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