
694 reduces its utility.
695
696 Bug fixes:
697 - Modify chunk allocation to work when address space layout randomization
698 (ASLR) is in use.
699 - Fix thread cleanup bugs related to TLS destruction.
700 - Handle 0-size allocation requests in posix_memalign().
701 - Fix a chunk leak. The leaked chunks were never touched, so this impacted
702 virtual memory usage, but not physical memory usage.
703
Jason Evans63692862011-02-07 22:48:35 -0800704* linux_2008082[78]a (August 27/28, 2008)
Jason Evans379f8472010-10-24 16:18:29 -0700705
706 These snapshot releases are the simple result of incorporating Linux-specific
707 support into the FreeBSD malloc sources.
708
709--------------------------------------------------------------------------------
710vim:filetype=text:textwidth=80