1
Following are change highlights associated with official releases. Important
2
bug fixes are all mentioned, but internal enhancements are omitted here for
3
brevity (even though they are more fun to write about). Much more detail can be
4
found in the git revision history:
6
http://www.canonware.com/cgi-bin/gitweb.cgi?p=jemalloc.git
7
git://canonware.com/jemalloc.git
9
* 3.3.1 (March 6, 2013)
11
This version fixes bugs that are typically encountered only when utilizing
12
custom run-time options.
15
- Fix a locking order bug that could cause deadlock during fork if heap
16
profiling were enabled.
17
- Fix a chunk recycling bug that could cause the allocator to lose track of
18
whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause
19
corruption if allocating via sbrk(2) (unlikely unless running with the
20
"dss:primary" option specified). This was completely harmless on Linux
21
unless using mlockall(2) (and unlikely even then, unless the
22
--disable-munmap configure option or the "dss:primary" option was
23
specified). This regression was introduced in 3.1.0 by the
24
mlockall(2)/madvise(2) interaction fix.
25
- Fix TLS-related memory corruption that could occur during thread exit if the
26
thread never allocated memory. Only the quarantine and prof facilities were
28
- Fix two quarantine bugs:
29
+ Internal reallocation of the quarantined object array leaked the old
31
+ Reallocation failure for internal reallocation of the quarantined object
32
array (very unlikely) resulted in memory corruption.
33
- Fix Valgrind integration to annotate all internally allocated memory in a
34
way that keeps Valgrind happy about internal data structure access.
35
- Fix building for s390 systems.
37
* 3.3.0 (January 23, 2013)
39
This version includes a few minor performance improvements in addition to the
40
listed new features and bug fixes.
43
- Add clipping support to lg_chunk option processing.
44
- Add the --enable-ivsalloc option.
45
- Add the --without-export option.
46
- Add the --disable-zone-allocator option.
49
- Fix "arenas.extend" mallctl to output the number of arenas.
50
- Fix chunk_recycyle() to unconditionally inform Valgrind that returned memory
52
- Fix build break on FreeBSD related to alloca.h.
54
* 3.2.0 (November 9, 2012)
56
In addition to a couple of bug fixes, this version modifies page run
57
allocation and dirty page purging algorithms in order to better control
58
page-level virtual memory fragmentation.
61
- Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
64
- Fix dss/mmap allocation precedence code to use recyclable mmap memory only
65
after primary dss allocation fails.
66
- Fix deadlock in the "arenas.purge" mallctl. This regression was introduced
67
in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
69
* 3.1.0 (October 16, 2012)
72
- Auto-detect whether running inside Valgrind, thus removing the need to
73
manually specify MALLOC_CONF=valgrind:true.
74
- Add the "arenas.extend" mallctl, which allows applications to create
75
manually managed arenas.
76
- Add the ALLOCM_ARENA() flag for {,r,d}allocm().
77
- Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
78
which provide control over dss/mmap precedence.
79
- Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
80
- Define LG_QUANTUM for hppa.
83
- Disable tcache by default if running inside Valgrind, in order to avoid
84
making unallocated objects appear reachable to Valgrind.
85
- Drop const from malloc_usable_size() argument on Linux.
88
- Fix heap profiling crash if sampled object is freed via realloc(p, 0).
89
- Remove const from __*_hook variable declarations, so that glibc can modify
90
them during process forking.
91
- Fix mlockall(2)/madvise(2) interaction.
92
- Fix fork(2)-related deadlocks.
93
- Fix error return value for "thread.tcache.enabled" mallctl.
95
* 3.0.0 (May 11, 2012)
97
Although this version adds some major new features, the primary focus is on
98
internal code cleanup that facilitates maintainability and portability, most
99
of which is not reflected in the ChangeLog. This is the first release to
100
incorporate substantial contributions from numerous other developers, and the
101
result is a more broadly useful allocator (see the git revision history for
102
contribution details). Note that the license has been unified, thanks to
103
Facebook granting a license under the same terms as the other copyright
104
holders (see COPYING).
107
- Implement Valgrind support, redzones, and quarantine.
108
- Add support for additional platforms:
112
+ Windows (no support yet for replacing the system malloc)
113
- Add support for additional architectures:
117
- Add support for cross compiling.
118
- Add nallocm(), which rounds a request size up to the nearest size class
119
without actually allocating.
120
- Implement aligned_alloc() (blame C11).
121
- Add the "thread.tcache.enabled" mallctl.
122
- Add the "opt.prof_final" mallctl.
123
- Update pprof (from gperftools 2.0).
124
- Add the --with-mangling option.
125
- Add the --disable-experimental option.
126
- Add the --disable-munmap option, and make it the default on Linux.
127
- Add the --enable-mremap option, which disables use of mremap(2) by default.
129
Incompatible changes:
130
- Enable stats by default.
131
- Enable fill by default.
132
- Disable lazy locking by default.
133
- Rename the "tcache.flush" mallctl to "thread.tcache.flush".
134
- Rename the "arenas.pagesize" mallctl to "arenas.page".
135
- Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
136
- Change the "opt.prof_accum" default from true to false.
139
- Remove the swap feature, including the "config.swap", "swap.avail",
140
"swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
141
- Remove highruns statistics, including the
142
"stats.arenas.<i>.bins.<j>.highruns" and
143
"stats.arenas.<i>.lruns.<j>.highruns" mallctls.
144
- As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
145
"arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
146
"arenas.[tqcs]bins" mallctls.
147
- Remove the "arenas.chunksize" mallctl.
148
- Remove the "opt.lg_prof_tcmax" option.
149
- Remove the "opt.lg_prof_bt_max" option.
150
- Remove the "opt.lg_tcache_gc_sweep" option.
151
- Remove the --disable-tiny option, including the "config.tiny" mallctl.
152
- Remove the --enable-dynamic-page-shift configure option.
153
- Remove the --enable-sysv configure option.
156
- Fix a statistics-related bug in the "thread.arena" mallctl that could cause
157
invalid statistics and crashes.
158
- Work around TLS deallocation via free() on Linux. This bug could cause
159
write-after-free memory corruption.
160
- Fix a potential deadlock that could occur during interval- and
161
growth-triggered heap profile dumps.
162
- Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
163
- Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could
164
cause memory corruption and crashes with --enable-dss specified.
165
- Fix fork-related bugs that could cause deadlock in children between fork
167
- Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
168
- Fix realloc(p, 0) to act like free(p).
169
- Do not enforce minimum alignment in memalign().
170
- Check for NULL pointer in malloc_usable_size().
171
- Fix an off-by-one heap profile statistics bug that could be observed in
172
interval- and growth-triggered heap profiles.
173
- Fix the "epoch" mallctl to update cached stats even if the passed in epoch
175
- Fix bin->runcur management to fix a layout policy bug. This bug did not
177
- Fix a bug in choose_arena_hard() that potentially caused more arenas to be
178
initialized than necessary.
179
- Add missing "opt.lg_tcache_max" mallctl implementation.
180
- Use glibc allocator hooks to make mixed allocator usage less likely.
181
- Fix build issues for --disable-tcache.
182
- Don't mangle pthread_create() when --with-private-namespace is specified.
184
* 2.2.5 (November 14, 2011)
187
- Fix huge_ralloc() race when using mremap(2). This is a serious bug that
188
could cause memory corruption and/or crashes.
189
- Fix huge_ralloc() to maintain chunk statistics.
190
- Fix malloc_stats_print(..., "a") output.
192
* 2.2.4 (November 5, 2011)
195
- Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as
196
well as for --disable-tls builds in earlier releases.
197
- Do not assume a 4 KiB page size in test/rallocm.c.
199
* 2.2.3 (August 31, 2011)
201
This version fixes numerous bugs related to heap profiling.
204
- Fix a prof-related race condition. This bug could cause memory corruption,
205
but only occurred in non-default configurations (prof_accum:false).
206
- Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
207
excluded from backtraces).
208
- Fix a prof-related bug in realloc() (only triggered by OOM errors).
209
- Fix prof-related bugs in allocm() and rallocm().
210
- Fix prof_tdata_cleanup() for --disable-tls builds.
211
- Fix a relative include path, to fix objdir builds.
213
* 2.2.2 (July 30, 2011)
216
- Fix a build error for --disable-tcache.
217
- Fix assertions in arena_purge() (for real this time).
218
- Add the --with-private-namespace option. This is a workaround for symbol
219
conflicts that can inadvertently arise when using static libraries.
221
* 2.2.1 (March 30, 2011)
224
- Implement atomic operations for x86/x64. This fixes compilation failures
225
for versions of gcc that are still in wide use.
226
- Fix an assertion in arena_purge().
228
* 2.2.0 (March 22, 2011)
230
This version incorporates several improvements to algorithms and data
231
structures that tend to reduce fragmentation and increase speed.
234
- Add the "stats.cactive" mallctl.
235
- Update pprof (from google-perftools 1.7).
236
- Improve backtracing-related configuration logic, and add the
237
--disable-prof-libgcc option.
240
- Change default symbol visibility from "internal", to "hidden", which
241
decreases the overhead of library-internal function calls.
242
- Fix symbol visibility so that it is also set on OS X.
243
- Fix a build dependency regression caused by the introduction of the .pic.o
244
suffix for PIC object files.
245
- Add missing checks for mutex initialization failures.
246
- Don't use libgcc-based backtracing except on x64, where it is known to work.
247
- Fix deadlocks on OS X that were due to memory allocation in
248
pthread_mutex_lock().
249
- Heap profiling-specific fixes:
250
+ Fix memory corruption due to integer overflow in small region index
251
computation, when using a small enough sample interval that profiling
252
context pointers are stored in small run headers.
253
+ Fix a bootstrap ordering bug that only occurred with TLS disabled.
254
+ Fix a rallocm() rsize bug.
255
+ Fix error detection bugs for aligned memory allocation.
257
* 2.1.3 (March 14, 2011)
260
- Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
262
- Fix a "thread.arena" mallctl bug.
263
- Fix a thread cache stats merging bug.
265
* 2.1.2 (March 2, 2011)
268
- Fix "thread.{de,}allocatedp" mallctl for OS X.
269
- Add missing jemalloc.a to build system.
271
* 2.1.1 (January 31, 2011)
274
- Fix aligned huge reallocation (affected allocm()).
275
- Fix the ALLOCM_LG_ALIGN macro definition.
276
- Fix a heap dumping deadlock.
277
- Fix a "thread.arena" mallctl bug.
279
* 2.1.0 (December 3, 2010)
281
This version incorporates some optimizations that can't quite be considered
285
- Use Linux's mremap(2) for huge object reallocation when possible.
286
- Avoid locking in mallctl*() when possible.
287
- Add the "thread.[de]allocatedp" mallctl's.
288
- Convert the manual page source from roff to DocBook, and generate both roff
292
- Fix a crash due to incorrect bootstrap ordering. This only impacted
293
--enable-debug --enable-dss configurations.
294
- Fix a minor statistics bug for mallctl("swap.avail", ...).
296
* 2.0.1 (October 29, 2010)
299
- Fix a race condition in heap profiling that could cause undefined behavior
300
if "opt.prof_accum" were disabled.
301
- Add missing mutex unlocks for some OOM error paths in the heap profiling
303
- Fix a compilation error for non-C99 builds.
305
* 2.0.0 (October 24, 2010)
307
This version focuses on the experimental *allocm() API, and on improved
308
run-time configuration/introspection. Nonetheless, numerous performance
309
improvements are also included.
312
- Implement the experimental {,r,s,d}allocm() API, which provides a superset
313
of the functionality available via malloc(), calloc(), posix_memalign(),
314
realloc(), malloc_usable_size(), and free(). These functions can be used to
315
allocate/reallocate aligned zeroed memory, ask for optional extra memory
316
during reallocation, prevent object movement during reallocation, etc.
317
- Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
318
more human-readable, and more flexible. For example:
321
MALLOC_CONF=abort:true,fill:true,stats_print:true
322
- Port to Apple OS X. Sponsored by Mozilla.
323
- Make it possible for the application to control thread-->arena mappings via
324
the "thread.arena" mallctl.
325
- Add compile-time support for all TLS-related functionality via pthreads TSD.
326
This is mainly of interest for OS X, which does not support TLS, but has a
327
TSD implementation with similar performance.
328
- Override memalign() and valloc() if they are provided by the system.
329
- Add the "arenas.purge" mallctl, which can be used to synchronously purge all
331
- Make cumulative heap profiling data optional, so that it is possible to
332
limit the amount of memory consumed by heap profiling data structures.
333
- Add per thread allocation counters that can be accessed via the
334
"thread.allocated" and "thread.deallocated" mallctls.
336
Incompatible changes:
337
- Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
338
- Increase default backtrace depth from 4 to 128 for heap profiling.
339
- Disable interval-based profile dumps by default.
342
- Remove bad assertions in fork handler functions. These assertions could
343
cause aborts for some combinations of configure settings.
344
- Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
345
- Fix leak context reporting. This bug tended to cause the number of contexts
346
to be underreported (though the reported number of objects and bytes were
348
- Fix a realloc() bug for large in-place growing reallocation. This bug could
349
cause memory corruption, but it was hard to trigger.
350
- Fix an allocation bug for small allocations that could be triggered if
351
multiple threads raced to create a new run of backing pages.
352
- Enhance the heap profiler to trigger samples based on usable size, rather
354
- Fix a heap profiling bug due to sometimes losing track of requested object
355
size for sampled objects.
357
* 1.0.3 (August 12, 2010)
360
- Fix the libunwind-based implementation of stack backtracing (used for heap
361
profiling). This bug could cause zero-length backtraces to be reported.
362
- Add a missing mutex unlock in library initialization code. If multiple
363
threads raced to initialize malloc, some of them could end up permanently
366
* 1.0.2 (May 11, 2010)
369
- Fix junk filling of large objects, which could cause memory corruption.
370
- Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
371
memory limits could cause swap file configuration to fail. Contributed by
374
* 1.0.1 (April 14, 2010)
377
- Fix compilation when --enable-fill is specified.
378
- Fix threads-related profiling bugs that affected accuracy and caused memory
379
to be leaked during thread exit.
380
- Fix dirty page purging race conditions that could cause crashes.
381
- Fix crash in tcache flushing code during thread destruction.
383
* 1.0.0 (April 11, 2010)
385
This release focuses on speed and run-time introspection. Numerous
386
algorithmic improvements make this release substantially faster than its
390
- Implement autoconf-based configuration system.
391
- Add mallctl*(), for the purposes of introspection and run-time
393
- Make it possible for the application to manually flush a thread's cache, via
394
the "tcache.flush" mallctl.
395
- Base maximum dirty page count on proportion of active memory.
396
- Compute various addtional run-time statistics, including per size class
397
statistics for large objects.
398
- Expose malloc_stats_print(), which can be called repeatedly by the
400
- Simplify the malloc_message() signature to only take one string argument,
401
and incorporate an opaque data pointer argument for use by the application
402
in combination with malloc_stats_print().
403
- Add support for allocation backed by one or more swap files, and allow the
404
application to disable over-commit if swap files are in use.
405
- Implement allocation profiling and leak checking.
408
- Remove the dynamic arena rebalancing code, since thread-specific caching
412
- Modify chunk allocation to work when address space layout randomization
414
- Fix thread cleanup bugs related to TLS destruction.
415
- Handle 0-size allocation requests in posix_memalign().
416
- Fix a chunk leak. The leaked chunks were never touched, so this impacted
417
virtual memory usage, but not physical memory usage.
419
* linux_2008082[78]a (August 27/28, 2008)
421
These snapshot releases are the simple result of incorporating Linux-specific
422
support into the FreeBSD malloc sources.
424
--------------------------------------------------------------------------------
425
vim:filetype=text:textwidth=80
added
removed
extra/jemalloc/ChangeLog
