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 * 3.6.0 (March 31, 2014) 9 10 This version contains a critical bug fix for a regression present in 3.5.0 and 11 3.5.1. 12 13 Bug fixes: 14 - Fix a regression in arena_chunk_alloc() that caused crashes during 15 small/large allocation if chunk allocation failed. In the absence of this 16 bug, chunk allocation failure would result in allocation failure, e.g. NULL 17 return from malloc(). This regression was introduced in 3.5.0. 18 - Fix backtracing for gcc intrinsics-based backtracing by specifying 19 -fno-omit-frame-pointer to gcc. Note that the application (and all the 20 libraries it links to) must also be compiled with this option for 21 backtracing to be reliable. 22 - Use dss allocation precedence for huge allocations as well as small/large 23 allocations. 24 - Fix test assertion failure message formatting. This bug did not manifect on 25 x86_64 systems because of implementation subtleties in va_list. 26 - Fix inconsequential test failures for hash and SFMT code. 27 28 New features: 29 - Support heap profiling on FreeBSD. This feature depends on the proc 30 filesystem being mounted during heap profile dumping. 31 32 * 3.5.1 (February 25, 2014) 33 34 This version primarily addresses minor bugs in test code. 35 36 Bug fixes: 37 - Configure Solaris/Illumos to use MADV_FREE. 38 - Fix junk filling for mremap(2)-based huge reallocation. This is only 39 relevant if configuring with the --enable-mremap option specified. 40 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the 41 compiler. 42 - Add a configure test for SSE2 rather than assuming it is usable on i686 43 systems. This fixes test compilation errors, especially on 32-bit Linux 44 systems. 45 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit 46 test. 47 - Fix/remove flawed alignment-related overflow tests. 48 - Prevent compiler optimizations that could change backtraces in the 49 prof_accum unit test. 50 51 * 3.5.0 (January 22, 2014) 52 53 This version focuses on refactoring and automated testing, though it also 54 includes some non-trivial heap profiling optimizations not mentioned below. 55 56 New features: 57 - Add the *allocx() API, which is a successor to the experimental *allocm() 58 API. The *allocx() functions are slightly simpler to use because they have 59 fewer parameters, they directly return the results of primary interest, and 60 mallocx()/rallocx() avoid the strict aliasing pitfall that 61 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is 62 slated for removal in the next non-bugfix release. 63 - Add support for LinuxThreads. 64 65 Bug fixes: 66 - Unless heap profiling is enabled, disable floating point code and don't link 67 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64 68 systems, makes it possible to completely disable floating point register 69 use. Some versions of glibc neglect to save/restore caller-saved floating 70 point registers during dynamic lazy symbol loading, and the symbol loading 71 code uses whatever malloc the application happens to have linked/loaded 72 with, the result being potential floating point register corruption. 73 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling 74 backtrace creation in imemalign(). This bug impacted posix_memalign() and 75 aligned_alloc(). 76 - Fix a file descriptor leak in a prof_dump_maps() error path. 77 - Fix prof_dump() to close the dump file descriptor for all relevant error 78 paths. 79 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for 80 allocation, not just deallocation. 81 - Fix a data race for large allocation stats counters. 82 - Fix a potential infinite loop during thread exit. This bug occurred on 83 Solaris, and could affect other platforms with similar pthreads TSD 84 implementations. 85 - Don't junk-fill reallocations unless usable size changes. This fixes a 86 violation of the *allocx()/*allocm() semantics. 87 - Fix growing large reallocation to junk fill new space. 88 - Fix huge deallocation to junk fill when munmap is disabled. 89 - Change the default private namespace prefix from empty to je_, and change 90 --with-private-namespace-prefix so that it prepends an additional prefix 91 rather than replacing je_. This reduces the likelihood of applications 92 which statically link jemalloc experiencing symbol name collisions. 93 - Add missing private namespace mangling (relevant when 94 --with-private-namespace is specified). 95 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as 96 static even for debug builds. 97 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice 98 this error path is never executed. 99 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These 100 bugs had no impact except for malformed inputs. 101 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by 102 existing calls, so they had no impact. 103 104 * 3.4.1 (October 20, 2013) 105 106 Bug fixes: 107 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption 108 of internal data structures and subsequent crashes. 109 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of 110 uninitialized memory in: 111 + arena chunk headers 112 + internal zero-initialized data structures (relevant to tcache and prof 113 code) 114 - Preserve errno during the first allocation. A readlink(2) call during 115 initialization fails unless /etc/malloc.conf exists, so errno was typically 116 set during the first allocation prior to this fix. 117 - Fix compilation warnings reported by gcc 4.8.1. 118 119 * 3.4.0 (June 2, 2013) 120 121 This version is essentially a small bugfix release, but the addition of 122 aarch64 support requires that the minor version be incremented. 123 124 Bug fixes: 125 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were 126 typically triggered by multiple threads concurrently deallocating huge 127 objects. 128 129 New features: 130 - Add support for the aarch64 architecture. 131 132 * 3.3.1 (March 6, 2013) 133 134 This version fixes bugs that are typically encountered only when utilizing 135 custom run-time options. 136 137 Bug fixes: 138 - Fix a locking order bug that could cause deadlock during fork if heap 139 profiling were enabled. 140 - Fix a chunk recycling bug that could cause the allocator to lose track of 141 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause 142 corruption if allocating via sbrk(2) (unlikely unless running with the 143 "dss:primary" option specified). This was completely harmless on Linux 144 unless using mlockall(2) (and unlikely even then, unless the 145 --disable-munmap configure option or the "dss:primary" option was 146 specified). This regression was introduced in 3.1.0 by the 147 mlockall(2)/madvise(2) interaction fix. 148 - Fix TLS-related memory corruption that could occur during thread exit if the 149 thread never allocated memory. Only the quarantine and prof facilities were 150 susceptible. 151 - Fix two quarantine bugs: 152 + Internal reallocation of the quarantined object array leaked the old 153 array. 154 + Reallocation failure for internal reallocation of the quarantined object 155 array (very unlikely) resulted in memory corruption. 156 - Fix Valgrind integration to annotate all internally allocated memory in a 157 way that keeps Valgrind happy about internal data structure access. 158 - Fix building for s390 systems. 159 160 * 3.3.0 (January 23, 2013) 161 162 This version includes a few minor performance improvements in addition to the 163 listed new features and bug fixes. 164 165 New features: 166 - Add clipping support to lg_chunk option processing. 167 - Add the --enable-ivsalloc option. 168 - Add the --without-export option. 169 - Add the --disable-zone-allocator option. 170 171 Bug fixes: 172 - Fix "arenas.extend" mallctl to output the number of arenas. 173 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory 174 is undefined. 175 - Fix build break on FreeBSD related to alloca.h. 176 177 * 3.2.0 (November 9, 2012) 178 179 In addition to a couple of bug fixes, this version modifies page run 180 allocation and dirty page purging algorithms in order to better control 181 page-level virtual memory fragmentation. 182 183 Incompatible changes: 184 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1). 185 186 Bug fixes: 187 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only 188 after primary dss allocation fails. 189 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced 190 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl. 191 192 * 3.1.0 (October 16, 2012) 193 194 New features: 195 - Auto-detect whether running inside Valgrind, thus removing the need to 196 manually specify MALLOC_CONF=valgrind:true. 197 - Add the "arenas.extend" mallctl, which allows applications to create 198 manually managed arenas. 199 - Add the ALLOCM_ARENA() flag for {,r,d}allocm(). 200 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls, 201 which provide control over dss/mmap precedence. 202 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge". 203 - Define LG_QUANTUM for hppa. 204 205 Incompatible changes: 206 - Disable tcache by default if running inside Valgrind, in order to avoid 207 making unallocated objects appear reachable to Valgrind. 208 - Drop const from malloc_usable_size() argument on Linux. 209 210 Bug fixes: 211 - Fix heap profiling crash if sampled object is freed via realloc(p, 0). 212 - Remove const from __*_hook variable declarations, so that glibc can modify 213 them during process forking. 214 - Fix mlockall(2)/madvise(2) interaction. 215 - Fix fork(2)-related deadlocks. 216 - Fix error return value for "thread.tcache.enabled" mallctl. 217 218 * 3.0.0 (May 11, 2012) 219 220 Although this version adds some major new features, the primary focus is on 221 internal code cleanup that facilitates maintainability and portability, most 222 of which is not reflected in the ChangeLog. This is the first release to 223 incorporate substantial contributions from numerous other developers, and the 224 result is a more broadly useful allocator (see the git revision history for 225 contribution details). Note that the license has been unified, thanks to 226 Facebook granting a license under the same terms as the other copyright 227 holders (see COPYING). 228 229 New features: 230 - Implement Valgrind support, redzones, and quarantine. 231 - Add support for additional platforms: 232 + FreeBSD 233 + Mac OS X Lion 234 + MinGW 235 + Windows (no support yet for replacing the system malloc) 236 - Add support for additional architectures: 237 + MIPS 238 + SH4 239 + Tilera 240 - Add support for cross compiling. 241 - Add nallocm(), which rounds a request size up to the nearest size class 242 without actually allocating. 243 - Implement aligned_alloc() (blame C11). 244 - Add the "thread.tcache.enabled" mallctl. 245 - Add the "opt.prof_final" mallctl. 246 - Update pprof (from gperftools 2.0). 247 - Add the --with-mangling option. 248 - Add the --disable-experimental option. 249 - Add the --disable-munmap option, and make it the default on Linux. 250 - Add the --enable-mremap option, which disables use of mremap(2) by default. 251 252 Incompatible changes: 253 - Enable stats by default. 254 - Enable fill by default. 255 - Disable lazy locking by default. 256 - Rename the "tcache.flush" mallctl to "thread.tcache.flush". 257 - Rename the "arenas.pagesize" mallctl to "arenas.page". 258 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). 259 - Change the "opt.prof_accum" default from true to false. 260 261 Removed features: 262 - Remove the swap feature, including the "config.swap", "swap.avail", 263 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls. 264 - Remove highruns statistics, including the 265 "stats.arenas.<i>.bins.<j>.highruns" and 266 "stats.arenas.<i>.lruns.<j>.highruns" mallctls. 267 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max", 268 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and 269 "arenas.[tqcs]bins" mallctls. 270 - Remove the "arenas.chunksize" mallctl. 271 - Remove the "opt.lg_prof_tcmax" option. 272 - Remove the "opt.lg_prof_bt_max" option. 273 - Remove the "opt.lg_tcache_gc_sweep" option. 274 - Remove the --disable-tiny option, including the "config.tiny" mallctl. 275 - Remove the --enable-dynamic-page-shift configure option. 276 - Remove the --enable-sysv configure option. 277 278 Bug fixes: 279 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause 280 invalid statistics and crashes. 281 - Work around TLS deallocation via free() on Linux. This bug could cause 282 write-after-free memory corruption. 283 - Fix a potential deadlock that could occur during interval- and 284 growth-triggered heap profile dumps. 285 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags. 286 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could 287 cause memory corruption and crashes with --enable-dss specified. 288 - Fix fork-related bugs that could cause deadlock in children between fork 289 and exec. 290 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter. 291 - Fix realloc(p, 0) to act like free(p). 292 - Do not enforce minimum alignment in memalign(). 293 - Check for NULL pointer in malloc_usable_size(). 294 - Fix an off-by-one heap profile statistics bug that could be observed in 295 interval- and growth-triggered heap profiles. 296 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch 297 is 0. 298 - Fix bin->runcur management to fix a layout policy bug. This bug did not 299 affect correctness. 300 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be 301 initialized than necessary. 302 - Add missing "opt.lg_tcache_max" mallctl implementation. 303 - Use glibc allocator hooks to make mixed allocator usage less likely. 304 - Fix build issues for --disable-tcache. 305 - Don't mangle pthread_create() when --with-private-namespace is specified. 306 307 * 2.2.5 (November 14, 2011) 308 309 Bug fixes: 310 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that 311 could cause memory corruption and/or crashes. 312 - Fix huge_ralloc() to maintain chunk statistics. 313 - Fix malloc_stats_print(..., "a") output. 314 315 * 2.2.4 (November 5, 2011) 316 317 Bug fixes: 318 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as 319 well as for --disable-tls builds in earlier releases. 320 - Do not assume a 4 KiB page size in test/rallocm.c. 321 322 * 2.2.3 (August 31, 2011) 323 324 This version fixes numerous bugs related to heap profiling. 325 326 Bug fixes: 327 - Fix a prof-related race condition. This bug could cause memory corruption, 328 but only occurred in non-default configurations (prof_accum:false). 329 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is 330 excluded from backtraces). 331 - Fix a prof-related bug in realloc() (only triggered by OOM errors). 332 - Fix prof-related bugs in allocm() and rallocm(). 333 - Fix prof_tdata_cleanup() for --disable-tls builds. 334 - Fix a relative include path, to fix objdir builds. 335 336 * 2.2.2 (July 30, 2011) 337 338 Bug fixes: 339 - Fix a build error for --disable-tcache. 340 - Fix assertions in arena_purge() (for real this time). 341 - Add the --with-private-namespace option. This is a workaround for symbol 342 conflicts that can inadvertently arise when using static libraries. 343 344 * 2.2.1 (March 30, 2011) 345 346 Bug fixes: 347 - Implement atomic operations for x86/x64. This fixes compilation failures 348 for versions of gcc that are still in wide use. 349 - Fix an assertion in arena_purge(). 350 351 * 2.2.0 (March 22, 2011) 352 353 This version incorporates several improvements to algorithms and data 354 structures that tend to reduce fragmentation and increase speed. 355 356 New features: 357 - Add the "stats.cactive" mallctl. 358 - Update pprof (from google-perftools 1.7). 359 - Improve backtracing-related configuration logic, and add the 360 --disable-prof-libgcc option. 361 362 Bug fixes: 363 - Change default symbol visibility from "internal", to "hidden", which 364 decreases the overhead of library-internal function calls. 365 - Fix symbol visibility so that it is also set on OS X. 366 - Fix a build dependency regression caused by the introduction of the .pic.o 367 suffix for PIC object files. 368 - Add missing checks for mutex initialization failures. 369 - Don't use libgcc-based backtracing except on x64, where it is known to work. 370 - Fix deadlocks on OS X that were due to memory allocation in 371 pthread_mutex_lock(). 372 - Heap profiling-specific fixes: 373 + Fix memory corruption due to integer overflow in small region index 374 computation, when using a small enough sample interval that profiling 375 context pointers are stored in small run headers. 376 + Fix a bootstrap ordering bug that only occurred with TLS disabled. 377 + Fix a rallocm() rsize bug. 378 + Fix error detection bugs for aligned memory allocation. 379 380 * 2.1.3 (March 14, 2011) 381 382 Bug fixes: 383 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix 384 for OS X in 2.1.2). 385 - Fix a "thread.arena" mallctl bug. 386 - Fix a thread cache stats merging bug. 387 388 * 2.1.2 (March 2, 2011) 389 390 Bug fixes: 391 - Fix "thread.{de,}allocatedp" mallctl for OS X. 392 - Add missing jemalloc.a to build system. 393 394 * 2.1.1 (January 31, 2011) 395 396 Bug fixes: 397 - Fix aligned huge reallocation (affected allocm()). 398 - Fix the ALLOCM_LG_ALIGN macro definition. 399 - Fix a heap dumping deadlock. 400 - Fix a "thread.arena" mallctl bug. 401 402 * 2.1.0 (December 3, 2010) 403 404 This version incorporates some optimizations that can't quite be considered 405 bug fixes. 406 407 New features: 408 - Use Linux's mremap(2) for huge object reallocation when possible. 409 - Avoid locking in mallctl*() when possible. 410 - Add the "thread.[de]allocatedp" mallctl's. 411 - Convert the manual page source from roff to DocBook, and generate both roff 412 and HTML manuals. 413 414 Bug fixes: 415 - Fix a crash due to incorrect bootstrap ordering. This only impacted 416 --enable-debug --enable-dss configurations. 417 - Fix a minor statistics bug for mallctl("swap.avail", ...). 418 419 * 2.0.1 (October 29, 2010) 420 421 Bug fixes: 422 - Fix a race condition in heap profiling that could cause undefined behavior 423 if "opt.prof_accum" were disabled. 424 - Add missing mutex unlocks for some OOM error paths in the heap profiling 425 code. 426 - Fix a compilation error for non-C99 builds. 427 428 * 2.0.0 (October 24, 2010) 429 430 This version focuses on the experimental *allocm() API, and on improved 431 run-time configuration/introspection. Nonetheless, numerous performance 432 improvements are also included. 433 434 New features: 435 - Implement the experimental {,r,s,d}allocm() API, which provides a superset 436 of the functionality available via malloc(), calloc(), posix_memalign(), 437 realloc(), malloc_usable_size(), and free(). These functions can be used to 438 allocate/reallocate aligned zeroed memory, ask for optional extra memory 439 during reallocation, prevent object movement during reallocation, etc. 440 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is 441 more human-readable, and more flexible. For example: 442 JEMALLOC_OPTIONS=AJP 443 is now: 444 MALLOC_CONF=abort:true,fill:true,stats_print:true 445 - Port to Apple OS X. Sponsored by Mozilla. 446 - Make it possible for the application to control thread-->arena mappings via 447 the "thread.arena" mallctl. 448 - Add compile-time support for all TLS-related functionality via pthreads TSD. 449 This is mainly of interest for OS X, which does not support TLS, but has a 450 TSD implementation with similar performance. 451 - Override memalign() and valloc() if they are provided by the system. 452 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all 453 dirty unused pages. 454 - Make cumulative heap profiling data optional, so that it is possible to 455 limit the amount of memory consumed by heap profiling data structures. 456 - Add per thread allocation counters that can be accessed via the 457 "thread.allocated" and "thread.deallocated" mallctls. 458 459 Incompatible changes: 460 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above). 461 - Increase default backtrace depth from 4 to 128 for heap profiling. 462 - Disable interval-based profile dumps by default. 463 464 Bug fixes: 465 - Remove bad assertions in fork handler functions. These assertions could 466 cause aborts for some combinations of configure settings. 467 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc. 468 - Fix leak context reporting. This bug tended to cause the number of contexts 469 to be underreported (though the reported number of objects and bytes were 470 correct). 471 - Fix a realloc() bug for large in-place growing reallocation. This bug could 472 cause memory corruption, but it was hard to trigger. 473 - Fix an allocation bug for small allocations that could be triggered if 474 multiple threads raced to create a new run of backing pages. 475 - Enhance the heap profiler to trigger samples based on usable size, rather 476 than request size. 477 - Fix a heap profiling bug due to sometimes losing track of requested object 478 size for sampled objects. 479 480 * 1.0.3 (August 12, 2010) 481 482 Bug fixes: 483 - Fix the libunwind-based implementation of stack backtracing (used for heap 484 profiling). This bug could cause zero-length backtraces to be reported. 485 - Add a missing mutex unlock in library initialization code. If multiple 486 threads raced to initialize malloc, some of them could end up permanently 487 blocked. 488 489 * 1.0.2 (May 11, 2010) 490 491 Bug fixes: 492 - Fix junk filling of large objects, which could cause memory corruption. 493 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual 494 memory limits could cause swap file configuration to fail. Contributed by 495 Jordan DeLong. 496 497 * 1.0.1 (April 14, 2010) 498 499 Bug fixes: 500 - Fix compilation when --enable-fill is specified. 501 - Fix threads-related profiling bugs that affected accuracy and caused memory 502 to be leaked during thread exit. 503 - Fix dirty page purging race conditions that could cause crashes. 504 - Fix crash in tcache flushing code during thread destruction. 505 506 * 1.0.0 (April 11, 2010) 507 508 This release focuses on speed and run-time introspection. Numerous 509 algorithmic improvements make this release substantially faster than its 510 predecessors. 511 512 New features: 513 - Implement autoconf-based configuration system. 514 - Add mallctl*(), for the purposes of introspection and run-time 515 configuration. 516 - Make it possible for the application to manually flush a thread's cache, via 517 the "tcache.flush" mallctl. 518 - Base maximum dirty page count on proportion of active memory. 519 - Compute various addtional run-time statistics, including per size class 520 statistics for large objects. 521 - Expose malloc_stats_print(), which can be called repeatedly by the 522 application. 523 - Simplify the malloc_message() signature to only take one string argument, 524 and incorporate an opaque data pointer argument for use by the application 525 in combination with malloc_stats_print(). 526 - Add support for allocation backed by one or more swap files, and allow the 527 application to disable over-commit if swap files are in use. 528 - Implement allocation profiling and leak checking. 529 530 Removed features: 531 - Remove the dynamic arena rebalancing code, since thread-specific caching 532 reduces its utility. 533 534 Bug fixes: 535 - Modify chunk allocation to work when address space layout randomization 536 (ASLR) is in use. 537 - Fix thread cleanup bugs related to TLS destruction. 538 - Handle 0-size allocation requests in posix_memalign(). 539 - Fix a chunk leak. The leaked chunks were never touched, so this impacted 540 virtual memory usage, but not physical memory usage. 541 542 * linux_2008082[78]a (August 27/28, 2008) 543 544 These snapshot releases are the simple result of incorporating Linux-specific 545 support into the FreeBSD malloc sources. 546 547 -------------------------------------------------------------------------------- 548 vim:filetype=text:textwidth=80 549
