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