xref: /redis-3.2.3/deps/jemalloc/ChangeLog (revision 5268379e)
1Following are change highlights associated with official releases.  Important
2bug fixes are all mentioned, but some internal enhancements are omitted here for
3brevity.  Much more detail can be found in the git revision history:
4
5    https://github.com/jemalloc/jemalloc
6
7* 4.0.3 (September 24, 2015)
8
9  This bugfix release continues the trend of xallocx() and heap profiling fixes.
10
11  Bug fixes:
12  - Fix xallocx(..., MALLOCX_ZERO) to zero all trailing bytes of large
13    allocations when --enable-cache-oblivious configure option is enabled.
14  - Fix xallocx(..., MALLOCX_ZERO) to zero trailing bytes of huge allocations
15    when resizing from/to a size class that is not a multiple of the chunk size.
16  - Fix prof_tctx_dump_iter() to filter out nodes that were created after heap
17    profile dumping started.
18  - Work around a potentially bad thread-specific data initialization
19    interaction with NPTL (glibc's pthreads implementation).
20
21* 4.0.2 (September 21, 2015)
22
23  This bugfix release addresses a few bugs specific to heap profiling.
24
25  Bug fixes:
26  - Fix ixallocx_prof_sample() to never modify nor create sampled small
27    allocations.  xallocx() is in general incapable of moving small allocations,
28    so this fix removes buggy code without loss of generality.
29  - Fix irallocx_prof_sample() to always allocate large regions, even when
30    alignment is non-zero.
31  - Fix prof_alloc_rollback() to read tdata from thread-specific data rather
32    than dereferencing a potentially invalid tctx.
33
34* 4.0.1 (September 15, 2015)
35
36  This is a bugfix release that is somewhat high risk due to the amount of
37  refactoring required to address deep xallocx() problems.  As a side effect of
38  these fixes, xallocx() now tries harder to partially fulfill requests for
39  optional extra space.  Note that a couple of minor heap profiling
40  optimizations are included, but these are better thought of as performance
41  fixes that were integral to disovering most of the other bugs.
42
43  Optimizations:
44  - Avoid a chunk metadata read in arena_prof_tctx_set(), since it is in the
45    fast path when heap profiling is enabled.  Additionally, split a special
46    case out into arena_prof_tctx_reset(), which also avoids chunk metadata
47    reads.
48  - Optimize irallocx_prof() to optimistically update the sampler state.  The
49    prior implementation appears to have been a holdover from when
50    rallocx()/xallocx() functionality was combined as rallocm().
51
52  Bug fixes:
53  - Fix TLS configuration such that it is enabled by default for platforms on
54    which it works correctly.
55  - Fix arenas_cache_cleanup() and arena_get_hard() to handle
56    allocation/deallocation within the application's thread-specific data
57    cleanup functions even after arenas_cache is torn down.
58  - Fix xallocx() bugs related to size+extra exceeding HUGE_MAXCLASS.
59  - Fix chunk purge hook calls for in-place huge shrinking reallocation to
60    specify the old chunk size rather than the new chunk size.  This bug caused
61    no correctness issues for the default chunk purge function, but was
62    visible to custom functions set via the "arena.<i>.chunk_hooks" mallctl.
63  - Fix heap profiling bugs:
64    + Fix heap profiling to distinguish among otherwise identical sample sites
65      with interposed resets (triggered via the "prof.reset" mallctl).  This bug
66      could cause data structure corruption that would most likely result in a
67      segfault.
68    + Fix irealloc_prof() to prof_alloc_rollback() on OOM.
69    + Make one call to prof_active_get_unlocked() per allocation event, and use
70      the result throughout the relevant functions that handle an allocation
71      event.  Also add a missing check in prof_realloc().  These fixes protect
72      allocation events against concurrent prof_active changes.
73    + Fix ixallocx_prof() to pass usize_max and zero to ixallocx_prof_sample()
74      in the correct order.
75    + Fix prof_realloc() to call prof_free_sampled_object() after calling
76      prof_malloc_sample_object().  Prior to this fix, if tctx and old_tctx were
77      the same, the tctx could have been prematurely destroyed.
78  - Fix portability bugs:
79    + Don't bitshift by negative amounts when encoding/decoding run sizes in
80      chunk header maps.  This affected systems with page sizes greater than 8
81      KiB.
82    + Rename index_t to szind_t to avoid an existing type on Solaris.
83    + Add JEMALLOC_CXX_THROW to the memalign() function prototype, in order to
84      match glibc and avoid compilation errors when including both
85      jemalloc/jemalloc.h and malloc.h in C++ code.
86    + Don't assume that /bin/sh is appropriate when running size_classes.sh
87      during configuration.
88    + Consider __sparcv9 a synonym for __sparc64__ when defining LG_QUANTUM.
89    + Link tests to librt if it contains clock_gettime(2).
90
91* 4.0.0 (August 17, 2015)
92
93  This version contains many speed and space optimizations, both minor and
94  major.  The major themes are generalization, unification, and simplification.
95  Although many of these optimizations cause no visible behavior change, their
96  cumulative effect is substantial.
97
98  New features:
99  - Normalize size class spacing to be consistent across the complete size
100    range.  By default there are four size classes per size doubling, but this
101    is now configurable via the --with-lg-size-class-group option.  Also add the
102    --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and
103    --with-lg-tiny-min options, which can be used to tweak page and size class
104    settings.  Impacts:
105    + Worst case performance for incrementally growing/shrinking reallocation
106      is improved because there are far fewer size classes, and therefore
107      copying happens less often.
108    + Internal fragmentation is limited to 20% for all but the smallest size
109      classes (those less than four times the quantum).  (1B + 4 KiB)
110      and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation.
111    + Chunk fragmentation tends to be lower because there are fewer distinct run
112      sizes to pack.
113  - Add support for explicit tcaches.  The "tcache.create", "tcache.flush", and
114    "tcache.destroy" mallctls control tcache lifetime and flushing, and the
115    MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API
116    control which tcache is used for each operation.
117  - Implement per thread heap profiling, as well as the ability to
118    enable/disable heap profiling on a per thread basis.  Add the "prof.reset",
119    "prof.lg_sample", "thread.prof.name", "thread.prof.active",
120    "opt.prof_thread_active_init", "prof.thread_active_init", and
121    "thread.prof.active" mallctls.
122  - Add support for per arena application-specified chunk allocators, configured
123    via the "arena.<i>.chunk_hooks" mallctl.
124  - Refactor huge allocation to be managed by arenas, so that arenas now
125    function as general purpose independent allocators.  This is important in
126    the context of user-specified chunk allocators, aside from the scalability
127    benefits.  Related new statistics:
128    + The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc",
129      "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests"
130      mallctls provide high level per arena huge allocation statistics.
131    + The "arenas.nhchunks", "arenas.hchunk.<i>.size",
132      "stats.arenas.<i>.hchunks.<j>.nmalloc",
133      "stats.arenas.<i>.hchunks.<j>.ndalloc",
134      "stats.arenas.<i>.hchunks.<j>.nrequests", and
135      "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class
136      statistics.
137  - Add the 'util' column to malloc_stats_print() output, which reports the
138    proportion of available regions that are currently in use for each small
139    size class.
140  - Add "alloc" and "free" modes for for junk filling (see the "opt.junk"
141    mallctl), so that it is possible to separately enable junk filling for
142    allocation versus deallocation.
143  - Add the jemalloc-config script, which provides information about how
144    jemalloc was configured, and how to integrate it into application builds.
145  - Add metadata statistics, which are accessible via the "stats.metadata",
146    "stats.arenas.<i>.metadata.mapped", and
147    "stats.arenas.<i>.metadata.allocated" mallctls.
148  - Add the "stats.resident" mallctl, which reports the upper limit of
149    physically resident memory mapped by the allocator.
150  - Add per arena control over unused dirty page purging, via the
151    "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and
152    "stats.arenas.<i>.lg_dirty_mult" mallctls.
153  - Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump
154    feature on/off during program execution.
155  - Add sdallocx(), which implements sized deallocation.  The primary
156    optimization over dallocx() is the removal of a metadata read, which often
157    suffers an L1 cache miss.
158  - Add missing header includes in jemalloc/jemalloc.h, so that applications
159    only have to #include <jemalloc/jemalloc.h>.
160  - Add support for additional platforms:
161    + Bitrig
162    + Cygwin
163    + DragonFlyBSD
164    + iOS
165    + OpenBSD
166    + OpenRISC/or1k
167
168  Optimizations:
169  - Maintain dirty runs in per arena LRUs rather than in per arena trees of
170    dirty-run-containing chunks.  In practice this change significantly reduces
171    dirty page purging volume.
172  - Integrate whole chunks into the unused dirty page purging machinery.  This
173    reduces the cost of repeated huge allocation/deallocation, because it
174    effectively introduces a cache of chunks.
175  - Split the arena chunk map into two separate arrays, in order to increase
176    cache locality for the frequently accessed bits.
177  - Move small run metadata out of runs, into arena chunk headers.  This reduces
178    run fragmentation, smaller runs reduce external fragmentation for small size
179    classes, and packed (less uniformly aligned) metadata layout improves CPU
180    cache set distribution.
181  - Randomly distribute large allocation base pointer alignment relative to page
182    boundaries in order to more uniformly utilize CPU cache sets.  This can be
183    disabled via the --disable-cache-oblivious configure option, and queried via
184    the "config.cache_oblivious" mallctl.
185  - Micro-optimize the fast paths for the public API functions.
186  - Refactor thread-specific data to reside in a single structure.  This assures
187    that only a single TLS read is necessary per call into the public API.
188  - Implement in-place huge allocation growing and shrinking.
189  - Refactor rtree (radix tree for chunk lookups) to be lock-free, and make
190    additional optimizations that reduce maximum lookup depth to one or two
191    levels.  This resolves what was a concurrency bottleneck for per arena huge
192    allocation, because a global data structure is critical for determining
193    which arenas own which huge allocations.
194
195  Incompatible changes:
196  - Replace --enable-cc-silence with --disable-cc-silence to suppress spurious
197    warnings by default.
198  - Assure that the constness of malloc_usable_size()'s return type matches that
199    of the system implementation.
200  - Change the heap profile dump format to support per thread heap profiling,
201    rename pprof to jeprof, and enhance it with the --thread=<n> option.  As a
202    result, the bundled jeprof must now be used rather than the upstream
203    (gperftools) pprof.
204  - Disable "opt.prof_final" by default, in order to avoid atexit(3), which can
205    internally deadlock on some platforms.
206  - Change the "arenas.nlruns" mallctl type from size_t to unsigned.
207  - Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with
208    "stats.arenas.<i>.bins.<j>.curregs".
209  - Ignore MALLOC_CONF in set{uid,gid,cap} binaries.
210  - Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the
211    MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage.
212
213  Removed features:
214  - Remove the *allocm() API, which is superseded by the *allocx() API.
215  - Remove the --enable-dss options, and make dss non-optional on all platforms
216    which support sbrk(2).
217  - Remove the "arenas.purge" mallctl, which was obsoleted by the
218    "arena.<i>.purge" mallctl in 3.1.0.
219  - Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically
220    detects whether it is running inside Valgrind.
221  - Remove the "stats.huge.allocated", "stats.huge.nmalloc", and
222    "stats.huge.ndalloc" mallctls.
223  - Remove the --enable-mremap option.
224  - Remove the "stats.chunks.current", "stats.chunks.total", and
225    "stats.chunks.high" mallctls.
226
227  Bug fixes:
228  - Fix the cactive statistic to decrease (rather than increase) when active
229    memory decreases.  This regression was first released in 3.5.0.
230  - Fix OOM handling in memalign() and valloc().  A variant of this bug existed
231    in all releases since 2.0.0, which introduced these functions.
232  - Fix an OOM-related regression in arena_tcache_fill_small(), which could
233    cause cache corruption on OOM.  This regression was present in all releases
234    from 2.2.0 through 3.6.0.
235  - Fix size class overflow handling for malloc(), posix_memalign(), memalign(),
236    calloc(), and realloc() when profiling is enabled.
237  - Fix the "arena.<i>.dss" mallctl to return an error if "primary" or
238    "secondary" precedence is specified, but sbrk(2) is not supported.
239  - Fix fallback lg_floor() implementations to handle extremely large inputs.
240  - Ensure the default purgeable zone is after the default zone on OS X.
241  - Fix latent bugs in atomic_*().
242  - Fix the "arena.<i>.dss" mallctl to handle read-only calls.
243  - Fix tls_model configuration to enable the initial-exec model when possible.
244  - Mark malloc_conf as a weak symbol so that the application can override it.
245  - Correctly detect glibc's adaptive pthread mutexes.
246  - Fix the --without-export configure option.
247
248* 3.6.0 (March 31, 2014)
249
250  This version contains a critical bug fix for a regression present in 3.5.0 and
251  3.5.1.
252
253  Bug fixes:
254  - Fix a regression in arena_chunk_alloc() that caused crashes during
255    small/large allocation if chunk allocation failed.  In the absence of this
256    bug, chunk allocation failure would result in allocation failure, e.g.  NULL
257    return from malloc().  This regression was introduced in 3.5.0.
258  - Fix backtracing for gcc intrinsics-based backtracing by specifying
259    -fno-omit-frame-pointer to gcc.  Note that the application (and all the
260    libraries it links to) must also be compiled with this option for
261    backtracing to be reliable.
262  - Use dss allocation precedence for huge allocations as well as small/large
263    allocations.
264  - Fix test assertion failure message formatting.  This bug did not manifest on
265    x86_64 systems because of implementation subtleties in va_list.
266  - Fix inconsequential test failures for hash and SFMT code.
267
268  New features:
269  - Support heap profiling on FreeBSD.  This feature depends on the proc
270    filesystem being mounted during heap profile dumping.
271
272* 3.5.1 (February 25, 2014)
273
274  This version primarily addresses minor bugs in test code.
275
276  Bug fixes:
277  - Configure Solaris/Illumos to use MADV_FREE.
278  - Fix junk filling for mremap(2)-based huge reallocation.  This is only
279    relevant if configuring with the --enable-mremap option specified.
280  - Avoid compilation failure if 'restrict' C99 keyword is not supported by the
281    compiler.
282  - Add a configure test for SSE2 rather than assuming it is usable on i686
283    systems.  This fixes test compilation errors, especially on 32-bit Linux
284    systems.
285  - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit
286    test.
287  - Fix/remove flawed alignment-related overflow tests.
288  - Prevent compiler optimizations that could change backtraces in the
289    prof_accum unit test.
290
291* 3.5.0 (January 22, 2014)
292
293  This version focuses on refactoring and automated testing, though it also
294  includes some non-trivial heap profiling optimizations not mentioned below.
295
296  New features:
297  - Add the *allocx() API, which is a successor to the experimental *allocm()
298    API.  The *allocx() functions are slightly simpler to use because they have
299    fewer parameters, they directly return the results of primary interest, and
300    mallocx()/rallocx() avoid the strict aliasing pitfall that
301    allocm()/rallocm() share with posix_memalign().  Note that *allocm() is
302    slated for removal in the next non-bugfix release.
303  - Add support for LinuxThreads.
304
305  Bug fixes:
306  - Unless heap profiling is enabled, disable floating point code and don't link
307    with libm.  This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64
308    systems, makes it possible to completely disable floating point register
309    use.  Some versions of glibc neglect to save/restore caller-saved floating
310    point registers during dynamic lazy symbol loading, and the symbol loading
311    code uses whatever malloc the application happens to have linked/loaded
312    with, the result being potential floating point register corruption.
313  - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling
314    backtrace creation in imemalign().  This bug impacted posix_memalign() and
315    aligned_alloc().
316  - Fix a file descriptor leak in a prof_dump_maps() error path.
317  - Fix prof_dump() to close the dump file descriptor for all relevant error
318    paths.
319  - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for
320    allocation, not just deallocation.
321  - Fix a data race for large allocation stats counters.
322  - Fix a potential infinite loop during thread exit.  This bug occurred on
323    Solaris, and could affect other platforms with similar pthreads TSD
324    implementations.
325  - Don't junk-fill reallocations unless usable size changes.  This fixes a
326    violation of the *allocx()/*allocm() semantics.
327  - Fix growing large reallocation to junk fill new space.
328  - Fix huge deallocation to junk fill when munmap is disabled.
329  - Change the default private namespace prefix from empty to je_, and change
330    --with-private-namespace-prefix so that it prepends an additional prefix
331    rather than replacing je_.  This reduces the likelihood of applications
332    which statically link jemalloc experiencing symbol name collisions.
333  - Add missing private namespace mangling (relevant when
334    --with-private-namespace is specified).
335  - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as
336    static even for debug builds.
337  - Add a missing mutex unlock in a malloc_init_hard() error path.  In practice
338    this error path is never executed.
339  - Fix numerous bugs in malloc_strotumax() error handling/reporting.  These
340    bugs had no impact except for malformed inputs.
341  - Fix numerous bugs in malloc_snprintf().  These bugs were not exercised by
342    existing calls, so they had no impact.
343
344* 3.4.1 (October 20, 2013)
345
346  Bug fixes:
347  - Fix a race in the "arenas.extend" mallctl that could cause memory corruption
348    of internal data structures and subsequent crashes.
349  - Fix Valgrind integration flaws that caused Valgrind warnings about reads of
350    uninitialized memory in:
351    + arena chunk headers
352    + internal zero-initialized data structures (relevant to tcache and prof
353      code)
354  - Preserve errno during the first allocation.  A readlink(2) call during
355    initialization fails unless /etc/malloc.conf exists, so errno was typically
356    set during the first allocation prior to this fix.
357  - Fix compilation warnings reported by gcc 4.8.1.
358
359* 3.4.0 (June 2, 2013)
360
361  This version is essentially a small bugfix release, but the addition of
362  aarch64 support requires that the minor version be incremented.
363
364  Bug fixes:
365  - Fix race-triggered deadlocks in chunk_record().  These deadlocks were
366    typically triggered by multiple threads concurrently deallocating huge
367    objects.
368
369  New features:
370  - Add support for the aarch64 architecture.
371
372* 3.3.1 (March 6, 2013)
373
374  This version fixes bugs that are typically encountered only when utilizing
375  custom run-time options.
376
377  Bug fixes:
378  - Fix a locking order bug that could cause deadlock during fork if heap
379    profiling were enabled.
380  - Fix a chunk recycling bug that could cause the allocator to lose track of
381    whether a chunk was zeroed.  On FreeBSD, NetBSD, and OS X, it could cause
382    corruption if allocating via sbrk(2) (unlikely unless running with the
383    "dss:primary" option specified).  This was completely harmless on Linux
384    unless using mlockall(2) (and unlikely even then, unless the
385    --disable-munmap configure option or the "dss:primary" option was
386    specified).  This regression was introduced in 3.1.0 by the
387    mlockall(2)/madvise(2) interaction fix.
388  - Fix TLS-related memory corruption that could occur during thread exit if the
389    thread never allocated memory.  Only the quarantine and prof facilities were
390    susceptible.
391  - Fix two quarantine bugs:
392    + Internal reallocation of the quarantined object array leaked the old
393      array.
394    + Reallocation failure for internal reallocation of the quarantined object
395      array (very unlikely) resulted in memory corruption.
396  - Fix Valgrind integration to annotate all internally allocated memory in a
397    way that keeps Valgrind happy about internal data structure access.
398  - Fix building for s390 systems.
399
400* 3.3.0 (January 23, 2013)
401
402  This version includes a few minor performance improvements in addition to the
403  listed new features and bug fixes.
404
405  New features:
406  - Add clipping support to lg_chunk option processing.
407  - Add the --enable-ivsalloc option.
408  - Add the --without-export option.
409  - Add the --disable-zone-allocator option.
410
411  Bug fixes:
412  - Fix "arenas.extend" mallctl to output the number of arenas.
413  - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory
414    is undefined.
415  - Fix build break on FreeBSD related to alloca.h.
416
417* 3.2.0 (November 9, 2012)
418
419  In addition to a couple of bug fixes, this version modifies page run
420  allocation and dirty page purging algorithms in order to better control
421  page-level virtual memory fragmentation.
422
423  Incompatible changes:
424  - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1).
425
426  Bug fixes:
427  - Fix dss/mmap allocation precedence code to use recyclable mmap memory only
428    after primary dss allocation fails.
429  - Fix deadlock in the "arenas.purge" mallctl.  This regression was introduced
430    in 3.1.0 by the addition of the "arena.<i>.purge" mallctl.
431
432* 3.1.0 (October 16, 2012)
433
434  New features:
435  - Auto-detect whether running inside Valgrind, thus removing the need to
436    manually specify MALLOC_CONF=valgrind:true.
437  - Add the "arenas.extend" mallctl, which allows applications to create
438    manually managed arenas.
439  - Add the ALLOCM_ARENA() flag for {,r,d}allocm().
440  - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls,
441    which provide control over dss/mmap precedence.
442  - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".
443  - Define LG_QUANTUM for hppa.
444
445  Incompatible changes:
446  - Disable tcache by default if running inside Valgrind, in order to avoid
447    making unallocated objects appear reachable to Valgrind.
448  - Drop const from malloc_usable_size() argument on Linux.
449
450  Bug fixes:
451  - Fix heap profiling crash if sampled object is freed via realloc(p, 0).
452  - Remove const from __*_hook variable declarations, so that glibc can modify
453    them during process forking.
454  - Fix mlockall(2)/madvise(2) interaction.
455  - Fix fork(2)-related deadlocks.
456  - Fix error return value for "thread.tcache.enabled" mallctl.
457
458* 3.0.0 (May 11, 2012)
459
460  Although this version adds some major new features, the primary focus is on
461  internal code cleanup that facilitates maintainability and portability, most
462  of which is not reflected in the ChangeLog.  This is the first release to
463  incorporate substantial contributions from numerous other developers, and the
464  result is a more broadly useful allocator (see the git revision history for
465  contribution details).  Note that the license has been unified, thanks to
466  Facebook granting a license under the same terms as the other copyright
467  holders (see COPYING).
468
469  New features:
470  - Implement Valgrind support, redzones, and quarantine.
471  - Add support for additional platforms:
472    + FreeBSD
473    + Mac OS X Lion
474    + MinGW
475    + Windows (no support yet for replacing the system malloc)
476  - Add support for additional architectures:
477    + MIPS
478    + SH4
479    + Tilera
480  - Add support for cross compiling.
481  - Add nallocm(), which rounds a request size up to the nearest size class
482    without actually allocating.
483  - Implement aligned_alloc() (blame C11).
484  - Add the "thread.tcache.enabled" mallctl.
485  - Add the "opt.prof_final" mallctl.
486  - Update pprof (from gperftools 2.0).
487  - Add the --with-mangling option.
488  - Add the --disable-experimental option.
489  - Add the --disable-munmap option, and make it the default on Linux.
490  - Add the --enable-mremap option, which disables use of mremap(2) by default.
491
492  Incompatible changes:
493  - Enable stats by default.
494  - Enable fill by default.
495  - Disable lazy locking by default.
496  - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
497  - Rename the "arenas.pagesize" mallctl to "arenas.page".
498  - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
499  - Change the "opt.prof_accum" default from true to false.
500
501  Removed features:
502  - Remove the swap feature, including the "config.swap", "swap.avail",
503    "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
504  - Remove highruns statistics, including the
505    "stats.arenas.<i>.bins.<j>.highruns" and
506    "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
507  - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
508    "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
509    "arenas.[tqcs]bins" mallctls.
510  - Remove the "arenas.chunksize" mallctl.
511  - Remove the "opt.lg_prof_tcmax" option.
512  - Remove the "opt.lg_prof_bt_max" option.
513  - Remove the "opt.lg_tcache_gc_sweep" option.
514  - Remove the --disable-tiny option, including the "config.tiny" mallctl.
515  - Remove the --enable-dynamic-page-shift configure option.
516  - Remove the --enable-sysv configure option.
517
518  Bug fixes:
519  - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
520    invalid statistics and crashes.
521  - Work around TLS deallocation via free() on Linux.  This bug could cause
522    write-after-free memory corruption.
523  - Fix a potential deadlock that could occur during interval- and
524    growth-triggered heap profile dumps.
525  - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
526  - Fix chunk_alloc_dss() to stop claiming memory is zeroed.  This bug could
527    cause memory corruption and crashes with --enable-dss specified.
528  - Fix fork-related bugs that could cause deadlock in children between fork
529    and exec.
530  - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
531  - Fix realloc(p, 0) to act like free(p).
532  - Do not enforce minimum alignment in memalign().
533  - Check for NULL pointer in malloc_usable_size().
534  - Fix an off-by-one heap profile statistics bug that could be observed in
535    interval- and growth-triggered heap profiles.
536  - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
537    is 0.
538  - Fix bin->runcur management to fix a layout policy bug.  This bug did not
539    affect correctness.
540  - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
541    initialized than necessary.
542  - Add missing "opt.lg_tcache_max" mallctl implementation.
543  - Use glibc allocator hooks to make mixed allocator usage less likely.
544  - Fix build issues for --disable-tcache.
545  - Don't mangle pthread_create() when --with-private-namespace is specified.
546
547* 2.2.5 (November 14, 2011)
548
549  Bug fixes:
550  - Fix huge_ralloc() race when using mremap(2).  This is a serious bug that
551    could cause memory corruption and/or crashes.
552  - Fix huge_ralloc() to maintain chunk statistics.
553  - Fix malloc_stats_print(..., "a") output.
554
555* 2.2.4 (November 5, 2011)
556
557  Bug fixes:
558  - Initialize arenas_tsd before using it.  This bug existed for 2.2.[0-3], as
559    well as for --disable-tls builds in earlier releases.
560  - Do not assume a 4 KiB page size in test/rallocm.c.
561
562* 2.2.3 (August 31, 2011)
563
564  This version fixes numerous bugs related to heap profiling.
565
566  Bug fixes:
567  - Fix a prof-related race condition.  This bug could cause memory corruption,
568    but only occurred in non-default configurations (prof_accum:false).
569  - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
570    excluded from backtraces).
571  - Fix a prof-related bug in realloc() (only triggered by OOM errors).
572  - Fix prof-related bugs in allocm() and rallocm().
573  - Fix prof_tdata_cleanup() for --disable-tls builds.
574  - Fix a relative include path, to fix objdir builds.
575
576* 2.2.2 (July 30, 2011)
577
578  Bug fixes:
579  - Fix a build error for --disable-tcache.
580  - Fix assertions in arena_purge() (for real this time).
581  - Add the --with-private-namespace option.  This is a workaround for symbol
582    conflicts that can inadvertently arise when using static libraries.
583
584* 2.2.1 (March 30, 2011)
585
586  Bug fixes:
587  - Implement atomic operations for x86/x64.  This fixes compilation failures
588    for versions of gcc that are still in wide use.
589  - Fix an assertion in arena_purge().
590
591* 2.2.0 (March 22, 2011)
592
593  This version incorporates several improvements to algorithms and data
594  structures that tend to reduce fragmentation and increase speed.
595
596  New features:
597  - Add the "stats.cactive" mallctl.
598  - Update pprof (from google-perftools 1.7).
599  - Improve backtracing-related configuration logic, and add the
600    --disable-prof-libgcc option.
601
602  Bug fixes:
603  - Change default symbol visibility from "internal", to "hidden", which
604    decreases the overhead of library-internal function calls.
605  - Fix symbol visibility so that it is also set on OS X.
606  - Fix a build dependency regression caused by the introduction of the .pic.o
607    suffix for PIC object files.
608  - Add missing checks for mutex initialization failures.
609  - Don't use libgcc-based backtracing except on x64, where it is known to work.
610  - Fix deadlocks on OS X that were due to memory allocation in
611    pthread_mutex_lock().
612  - Heap profiling-specific fixes:
613    + Fix memory corruption due to integer overflow in small region index
614      computation, when using a small enough sample interval that profiling
615      context pointers are stored in small run headers.
616    + Fix a bootstrap ordering bug that only occurred with TLS disabled.
617    + Fix a rallocm() rsize bug.
618    + Fix error detection bugs for aligned memory allocation.
619
620* 2.1.3 (March 14, 2011)
621
622  Bug fixes:
623  - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
624    for OS X in 2.1.2).
625  - Fix a "thread.arena" mallctl bug.
626  - Fix a thread cache stats merging bug.
627
628* 2.1.2 (March 2, 2011)
629
630  Bug fixes:
631  - Fix "thread.{de,}allocatedp" mallctl for OS X.
632  - Add missing jemalloc.a to build system.
633
634* 2.1.1 (January 31, 2011)
635
636  Bug fixes:
637  - Fix aligned huge reallocation (affected allocm()).
638  - Fix the ALLOCM_LG_ALIGN macro definition.
639  - Fix a heap dumping deadlock.
640  - Fix a "thread.arena" mallctl bug.
641
642* 2.1.0 (December 3, 2010)
643
644  This version incorporates some optimizations that can't quite be considered
645  bug fixes.
646
647  New features:
648  - Use Linux's mremap(2) for huge object reallocation when possible.
649  - Avoid locking in mallctl*() when possible.
650  - Add the "thread.[de]allocatedp" mallctl's.
651  - Convert the manual page source from roff to DocBook, and generate both roff
652    and HTML manuals.
653
654  Bug fixes:
655  - Fix a crash due to incorrect bootstrap ordering.  This only impacted
656    --enable-debug --enable-dss configurations.
657  - Fix a minor statistics bug for mallctl("swap.avail", ...).
658
659* 2.0.1 (October 29, 2010)
660
661  Bug fixes:
662  - Fix a race condition in heap profiling that could cause undefined behavior
663    if "opt.prof_accum" were disabled.
664  - Add missing mutex unlocks for some OOM error paths in the heap profiling
665    code.
666  - Fix a compilation error for non-C99 builds.
667
668* 2.0.0 (October 24, 2010)
669
670  This version focuses on the experimental *allocm() API, and on improved
671  run-time configuration/introspection.  Nonetheless, numerous performance
672  improvements are also included.
673
674  New features:
675  - Implement the experimental {,r,s,d}allocm() API, which provides a superset
676    of the functionality available via malloc(), calloc(), posix_memalign(),
677    realloc(), malloc_usable_size(), and free().  These functions can be used to
678    allocate/reallocate aligned zeroed memory, ask for optional extra memory
679    during reallocation, prevent object movement during reallocation, etc.
680  - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
681    more human-readable, and more flexible.  For example:
682      JEMALLOC_OPTIONS=AJP
683    is now:
684      MALLOC_CONF=abort:true,fill:true,stats_print:true
685  - Port to Apple OS X.  Sponsored by Mozilla.
686  - Make it possible for the application to control thread-->arena mappings via
687    the "thread.arena" mallctl.
688  - Add compile-time support for all TLS-related functionality via pthreads TSD.
689    This is mainly of interest for OS X, which does not support TLS, but has a
690    TSD implementation with similar performance.
691  - Override memalign() and valloc() if they are provided by the system.
692  - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
693    dirty unused pages.
694  - Make cumulative heap profiling data optional, so that it is possible to
695    limit the amount of memory consumed by heap profiling data structures.
696  - Add per thread allocation counters that can be accessed via the
697    "thread.allocated" and "thread.deallocated" mallctls.
698
699  Incompatible changes:
700  - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
701  - Increase default backtrace depth from 4 to 128 for heap profiling.
702  - Disable interval-based profile dumps by default.
703
704  Bug fixes:
705  - Remove bad assertions in fork handler functions.  These assertions could
706    cause aborts for some combinations of configure settings.
707  - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
708  - Fix leak context reporting.  This bug tended to cause the number of contexts
709    to be underreported (though the reported number of objects and bytes were
710    correct).
711  - Fix a realloc() bug for large in-place growing reallocation.  This bug could
712    cause memory corruption, but it was hard to trigger.
713  - Fix an allocation bug for small allocations that could be triggered if
714    multiple threads raced to create a new run of backing pages.
715  - Enhance the heap profiler to trigger samples based on usable size, rather
716    than request size.
717  - Fix a heap profiling bug due to sometimes losing track of requested object
718    size for sampled objects.
719
720* 1.0.3 (August 12, 2010)
721
722  Bug fixes:
723  - Fix the libunwind-based implementation of stack backtracing (used for heap
724    profiling).  This bug could cause zero-length backtraces to be reported.
725  - Add a missing mutex unlock in library initialization code.  If multiple
726    threads raced to initialize malloc, some of them could end up permanently
727    blocked.
728
729* 1.0.2 (May 11, 2010)
730
731  Bug fixes:
732  - Fix junk filling of large objects, which could cause memory corruption.
733  - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
734    memory limits could cause swap file configuration to fail.  Contributed by
735    Jordan DeLong.
736
737* 1.0.1 (April 14, 2010)
738
739  Bug fixes:
740  - Fix compilation when --enable-fill is specified.
741  - Fix threads-related profiling bugs that affected accuracy and caused memory
742    to be leaked during thread exit.
743  - Fix dirty page purging race conditions that could cause crashes.
744  - Fix crash in tcache flushing code during thread destruction.
745
746* 1.0.0 (April 11, 2010)
747
748  This release focuses on speed and run-time introspection.  Numerous
749  algorithmic improvements make this release substantially faster than its
750  predecessors.
751
752  New features:
753  - Implement autoconf-based configuration system.
754  - Add mallctl*(), for the purposes of introspection and run-time
755    configuration.
756  - Make it possible for the application to manually flush a thread's cache, via
757    the "tcache.flush" mallctl.
758  - Base maximum dirty page count on proportion of active memory.
759  - Compute various additional run-time statistics, including per size class
760    statistics for large objects.
761  - Expose malloc_stats_print(), which can be called repeatedly by the
762    application.
763  - Simplify the malloc_message() signature to only take one string argument,
764    and incorporate an opaque data pointer argument for use by the application
765    in combination with malloc_stats_print().
766  - Add support for allocation backed by one or more swap files, and allow the
767    application to disable over-commit if swap files are in use.
768  - Implement allocation profiling and leak checking.
769
770  Removed features:
771  - Remove the dynamic arena rebalancing code, since thread-specific caching
772    reduces its utility.
773
774  Bug fixes:
775  - Modify chunk allocation to work when address space layout randomization
776    (ASLR) is in use.
777  - Fix thread cleanup bugs related to TLS destruction.
778  - Handle 0-size allocation requests in posix_memalign().
779  - Fix a chunk leak.  The leaked chunks were never touched, so this impacted
780    virtual memory usage, but not physical memory usage.
781
782* linux_2008082[78]a (August 27/28, 2008)
783
784  These snapshot releases are the simple result of incorporating Linux-specific
785  support into the FreeBSD malloc sources.
786
787--------------------------------------------------------------------------------
788vim:filetype=text:textwidth=80
789