1Following are change highlights associated with official releases. Important
2bug fixes are all mentioned, but internal enhancements are omitted here for
3brevity (even though they are more fun to write about). Much more detail can be
4found in the git revision history:
5
6 https://github.com/jemalloc/jemalloc
7
8* 3.5.1 (February 25, 2014)
9
10 This version primarily addresses minor bugs in test code.
11
12 Bug fixes:
13 - Configure Solaris/Illumos to use MADV_FREE.
14 - Fix junk filling for mremap(2)-based huge reallocation. This is only
15 relevant if configuring with the --enable-mremap option specified.
16 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the
17 compiler.
18 - Add a configure test for SSE2 rather than assuming it is usable on i686
19 systems. This fixes test compilation errors, especially on 32-bit Linux
20 systems.
21 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit
22 test.
23 - Fix/remove flawed alignment-related overflow tests.
24 - Prevent compiler optimizations that could change backtraces in the
25 prof_accum unit test.
26
27* 3.5.0 (January 22, 2014)
28
29 This version focuses on refactoring and automated testing, though it also
30 includes some non-trivial heap profiling optimizations not mentioned below.
31
32 New features:
33 - Add the *allocx() API, which is a successor to the experimental *allocm()
34 API. The *allocx() functions are slightly simpler to use because they have
35 fewer parameters, they directly return the results of primary interest, and
36 mallocx()/rallocx() avoid the strict aliasing pitfall that
37 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is
38 slated for removal in the next non-bugfix release.
39 - Add support for LinuxThreads.
40
41 Bug fixes:
42 - Unless heap profiling is enabled, disable floating point code and don't link
43 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64
44 systems, makes it possible to completely disable floating point register
45 use. Some versions of glibc neglect to save/restore caller-saved floating
46 point registers during dynamic lazy symbol loading, and the symbol loading
47 code uses whatever malloc the application happens to have linked/loaded
48 with, the result being potential floating point register corruption.
49 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling
50 backtrace creation in imemalign(). This bug impacted posix_memalign() and
51 aligned_alloc().
52 - Fix a file descriptor leak in a prof_dump_maps() error path.
53 - Fix prof_dump() to close the dump file descriptor for all relevant error
54 paths.
55 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for
56 allocation, not just deallocation.
57 - Fix a data race for large allocation stats counters.
58 - Fix a potential infinite loop during thread exit. This bug occurred on
59 Solaris, and could affect other platforms with similar pthreads TSD
60 implementations.
61 - Don't junk-fill reallocations unless usable size changes. This fixes a
62 violation of the *allocx()/*allocm() semantics.
63 - Fix growing large reallocation to junk fill new space.
64 - Fix huge deallocation to junk fill when munmap is disabled.
65 - Change the default private namespace prefix from empty to je_, and change
66 --with-private-namespace-prefix so that it prepends an additional prefix
67 rather than replacing je_. This reduces the likelihood of applications
68 which statically link jemalloc experiencing symbol name collisions.
69 - Add missing private namespace mangling (relevant when
70 --with-private-namespace is specified).
71 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as
72 static even for debug builds.
73 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice
74 this error path is never executed.
75 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These
76 bugs had no impact except for malformed inputs.
77 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by
78 existing calls, so they had no impact.
79
80* 3.4.1 (October 20, 2013)
81
82 Bug fixes:
83 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption
84 of internal data structures and subsequent crashes.
85 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of
86 uninitialized memory in:
87 + arena chunk headers
88 + internal zero-initialized data structures (relevant to tcache and prof
89 code)
90 - Preserve errno during the first allocation. A readlink(2) call during
91 initialization fails unless /etc/malloc.conf exists, so errno was typically
92 set during the first allocation prior to this fix.
93 - Fix compilation warnings reported by gcc 4.8.1.
94
95* 3.4.0 (June 2, 2013)
96
97 This version is essentially a small bugfix release, but the addition of
98 aarch64 support requires that the minor version be incremented.
99
100 Bug fixes:
101 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were
102 typically triggered by multiple threads concurrently deallocating huge
103 objects.
104
105 New features:
106 - Add support for the aarch64 architecture.
107
108* 3.3.1 (March 6, 2013)
109
110 This version fixes bugs that are typically encountered only when utilizing
111 custom run-time options.
112
113 Bug fixes:
114 - Fix a locking order bug that could cause deadlock during fork if heap
115 profiling were enabled.
116 - Fix a chunk recycling bug that could cause the allocator to lose track of
117 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause
118 corruption if allocating via sbrk(2) (unlikely unless running with the
119 "dss:primary" option specified). This was completely harmless on Linux
120 unless using mlockall(2) (and unlikely even then, unless the
121 --disable-munmap configure option or the "dss:primary" option was
122 specified). This regression was introduced in 3.1.0 by the
123 mlockall(2)/madvise(2) interaction fix.
124 - Fix TLS-related memory corruption that could occur during thread exit if the
125 thread never allocated memory. Only the quarantine and prof facilities were
126 susceptible.
127 - Fix two quarantine bugs:
128 + Internal reallocation of the quarantined object array leaked the old
129 array.
130 + Reallocation failure for internal reallocation of the quarantined object
131 array (very unlikely) resulted in memory corruption.
132 - Fix Valgrind integration to annotate all internally allocated memory in a
133 way that keeps Valgrind happy about internal data structure access.
134 - Fix building for s390 systems.
135
136* 3.3.0 (January 23, 2013)
137
138 This version includes a few minor performance improvements in addition to the
139 listed new features and bug fixes.
140
141 New features:
142 - Add clipping support to lg_chunk option processing.
143 - Add the --enable-ivsalloc option.
144 - Add the --without-export option.
145 - Add the --disable-zone-allocator option.
146
147 Bug fixes:
148 - Fix "arenas.extend" mallctl to output the number of arenas.
149 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory
150 is undefined.
151 - Fix build break on FreeBSD related to alloca.h.
152
153* 3.2.0 (November 9, 2012)
154
155 In addition to a couple of bug fixes, this version modifies page run
156 allocation and dirty page purging algorithms in order to better control
157 page-level virtual memory fragmentation.
158
159 Incompatible changes:
160 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
161
162 Bug fixes:
163 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
164 after primary dss allocation fails.
165 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced
166 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
167
168* 3.1.0 (October 16, 2012)
169
170 New features:
171 - Auto-detect whether running inside Valgrind, thus removing the need to
172 manually specify MALLOC_CONF=valgrind:true.
173 - Add the "arenas.extend" mallctl, which allows applications to create
174 manually managed arenas.
175 - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
176 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
177 which provide control over dss/mmap precedence.
178 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
179 - Define LG_QUANTUM for hppa.
180
181 Incompatible changes:
182 - Disable tcache by default if running inside Valgrind, in order to avoid
183 making unallocated objects appear reachable to Valgrind.
184 - Drop const from malloc_usable_size() argument on Linux.
185
186 Bug fixes:
187 - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
188 - Remove const from __*_hook variable declarations, so that glibc can modify
189 them during process forking.
190 - Fix mlockall(2)/madvise(2) interaction.
191 - Fix fork(2)-related deadlocks.
192 - Fix error return value for "thread.tcache.enabled" mallctl.
193
194* 3.0.0 (May 11, 2012)
195
196 Although this version adds some major new features, the primary focus is on
197 internal code cleanup that facilitates maintainability and portability, most
198 of which is not reflected in the ChangeLog. This is the first release to
199 incorporate substantial contributions from numerous other developers, and the
200 result is a more broadly useful allocator (see the git revision history for
201 contribution details). Note that the license has been unified, thanks to
202 Facebook granting a license under the same terms as the other copyright
203 holders (see COPYING).
204
205 New features:
206 - Implement Valgrind support, redzones, and quarantine.
207 - Add support for additional platforms:
208 + FreeBSD
209 + Mac OS X Lion
210 + MinGW
211 + Windows (no support yet for replacing the system malloc)
212 - Add support for additional architectures:
213 + MIPS
214 + SH4
215 + Tilera
216 - Add support for cross compiling.
217 - Add nallocm(), which rounds a request size up to the nearest size class
218 without actually allocating.
219 - Implement aligned_alloc() (blame C11).
220 - Add the "thread.tcache.enabled" mallctl.
221 - Add the "opt.prof_final" mallctl.
222 - Update pprof (from gperftools 2.0).
223 - Add the --with-mangling option.
224 - Add the --disable-experimental option.
225 - Add the --disable-munmap option, and make it the default on Linux.
226 - Add the --enable-mremap option, which disables use of mremap(2) by default.
227
228 Incompatible changes:
229 - Enable stats by default.
230 - Enable fill by default.
231 - Disable lazy locking by default.
232 - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
233 - Rename the "arenas.pagesize" mallctl to "arenas.page".
234 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
235 - Change the "opt.prof_accum" default from true to false.
236
237 Removed features:
238 - Remove the swap feature, including the "config.swap", "swap.avail",
239 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
240 - Remove highruns statistics, including the
241 "stats.arenas.<i>.bins.<j>.highruns" and
242 "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
243 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
244 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
245 "arenas.[tqcs]bins" mallctls.
246 - Remove the "arenas.chunksize" mallctl.
247 - Remove the "opt.lg_prof_tcmax" option.
248 - Remove the "opt.lg_prof_bt_max" option.
249 - Remove the "opt.lg_tcache_gc_sweep" option.
250 - Remove the --disable-tiny option, including the "config.tiny" mallctl.
251 - Remove the --enable-dynamic-page-shift configure option.
252 - Remove the --enable-sysv configure option.
253
254 Bug fixes:
255 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
256 invalid statistics and crashes.
257 - Work around TLS deallocation via free() on Linux. This bug could cause
258 write-after-free memory corruption.
259 - Fix a potential deadlock that could occur during interval- and
260 growth-triggered heap profile dumps.
261 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
262 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could
263 cause memory corruption and crashes with --enable-dss specified.
264 - Fix fork-related bugs that could cause deadlock in children between fork
265 and exec.
266 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
267 - Fix realloc(p, 0) to act like free(p).
268 - Do not enforce minimum alignment in memalign().
269 - Check for NULL pointer in malloc_usable_size().
270 - Fix an off-by-one heap profile statistics bug that could be observed in
271 interval- and growth-triggered heap profiles.
272 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
273 is 0.
274 - Fix bin->runcur management to fix a layout policy bug. This bug did not
275 affect correctness.
276 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
277 initialized than necessary.
278 - Add missing "opt.lg_tcache_max" mallctl implementation.
279 - Use glibc allocator hooks to make mixed allocator usage less likely.
280 - Fix build issues for --disable-tcache.
281 - Don't mangle pthread_create() when --with-private-namespace is specified.
282
283* 2.2.5 (November 14, 2011)
284
285 Bug fixes:
286 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that
287 could cause memory corruption and/or crashes.
288 - Fix huge_ralloc() to maintain chunk statistics.
289 - Fix malloc_stats_print(..., "a") output.
290
291* 2.2.4 (November 5, 2011)
292
293 Bug fixes:
294 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as
295 well as for --disable-tls builds in earlier releases.
296 - Do not assume a 4 KiB page size in test/rallocm.c.
297
298* 2.2.3 (August 31, 2011)
299
300 This version fixes numerous bugs related to heap profiling.
301
302 Bug fixes:
303 - Fix a prof-related race condition. This bug could cause memory corruption,
304 but only occurred in non-default configurations (prof_accum:false).
305 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
306 excluded from backtraces).
307 - Fix a prof-related bug in realloc() (only triggered by OOM errors).
308 - Fix prof-related bugs in allocm() and rallocm().
309 - Fix prof_tdata_cleanup() for --disable-tls builds.
310 - Fix a relative include path, to fix objdir builds.
311
312* 2.2.2 (July 30, 2011)
313
314 Bug fixes:
315 - Fix a build error for --disable-tcache.
316 - Fix assertions in arena_purge() (for real this time).
317 - Add the --with-private-namespace option. This is a workaround for symbol
318 conflicts that can inadvertently arise when using static libraries.
319
320* 2.2.1 (March 30, 2011)
321
322 Bug fixes:
323 - Implement atomic operations for x86/x64. This fixes compilation failures
324 for versions of gcc that are still in wide use.
325 - Fix an assertion in arena_purge().
326
327* 2.2.0 (March 22, 2011)
328
329 This version incorporates several improvements to algorithms and data
330 structures that tend to reduce fragmentation and increase speed.
331
332 New features:
333 - Add the "stats.cactive" mallctl.
334 - Update pprof (from google-perftools 1.7).
335 - Improve backtracing-related configuration logic, and add the
336 --disable-prof-libgcc option.
337
338 Bug fixes:
339 - Change default symbol visibility from "internal", to "hidden", which
340 decreases the overhead of library-internal function calls.
341 - Fix symbol visibility so that it is also set on OS X.
342 - Fix a build dependency regression caused by the introduction of the .pic.o
343 suffix for PIC object files.
344 - Add missing checks for mutex initialization failures.
345 - Don't use libgcc-based backtracing except on x64, where it is known to work.
346 - Fix deadlocks on OS X that were due to memory allocation in
347 pthread_mutex_lock().
348 - Heap profiling-specific fixes:
349 + Fix memory corruption due to integer overflow in small region index
350 computation, when using a small enough sample interval that profiling
351 context pointers are stored in small run headers.
352 + Fix a bootstrap ordering bug that only occurred with TLS disabled.
353 + Fix a rallocm() rsize bug.
354 + Fix error detection bugs for aligned memory allocation.
355
356* 2.1.3 (March 14, 2011)
357
358 Bug fixes:
359 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
360 for OS X in 2.1.2).
361 - Fix a "thread.arena" mallctl bug.
362 - Fix a thread cache stats merging bug.
363
364* 2.1.2 (March 2, 2011)
365
366 Bug fixes:
367 - Fix "thread.{de,}allocatedp" mallctl for OS X.
368 - Add missing jemalloc.a to build system.
369
370* 2.1.1 (January 31, 2011)
371
372 Bug fixes:
373 - Fix aligned huge reallocation (affected allocm()).
374 - Fix the ALLOCM_LG_ALIGN macro definition.
375 - Fix a heap dumping deadlock.
376 - Fix a "thread.arena" mallctl bug.
377
378* 2.1.0 (December 3, 2010)
379
380 This version incorporates some optimizations that can't quite be considered
381 bug fixes.
382
383 New features:
384 - Use Linux's mremap(2) for huge object reallocation when possible.
385 - Avoid locking in mallctl*() when possible.
386 - Add the "thread.[de]allocatedp" mallctl's.
387 - Convert the manual page source from roff to DocBook, and generate both roff
388 and HTML manuals.
389
390 Bug fixes:
391 - Fix a crash due to incorrect bootstrap ordering. This only impacted
392 --enable-debug --enable-dss configurations.
393 - Fix a minor statistics bug for mallctl("swap.avail", ...).
394
395* 2.0.1 (October 29, 2010)
396
397 Bug fixes:
398 - Fix a race condition in heap profiling that could cause undefined behavior
399 if "opt.prof_accum" were disabled.
400 - Add missing mutex unlocks for some OOM error paths in the heap profiling
401 code.
402 - Fix a compilation error for non-C99 builds.
403
404* 2.0.0 (October 24, 2010)
405
406 This version focuses on the experimental *allocm() API, and on improved
407 run-time configuration/introspection. Nonetheless, numerous performance
408 improvements are also included.
409
410 New features:
411 - Implement the experimental {,r,s,d}allocm() API, which provides a superset
412 of the functionality available via malloc(), calloc(), posix_memalign(),
413 realloc(), malloc_usable_size(), and free(). These functions can be used to
414 allocate/reallocate aligned zeroed memory, ask for optional extra memory
415 during reallocation, prevent object movement during reallocation, etc.
416 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
417 more human-readable, and more flexible. For example:
418 JEMALLOC_OPTIONS=AJP
419 is now:
420 MALLOC_CONF=abort:true,fill:true,stats_print:true
421 - Port to Apple OS X. Sponsored by Mozilla.
422 - Make it possible for the application to control thread-->arena mappings via
423 the "thread.arena" mallctl.
424 - Add compile-time support for all TLS-related functionality via pthreads TSD.
425 This is mainly of interest for OS X, which does not support TLS, but has a
426 TSD implementation with similar performance.
427 - Override memalign() and valloc() if they are provided by the system.
428 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
429 dirty unused pages.
430 - Make cumulative heap profiling data optional, so that it is possible to
431 limit the amount of memory consumed by heap profiling data structures.
432 - Add per thread allocation counters that can be accessed via the
433 "thread.allocated" and "thread.deallocated" mallctls.
434
435 Incompatible changes:
436 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
437 - Increase default backtrace depth from 4 to 128 for heap profiling.
438 - Disable interval-based profile dumps by default.
439
440 Bug fixes:
441 - Remove bad assertions in fork handler functions. These assertions could
442 cause aborts for some combinations of configure settings.
443 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
444 - Fix leak context reporting. This bug tended to cause the number of contexts
445 to be underreported (though the reported number of objects and bytes were
446 correct).
447 - Fix a realloc() bug for large in-place growing reallocation. This bug could
448 cause memory corruption, but it was hard to trigger.
449 - Fix an allocation bug for small allocations that could be triggered if
450 multiple threads raced to create a new run of backing pages.
451 - Enhance the heap profiler to trigger samples based on usable size, rather
452 than request size.
453 - Fix a heap profiling bug due to sometimes losing track of requested object
454 size for sampled objects.
455
456* 1.0.3 (August 12, 2010)
457
458 Bug fixes:
459 - Fix the libunwind-based implementation of stack backtracing (used for heap
460 profiling). This bug could cause zero-length backtraces to be reported.
461 - Add a missing mutex unlock in library initialization code. If multiple
462 threads raced to initialize malloc, some of them could end up permanently
463 blocked.
464
465* 1.0.2 (May 11, 2010)
466
467 Bug fixes:
468 - Fix junk filling of large objects, which could cause memory corruption.
469 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
470 memory limits could cause swap file configuration to fail. Contributed by
471 Jordan DeLong.
472
473* 1.0.1 (April 14, 2010)
474
475 Bug fixes:
476 - Fix compilation when --enable-fill is specified.
477 - Fix threads-related profiling bugs that affected accuracy and caused memory
478 to be leaked during thread exit.
479 - Fix dirty page purging race conditions that could cause crashes.
480 - Fix crash in tcache flushing code during thread destruction.
481
482* 1.0.0 (April 11, 2010)
483
484 This release focuses on speed and run-time introspection. Numerous
485 algorithmic improvements make this release substantially faster than its
486 predecessors.
487
488 New features:
489 - Implement autoconf-based configuration system.
490 - Add mallctl*(), for the purposes of introspection and run-time
491 configuration.
492 - Make it possible for the application to manually flush a thread's cache, via
493 the "tcache.flush" mallctl.
494 - Base maximum dirty page count on proportion of active memory.
495 - Compute various addtional run-time statistics, including per size class
496 statistics for large objects.
497 - Expose malloc_stats_print(), which can be called repeatedly by the
498 application.
499 - Simplify the malloc_message() signature to only take one string argument,
500 and incorporate an opaque data pointer argument for use by the application
501 in combination with malloc_stats_print().
502 - Add support for allocation backed by one or more swap files, and allow the
503 application to disable over-commit if swap files are in use.
504 - Implement allocation profiling and leak checking.
505
506 Removed features:
507 - Remove the dynamic arena rebalancing code, since thread-specific caching
508 reduces its utility.
509
510 Bug fixes:
511 - Modify chunk allocation to work when address space layout randomization
512 (ASLR) is in use.
513 - Fix thread cleanup bugs related to TLS destruction.
514 - Handle 0-size allocation requests in posix_memalign().
515 - Fix a chunk leak. The leaked chunks were never touched, so this impacted
516 virtual memory usage, but not physical memory usage.
517
518* linux_2008082[78]a (August 27/28, 2008)
519
520 These snapshot releases are the simple result of incorporating Linux-specific
521 support into the FreeBSD malloc sources.
522
523--------------------------------------------------------------------------------
524vim:filetype=text:textwidth=80
2bug fixes are all mentioned, but internal enhancements are omitted here for
3brevity (even though they are more fun to write about). Much more detail can be
4found in the git revision history:
5
6 https://github.com/jemalloc/jemalloc
7
8* 3.5.1 (February 25, 2014)
9
10 This version primarily addresses minor bugs in test code.
11
12 Bug fixes:
13 - Configure Solaris/Illumos to use MADV_FREE.
14 - Fix junk filling for mremap(2)-based huge reallocation. This is only
15 relevant if configuring with the --enable-mremap option specified.
16 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the
17 compiler.
18 - Add a configure test for SSE2 rather than assuming it is usable on i686
19 systems. This fixes test compilation errors, especially on 32-bit Linux
20 systems.
21 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit
22 test.
23 - Fix/remove flawed alignment-related overflow tests.
24 - Prevent compiler optimizations that could change backtraces in the
25 prof_accum unit test.
26
27* 3.5.0 (January 22, 2014)
28
29 This version focuses on refactoring and automated testing, though it also
30 includes some non-trivial heap profiling optimizations not mentioned below.
31
32 New features:
33 - Add the *allocx() API, which is a successor to the experimental *allocm()
34 API. The *allocx() functions are slightly simpler to use because they have
35 fewer parameters, they directly return the results of primary interest, and
36 mallocx()/rallocx() avoid the strict aliasing pitfall that
37 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is
38 slated for removal in the next non-bugfix release.
39 - Add support for LinuxThreads.
40
41 Bug fixes:
42 - Unless heap profiling is enabled, disable floating point code and don't link
43 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64
44 systems, makes it possible to completely disable floating point register
45 use. Some versions of glibc neglect to save/restore caller-saved floating
46 point registers during dynamic lazy symbol loading, and the symbol loading
47 code uses whatever malloc the application happens to have linked/loaded
48 with, the result being potential floating point register corruption.
49 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling
50 backtrace creation in imemalign(). This bug impacted posix_memalign() and
51 aligned_alloc().
52 - Fix a file descriptor leak in a prof_dump_maps() error path.
53 - Fix prof_dump() to close the dump file descriptor for all relevant error
54 paths.
55 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for
56 allocation, not just deallocation.
57 - Fix a data race for large allocation stats counters.
58 - Fix a potential infinite loop during thread exit. This bug occurred on
59 Solaris, and could affect other platforms with similar pthreads TSD
60 implementations.
61 - Don't junk-fill reallocations unless usable size changes. This fixes a
62 violation of the *allocx()/*allocm() semantics.
63 - Fix growing large reallocation to junk fill new space.
64 - Fix huge deallocation to junk fill when munmap is disabled.
65 - Change the default private namespace prefix from empty to je_, and change
66 --with-private-namespace-prefix so that it prepends an additional prefix
67 rather than replacing je_. This reduces the likelihood of applications
68 which statically link jemalloc experiencing symbol name collisions.
69 - Add missing private namespace mangling (relevant when
70 --with-private-namespace is specified).
71 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as
72 static even for debug builds.
73 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice
74 this error path is never executed.
75 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These
76 bugs had no impact except for malformed inputs.
77 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by
78 existing calls, so they had no impact.
79
80* 3.4.1 (October 20, 2013)
81
82 Bug fixes:
83 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption
84 of internal data structures and subsequent crashes.
85 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of
86 uninitialized memory in:
87 + arena chunk headers
88 + internal zero-initialized data structures (relevant to tcache and prof
89 code)
90 - Preserve errno during the first allocation. A readlink(2) call during
91 initialization fails unless /etc/malloc.conf exists, so errno was typically
92 set during the first allocation prior to this fix.
93 - Fix compilation warnings reported by gcc 4.8.1.
94
95* 3.4.0 (June 2, 2013)
96
97 This version is essentially a small bugfix release, but the addition of
98 aarch64 support requires that the minor version be incremented.
99
100 Bug fixes:
101 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were
102 typically triggered by multiple threads concurrently deallocating huge
103 objects.
104
105 New features:
106 - Add support for the aarch64 architecture.
107
108* 3.3.1 (March 6, 2013)
109
110 This version fixes bugs that are typically encountered only when utilizing
111 custom run-time options.
112
113 Bug fixes:
114 - Fix a locking order bug that could cause deadlock during fork if heap
115 profiling were enabled.
116 - Fix a chunk recycling bug that could cause the allocator to lose track of
117 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause
118 corruption if allocating via sbrk(2) (unlikely unless running with the
119 "dss:primary" option specified). This was completely harmless on Linux
120 unless using mlockall(2) (and unlikely even then, unless the
121 --disable-munmap configure option or the "dss:primary" option was
122 specified). This regression was introduced in 3.1.0 by the
123 mlockall(2)/madvise(2) interaction fix.
124 - Fix TLS-related memory corruption that could occur during thread exit if the
125 thread never allocated memory. Only the quarantine and prof facilities were
126 susceptible.
127 - Fix two quarantine bugs:
128 + Internal reallocation of the quarantined object array leaked the old
129 array.
130 + Reallocation failure for internal reallocation of the quarantined object
131 array (very unlikely) resulted in memory corruption.
132 - Fix Valgrind integration to annotate all internally allocated memory in a
133 way that keeps Valgrind happy about internal data structure access.
134 - Fix building for s390 systems.
135
136* 3.3.0 (January 23, 2013)
137
138 This version includes a few minor performance improvements in addition to the
139 listed new features and bug fixes.
140
141 New features:
142 - Add clipping support to lg_chunk option processing.
143 - Add the --enable-ivsalloc option.
144 - Add the --without-export option.
145 - Add the --disable-zone-allocator option.
146
147 Bug fixes:
148 - Fix "arenas.extend" mallctl to output the number of arenas.
149 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory
150 is undefined.
151 - Fix build break on FreeBSD related to alloca.h.
152
153* 3.2.0 (November 9, 2012)
154
155 In addition to a couple of bug fixes, this version modifies page run
156 allocation and dirty page purging algorithms in order to better control
157 page-level virtual memory fragmentation.
158
159 Incompatible changes:
160 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
161
162 Bug fixes:
163 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
164 after primary dss allocation fails.
165 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced
166 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
167
168* 3.1.0 (October 16, 2012)
169
170 New features:
171 - Auto-detect whether running inside Valgrind, thus removing the need to
172 manually specify MALLOC_CONF=valgrind:true.
173 - Add the "arenas.extend" mallctl, which allows applications to create
174 manually managed arenas.
175 - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
176 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
177 which provide control over dss/mmap precedence.
178 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
179 - Define LG_QUANTUM for hppa.
180
181 Incompatible changes:
182 - Disable tcache by default if running inside Valgrind, in order to avoid
183 making unallocated objects appear reachable to Valgrind.
184 - Drop const from malloc_usable_size() argument on Linux.
185
186 Bug fixes:
187 - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
188 - Remove const from __*_hook variable declarations, so that glibc can modify
189 them during process forking.
190 - Fix mlockall(2)/madvise(2) interaction.
191 - Fix fork(2)-related deadlocks.
192 - Fix error return value for "thread.tcache.enabled" mallctl.
193
194* 3.0.0 (May 11, 2012)
195
196 Although this version adds some major new features, the primary focus is on
197 internal code cleanup that facilitates maintainability and portability, most
198 of which is not reflected in the ChangeLog. This is the first release to
199 incorporate substantial contributions from numerous other developers, and the
200 result is a more broadly useful allocator (see the git revision history for
201 contribution details). Note that the license has been unified, thanks to
202 Facebook granting a license under the same terms as the other copyright
203 holders (see COPYING).
204
205 New features:
206 - Implement Valgrind support, redzones, and quarantine.
207 - Add support for additional platforms:
208 + FreeBSD
209 + Mac OS X Lion
210 + MinGW
211 + Windows (no support yet for replacing the system malloc)
212 - Add support for additional architectures:
213 + MIPS
214 + SH4
215 + Tilera
216 - Add support for cross compiling.
217 - Add nallocm(), which rounds a request size up to the nearest size class
218 without actually allocating.
219 - Implement aligned_alloc() (blame C11).
220 - Add the "thread.tcache.enabled" mallctl.
221 - Add the "opt.prof_final" mallctl.
222 - Update pprof (from gperftools 2.0).
223 - Add the --with-mangling option.
224 - Add the --disable-experimental option.
225 - Add the --disable-munmap option, and make it the default on Linux.
226 - Add the --enable-mremap option, which disables use of mremap(2) by default.
227
228 Incompatible changes:
229 - Enable stats by default.
230 - Enable fill by default.
231 - Disable lazy locking by default.
232 - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
233 - Rename the "arenas.pagesize" mallctl to "arenas.page".
234 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
235 - Change the "opt.prof_accum" default from true to false.
236
237 Removed features:
238 - Remove the swap feature, including the "config.swap", "swap.avail",
239 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
240 - Remove highruns statistics, including the
241 "stats.arenas.<i>.bins.<j>.highruns" and
242 "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
243 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
244 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
245 "arenas.[tqcs]bins" mallctls.
246 - Remove the "arenas.chunksize" mallctl.
247 - Remove the "opt.lg_prof_tcmax" option.
248 - Remove the "opt.lg_prof_bt_max" option.
249 - Remove the "opt.lg_tcache_gc_sweep" option.
250 - Remove the --disable-tiny option, including the "config.tiny" mallctl.
251 - Remove the --enable-dynamic-page-shift configure option.
252 - Remove the --enable-sysv configure option.
253
254 Bug fixes:
255 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
256 invalid statistics and crashes.
257 - Work around TLS deallocation via free() on Linux. This bug could cause
258 write-after-free memory corruption.
259 - Fix a potential deadlock that could occur during interval- and
260 growth-triggered heap profile dumps.
261 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
262 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could
263 cause memory corruption and crashes with --enable-dss specified.
264 - Fix fork-related bugs that could cause deadlock in children between fork
265 and exec.
266 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
267 - Fix realloc(p, 0) to act like free(p).
268 - Do not enforce minimum alignment in memalign().
269 - Check for NULL pointer in malloc_usable_size().
270 - Fix an off-by-one heap profile statistics bug that could be observed in
271 interval- and growth-triggered heap profiles.
272 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
273 is 0.
274 - Fix bin->runcur management to fix a layout policy bug. This bug did not
275 affect correctness.
276 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
277 initialized than necessary.
278 - Add missing "opt.lg_tcache_max" mallctl implementation.
279 - Use glibc allocator hooks to make mixed allocator usage less likely.
280 - Fix build issues for --disable-tcache.
281 - Don't mangle pthread_create() when --with-private-namespace is specified.
282
283* 2.2.5 (November 14, 2011)
284
285 Bug fixes:
286 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that
287 could cause memory corruption and/or crashes.
288 - Fix huge_ralloc() to maintain chunk statistics.
289 - Fix malloc_stats_print(..., "a") output.
290
291* 2.2.4 (November 5, 2011)
292
293 Bug fixes:
294 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as
295 well as for --disable-tls builds in earlier releases.
296 - Do not assume a 4 KiB page size in test/rallocm.c.
297
298* 2.2.3 (August 31, 2011)
299
300 This version fixes numerous bugs related to heap profiling.
301
302 Bug fixes:
303 - Fix a prof-related race condition. This bug could cause memory corruption,
304 but only occurred in non-default configurations (prof_accum:false).
305 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
306 excluded from backtraces).
307 - Fix a prof-related bug in realloc() (only triggered by OOM errors).
308 - Fix prof-related bugs in allocm() and rallocm().
309 - Fix prof_tdata_cleanup() for --disable-tls builds.
310 - Fix a relative include path, to fix objdir builds.
311
312* 2.2.2 (July 30, 2011)
313
314 Bug fixes:
315 - Fix a build error for --disable-tcache.
316 - Fix assertions in arena_purge() (for real this time).
317 - Add the --with-private-namespace option. This is a workaround for symbol
318 conflicts that can inadvertently arise when using static libraries.
319
320* 2.2.1 (March 30, 2011)
321
322 Bug fixes:
323 - Implement atomic operations for x86/x64. This fixes compilation failures
324 for versions of gcc that are still in wide use.
325 - Fix an assertion in arena_purge().
326
327* 2.2.0 (March 22, 2011)
328
329 This version incorporates several improvements to algorithms and data
330 structures that tend to reduce fragmentation and increase speed.
331
332 New features:
333 - Add the "stats.cactive" mallctl.
334 - Update pprof (from google-perftools 1.7).
335 - Improve backtracing-related configuration logic, and add the
336 --disable-prof-libgcc option.
337
338 Bug fixes:
339 - Change default symbol visibility from "internal", to "hidden", which
340 decreases the overhead of library-internal function calls.
341 - Fix symbol visibility so that it is also set on OS X.
342 - Fix a build dependency regression caused by the introduction of the .pic.o
343 suffix for PIC object files.
344 - Add missing checks for mutex initialization failures.
345 - Don't use libgcc-based backtracing except on x64, where it is known to work.
346 - Fix deadlocks on OS X that were due to memory allocation in
347 pthread_mutex_lock().
348 - Heap profiling-specific fixes:
349 + Fix memory corruption due to integer overflow in small region index
350 computation, when using a small enough sample interval that profiling
351 context pointers are stored in small run headers.
352 + Fix a bootstrap ordering bug that only occurred with TLS disabled.
353 + Fix a rallocm() rsize bug.
354 + Fix error detection bugs for aligned memory allocation.
355
356* 2.1.3 (March 14, 2011)
357
358 Bug fixes:
359 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
360 for OS X in 2.1.2).
361 - Fix a "thread.arena" mallctl bug.
362 - Fix a thread cache stats merging bug.
363
364* 2.1.2 (March 2, 2011)
365
366 Bug fixes:
367 - Fix "thread.{de,}allocatedp" mallctl for OS X.
368 - Add missing jemalloc.a to build system.
369
370* 2.1.1 (January 31, 2011)
371
372 Bug fixes:
373 - Fix aligned huge reallocation (affected allocm()).
374 - Fix the ALLOCM_LG_ALIGN macro definition.
375 - Fix a heap dumping deadlock.
376 - Fix a "thread.arena" mallctl bug.
377
378* 2.1.0 (December 3, 2010)
379
380 This version incorporates some optimizations that can't quite be considered
381 bug fixes.
382
383 New features:
384 - Use Linux's mremap(2) for huge object reallocation when possible.
385 - Avoid locking in mallctl*() when possible.
386 - Add the "thread.[de]allocatedp" mallctl's.
387 - Convert the manual page source from roff to DocBook, and generate both roff
388 and HTML manuals.
389
390 Bug fixes:
391 - Fix a crash due to incorrect bootstrap ordering. This only impacted
392 --enable-debug --enable-dss configurations.
393 - Fix a minor statistics bug for mallctl("swap.avail", ...).
394
395* 2.0.1 (October 29, 2010)
396
397 Bug fixes:
398 - Fix a race condition in heap profiling that could cause undefined behavior
399 if "opt.prof_accum" were disabled.
400 - Add missing mutex unlocks for some OOM error paths in the heap profiling
401 code.
402 - Fix a compilation error for non-C99 builds.
403
404* 2.0.0 (October 24, 2010)
405
406 This version focuses on the experimental *allocm() API, and on improved
407 run-time configuration/introspection. Nonetheless, numerous performance
408 improvements are also included.
409
410 New features:
411 - Implement the experimental {,r,s,d}allocm() API, which provides a superset
412 of the functionality available via malloc(), calloc(), posix_memalign(),
413 realloc(), malloc_usable_size(), and free(). These functions can be used to
414 allocate/reallocate aligned zeroed memory, ask for optional extra memory
415 during reallocation, prevent object movement during reallocation, etc.
416 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
417 more human-readable, and more flexible. For example:
418 JEMALLOC_OPTIONS=AJP
419 is now:
420 MALLOC_CONF=abort:true,fill:true,stats_print:true
421 - Port to Apple OS X. Sponsored by Mozilla.
422 - Make it possible for the application to control thread-->arena mappings via
423 the "thread.arena" mallctl.
424 - Add compile-time support for all TLS-related functionality via pthreads TSD.
425 This is mainly of interest for OS X, which does not support TLS, but has a
426 TSD implementation with similar performance.
427 - Override memalign() and valloc() if they are provided by the system.
428 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
429 dirty unused pages.
430 - Make cumulative heap profiling data optional, so that it is possible to
431 limit the amount of memory consumed by heap profiling data structures.
432 - Add per thread allocation counters that can be accessed via the
433 "thread.allocated" and "thread.deallocated" mallctls.
434
435 Incompatible changes:
436 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
437 - Increase default backtrace depth from 4 to 128 for heap profiling.
438 - Disable interval-based profile dumps by default.
439
440 Bug fixes:
441 - Remove bad assertions in fork handler functions. These assertions could
442 cause aborts for some combinations of configure settings.
443 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
444 - Fix leak context reporting. This bug tended to cause the number of contexts
445 to be underreported (though the reported number of objects and bytes were
446 correct).
447 - Fix a realloc() bug for large in-place growing reallocation. This bug could
448 cause memory corruption, but it was hard to trigger.
449 - Fix an allocation bug for small allocations that could be triggered if
450 multiple threads raced to create a new run of backing pages.
451 - Enhance the heap profiler to trigger samples based on usable size, rather
452 than request size.
453 - Fix a heap profiling bug due to sometimes losing track of requested object
454 size for sampled objects.
455
456* 1.0.3 (August 12, 2010)
457
458 Bug fixes:
459 - Fix the libunwind-based implementation of stack backtracing (used for heap
460 profiling). This bug could cause zero-length backtraces to be reported.
461 - Add a missing mutex unlock in library initialization code. If multiple
462 threads raced to initialize malloc, some of them could end up permanently
463 blocked.
464
465* 1.0.2 (May 11, 2010)
466
467 Bug fixes:
468 - Fix junk filling of large objects, which could cause memory corruption.
469 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
470 memory limits could cause swap file configuration to fail. Contributed by
471 Jordan DeLong.
472
473* 1.0.1 (April 14, 2010)
474
475 Bug fixes:
476 - Fix compilation when --enable-fill is specified.
477 - Fix threads-related profiling bugs that affected accuracy and caused memory
478 to be leaked during thread exit.
479 - Fix dirty page purging race conditions that could cause crashes.
480 - Fix crash in tcache flushing code during thread destruction.
481
482* 1.0.0 (April 11, 2010)
483
484 This release focuses on speed and run-time introspection. Numerous
485 algorithmic improvements make this release substantially faster than its
486 predecessors.
487
488 New features:
489 - Implement autoconf-based configuration system.
490 - Add mallctl*(), for the purposes of introspection and run-time
491 configuration.
492 - Make it possible for the application to manually flush a thread's cache, via
493 the "tcache.flush" mallctl.
494 - Base maximum dirty page count on proportion of active memory.
495 - Compute various addtional run-time statistics, including per size class
496 statistics for large objects.
497 - Expose malloc_stats_print(), which can be called repeatedly by the
498 application.
499 - Simplify the malloc_message() signature to only take one string argument,
500 and incorporate an opaque data pointer argument for use by the application
501 in combination with malloc_stats_print().
502 - Add support for allocation backed by one or more swap files, and allow the
503 application to disable over-commit if swap files are in use.
504 - Implement allocation profiling and leak checking.
505
506 Removed features:
507 - Remove the dynamic arena rebalancing code, since thread-specific caching
508 reduces its utility.
509
510 Bug fixes:
511 - Modify chunk allocation to work when address space layout randomization
512 (ASLR) is in use.
513 - Fix thread cleanup bugs related to TLS destruction.
514 - Handle 0-size allocation requests in posix_memalign().
515 - Fix a chunk leak. The leaked chunks were never touched, so this impacted
516 virtual memory usage, but not physical memory usage.
517
518* linux_2008082[78]a (August 27/28, 2008)
519
520 These snapshot releases are the simple result of incorporating Linux-specific
521 support into the FreeBSD malloc sources.
522
523--------------------------------------------------------------------------------
524vim:filetype=text:textwidth=80