xref: /freebsd/contrib/jemalloc/ChangeLog (revision 88ad2f8dccdd7a04f3687ff79efd55e68912a611)
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    http://www.canonware.com/cgi-bin/gitweb.cgi?p=jemalloc.git
7    git://canonware.com/jemalloc.git
8
9* 3.3.0 (January 23, 2013)
10
11  This version includes a few minor performance improvements in addition to the
12  listed new features and bug fixes.
13
14  New features:
15  - Add clipping support to lg_chunk option processing.
16  - Add the --enable-ivsalloc option.
17  - Add the --without-export option.
18  - Add the --disable-zone-allocator option.
19
20  Bug fixes:
21  - Fix "arenas.extend" mallctl to output the number of arenas.
22  - Fix chunk_recycyle() to unconditionally inform Valgrind that returned memory
23    is undefined.
24  - Fix build break on FreeBSD related to alloca.h.
25
26* 3.2.0 (November 9, 2012)
27
28  In addition to a couple of bug fixes, this version modifies page run
29  allocation and dirty page purging algorithms in order to better control
30  page-level virtual memory fragmentation.
31
32  Incompatible changes:
33  - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
34
35  Bug fixes:
36  - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
37    after primary dss allocation fails.
38  - Fix deadlock in the "arenas.purge" mallctl.  This regression was introduced
39    in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
40
41* 3.1.0 (October 16, 2012)
42
43  New features:
44  - Auto-detect whether running inside Valgrind, thus removing the need to
45    manually specify MALLOC_CONF=valgrind:true.
46  - Add the "arenas.extend" mallctl, which allows applications to create
47    manually managed arenas.
48  - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
49  - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
50    which provide control over dss/mmap precedence.
51  - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
52  - Define LG_QUANTUM for hppa.
53
54  Incompatible changes:
55  - Disable tcache by default if running inside Valgrind, in order to avoid
56    making unallocated objects appear reachable to Valgrind.
57  - Drop const from malloc_usable_size() argument on Linux.
58
59  Bug fixes:
60  - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
61  - Remove const from __*_hook variable declarations, so that glibc can modify
62    them during process forking.
63  - Fix mlockall(2)/madvise(2) interaction.
64  - Fix fork(2)-related deadlocks.
65  - Fix error return value for "thread.tcache.enabled" mallctl.
66
67* 3.0.0 (May 11, 2012)
68
69  Although this version adds some major new features, the primary focus is on
70  internal code cleanup that facilitates maintainability and portability, most
71  of which is not reflected in the ChangeLog.  This is the first release to
72  incorporate substantial contributions from numerous other developers, and the
73  result is a more broadly useful allocator (see the git revision history for
74  contribution details).  Note that the license has been unified, thanks to
75  Facebook granting a license under the same terms as the other copyright
76  holders (see COPYING).
77
78  New features:
79  - Implement Valgrind support, redzones, and quarantine.
80  - Add support for additional platforms:
81    + FreeBSD
82    + Mac OS X Lion
83    + MinGW
84    + Windows (no support yet for replacing the system malloc)
85  - Add support for additional architectures:
86    + MIPS
87    + SH4
88    + Tilera
89  - Add support for cross compiling.
90  - Add nallocm(), which rounds a request size up to the nearest size class
91    without actually allocating.
92  - Implement aligned_alloc() (blame C11).
93  - Add the "thread.tcache.enabled" mallctl.
94  - Add the "opt.prof_final" mallctl.
95  - Update pprof (from gperftools 2.0).
96  - Add the --with-mangling option.
97  - Add the --disable-experimental option.
98  - Add the --disable-munmap option, and make it the default on Linux.
99  - Add the --enable-mremap option, which disables use of mremap(2) by default.
100
101  Incompatible changes:
102  - Enable stats by default.
103  - Enable fill by default.
104  - Disable lazy locking by default.
105  - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
106  - Rename the "arenas.pagesize" mallctl to "arenas.page".
107  - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
108  - Change the "opt.prof_accum" default from true to false.
109
110  Removed features:
111  - Remove the swap feature, including the "config.swap", "swap.avail",
112    "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
113  - Remove highruns statistics, including the
114    "stats.arenas.<i>.bins.<j>.highruns" and
115    "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
116  - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
117    "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
118    "arenas.[tqcs]bins" mallctls.
119  - Remove the "arenas.chunksize" mallctl.
120  - Remove the "opt.lg_prof_tcmax" option.
121  - Remove the "opt.lg_prof_bt_max" option.
122  - Remove the "opt.lg_tcache_gc_sweep" option.
123  - Remove the --disable-tiny option, including the "config.tiny" mallctl.
124  - Remove the --enable-dynamic-page-shift configure option.
125  - Remove the --enable-sysv configure option.
126
127  Bug fixes:
128  - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
129    invalid statistics and crashes.
130  - Work around TLS deallocation via free() on Linux.  This bug could cause
131    write-after-free memory corruption.
132  - Fix a potential deadlock that could occur during interval- and
133    growth-triggered heap profile dumps.
134  - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
135  - Fix chunk_alloc_dss() to stop claiming memory is zeroed.  This bug could
136    cause memory corruption and crashes with --enable-dss specified.
137  - Fix fork-related bugs that could cause deadlock in children between fork
138    and exec.
139  - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
140  - Fix realloc(p, 0) to act like free(p).
141  - Do not enforce minimum alignment in memalign().
142  - Check for NULL pointer in malloc_usable_size().
143  - Fix an off-by-one heap profile statistics bug that could be observed in
144    interval- and growth-triggered heap profiles.
145  - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
146    is 0.
147  - Fix bin->runcur management to fix a layout policy bug.  This bug did not
148    affect correctness.
149  - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
150    initialized than necessary.
151  - Add missing "opt.lg_tcache_max" mallctl implementation.
152  - Use glibc allocator hooks to make mixed allocator usage less likely.
153  - Fix build issues for --disable-tcache.
154  - Don't mangle pthread_create() when --with-private-namespace is specified.
155
156* 2.2.5 (November 14, 2011)
157
158  Bug fixes:
159  - Fix huge_ralloc() race when using mremap(2).  This is a serious bug that
160    could cause memory corruption and/or crashes.
161  - Fix huge_ralloc() to maintain chunk statistics.
162  - Fix malloc_stats_print(..., "a") output.
163
164* 2.2.4 (November 5, 2011)
165
166  Bug fixes:
167  - Initialize arenas_tsd before using it.  This bug existed for 2.2.[0-3], as
168    well as for --disable-tls builds in earlier releases.
169  - Do not assume a 4 KiB page size in test/rallocm.c.
170
171* 2.2.3 (August 31, 2011)
172
173  This version fixes numerous bugs related to heap profiling.
174
175  Bug fixes:
176  - Fix a prof-related race condition.  This bug could cause memory corruption,
177    but only occurred in non-default configurations (prof_accum:false).
178  - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
179    excluded from backtraces).
180  - Fix a prof-related bug in realloc() (only triggered by OOM errors).
181  - Fix prof-related bugs in allocm() and rallocm().
182  - Fix prof_tdata_cleanup() for --disable-tls builds.
183  - Fix a relative include path, to fix objdir builds.
184
185* 2.2.2 (July 30, 2011)
186
187  Bug fixes:
188  - Fix a build error for --disable-tcache.
189  - Fix assertions in arena_purge() (for real this time).
190  - Add the --with-private-namespace option.  This is a workaround for symbol
191    conflicts that can inadvertently arise when using static libraries.
192
193* 2.2.1 (March 30, 2011)
194
195  Bug fixes:
196  - Implement atomic operations for x86/x64.  This fixes compilation failures
197    for versions of gcc that are still in wide use.
198  - Fix an assertion in arena_purge().
199
200* 2.2.0 (March 22, 2011)
201
202  This version incorporates several improvements to algorithms and data
203  structures that tend to reduce fragmentation and increase speed.
204
205  New features:
206  - Add the "stats.cactive" mallctl.
207  - Update pprof (from google-perftools 1.7).
208  - Improve backtracing-related configuration logic, and add the
209    --disable-prof-libgcc option.
210
211  Bug fixes:
212  - Change default symbol visibility from "internal", to "hidden", which
213    decreases the overhead of library-internal function calls.
214  - Fix symbol visibility so that it is also set on OS X.
215  - Fix a build dependency regression caused by the introduction of the .pic.o
216    suffix for PIC object files.
217  - Add missing checks for mutex initialization failures.
218  - Don't use libgcc-based backtracing except on x64, where it is known to work.
219  - Fix deadlocks on OS X that were due to memory allocation in
220    pthread_mutex_lock().
221  - Heap profiling-specific fixes:
222    + Fix memory corruption due to integer overflow in small region index
223      computation, when using a small enough sample interval that profiling
224      context pointers are stored in small run headers.
225    + Fix a bootstrap ordering bug that only occurred with TLS disabled.
226    + Fix a rallocm() rsize bug.
227    + Fix error detection bugs for aligned memory allocation.
228
229* 2.1.3 (March 14, 2011)
230
231  Bug fixes:
232  - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
233    for OS X in 2.1.2).
234  - Fix a "thread.arena" mallctl bug.
235  - Fix a thread cache stats merging bug.
236
237* 2.1.2 (March 2, 2011)
238
239  Bug fixes:
240  - Fix "thread.{de,}allocatedp" mallctl for OS X.
241  - Add missing jemalloc.a to build system.
242
243* 2.1.1 (January 31, 2011)
244
245  Bug fixes:
246  - Fix aligned huge reallocation (affected allocm()).
247  - Fix the ALLOCM_LG_ALIGN macro definition.
248  - Fix a heap dumping deadlock.
249  - Fix a "thread.arena" mallctl bug.
250
251* 2.1.0 (December 3, 2010)
252
253  This version incorporates some optimizations that can't quite be considered
254  bug fixes.
255
256  New features:
257  - Use Linux's mremap(2) for huge object reallocation when possible.
258  - Avoid locking in mallctl*() when possible.
259  - Add the "thread.[de]allocatedp" mallctl's.
260  - Convert the manual page source from roff to DocBook, and generate both roff
261    and HTML manuals.
262
263  Bug fixes:
264  - Fix a crash due to incorrect bootstrap ordering.  This only impacted
265    --enable-debug --enable-dss configurations.
266  - Fix a minor statistics bug for mallctl("swap.avail", ...).
267
268* 2.0.1 (October 29, 2010)
269
270  Bug fixes:
271  - Fix a race condition in heap profiling that could cause undefined behavior
272    if "opt.prof_accum" were disabled.
273  - Add missing mutex unlocks for some OOM error paths in the heap profiling
274    code.
275  - Fix a compilation error for non-C99 builds.
276
277* 2.0.0 (October 24, 2010)
278
279  This version focuses on the experimental *allocm() API, and on improved
280  run-time configuration/introspection.  Nonetheless, numerous performance
281  improvements are also included.
282
283  New features:
284  - Implement the experimental {,r,s,d}allocm() API, which provides a superset
285    of the functionality available via malloc(), calloc(), posix_memalign(),
286    realloc(), malloc_usable_size(), and free().  These functions can be used to
287    allocate/reallocate aligned zeroed memory, ask for optional extra memory
288    during reallocation, prevent object movement during reallocation, etc.
289  - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
290    more human-readable, and more flexible.  For example:
291      JEMALLOC_OPTIONS=AJP
292    is now:
293      MALLOC_CONF=abort:true,fill:true,stats_print:true
294  - Port to Apple OS X.  Sponsored by Mozilla.
295  - Make it possible for the application to control thread-->arena mappings via
296    the "thread.arena" mallctl.
297  - Add compile-time support for all TLS-related functionality via pthreads TSD.
298    This is mainly of interest for OS X, which does not support TLS, but has a
299    TSD implementation with similar performance.
300  - Override memalign() and valloc() if they are provided by the system.
301  - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
302    dirty unused pages.
303  - Make cumulative heap profiling data optional, so that it is possible to
304    limit the amount of memory consumed by heap profiling data structures.
305  - Add per thread allocation counters that can be accessed via the
306    "thread.allocated" and "thread.deallocated" mallctls.
307
308  Incompatible changes:
309  - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
310  - Increase default backtrace depth from 4 to 128 for heap profiling.
311  - Disable interval-based profile dumps by default.
312
313  Bug fixes:
314  - Remove bad assertions in fork handler functions.  These assertions could
315    cause aborts for some combinations of configure settings.
316  - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
317  - Fix leak context reporting.  This bug tended to cause the number of contexts
318    to be underreported (though the reported number of objects and bytes were
319    correct).
320  - Fix a realloc() bug for large in-place growing reallocation.  This bug could
321    cause memory corruption, but it was hard to trigger.
322  - Fix an allocation bug for small allocations that could be triggered if
323    multiple threads raced to create a new run of backing pages.
324  - Enhance the heap profiler to trigger samples based on usable size, rather
325    than request size.
326  - Fix a heap profiling bug due to sometimes losing track of requested object
327    size for sampled objects.
328
329* 1.0.3 (August 12, 2010)
330
331  Bug fixes:
332  - Fix the libunwind-based implementation of stack backtracing (used for heap
333    profiling).  This bug could cause zero-length backtraces to be reported.
334  - Add a missing mutex unlock in library initialization code.  If multiple
335    threads raced to initialize malloc, some of them could end up permanently
336    blocked.
337
338* 1.0.2 (May 11, 2010)
339
340  Bug fixes:
341  - Fix junk filling of large objects, which could cause memory corruption.
342  - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
343    memory limits could cause swap file configuration to fail.  Contributed by
344    Jordan DeLong.
345
346* 1.0.1 (April 14, 2010)
347
348  Bug fixes:
349  - Fix compilation when --enable-fill is specified.
350  - Fix threads-related profiling bugs that affected accuracy and caused memory
351    to be leaked during thread exit.
352  - Fix dirty page purging race conditions that could cause crashes.
353  - Fix crash in tcache flushing code during thread destruction.
354
355* 1.0.0 (April 11, 2010)
356
357  This release focuses on speed and run-time introspection.  Numerous
358  algorithmic improvements make this release substantially faster than its
359  predecessors.
360
361  New features:
362  - Implement autoconf-based configuration system.
363  - Add mallctl*(), for the purposes of introspection and run-time
364    configuration.
365  - Make it possible for the application to manually flush a thread's cache, via
366    the "tcache.flush" mallctl.
367  - Base maximum dirty page count on proportion of active memory.
368  - Compute various addtional run-time statistics, including per size class
369    statistics for large objects.
370  - Expose malloc_stats_print(), which can be called repeatedly by the
371    application.
372  - Simplify the malloc_message() signature to only take one string argument,
373    and incorporate an opaque data pointer argument for use by the application
374    in combination with malloc_stats_print().
375  - Add support for allocation backed by one or more swap files, and allow the
376    application to disable over-commit if swap files are in use.
377  - Implement allocation profiling and leak checking.
378
379  Removed features:
380  - Remove the dynamic arena rebalancing code, since thread-specific caching
381    reduces its utility.
382
383  Bug fixes:
384  - Modify chunk allocation to work when address space layout randomization
385    (ASLR) is in use.
386  - Fix thread cleanup bugs related to TLS destruction.
387  - Handle 0-size allocation requests in posix_memalign().
388  - Fix a chunk leak.  The leaked chunks were never touched, so this impacted
389    virtual memory usage, but not physical memory usage.
390
391* linux_2008082[78]a (August 27/28, 2008)
392
393  These snapshot releases are the simple result of incorporating Linux-specific
394  support into the FreeBSD malloc sources.
395
396--------------------------------------------------------------------------------
397vim:filetype=text:textwidth=80
398