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