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* 2.2.5 (November 14, 2011) 10 11 Bug fixes: 12 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that 13 could cause memory corruption and/or crashes. 14 - Fix huge_ralloc() to maintain chunk statistics. 15 - Fix malloc_stats_print(..., "a") output. 16 17* 2.2.4 (November 5, 2011) 18 19 Bug fixes: 20 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as 21 well as for --disable-tls builds in earlier releases. 22 - Do not assume a 4 KiB page size in test/rallocm.c. 23 24* 2.2.3 (August 31, 2011) 25 26 This version fixes numerous bugs related to heap profiling. 27 28 Bug fixes: 29 - Fix a prof-related race condition. This bug could cause memory corruption, 30 but only occurred in non-default configurations (prof_accum:false). 31 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is 32 excluded from backtraces). 33 - Fix a prof-related bug in realloc() (only triggered by OOM errors). 34 - Fix prof-related bugs in allocm() and rallocm(). 35 - Fix prof_tdata_cleanup() for --disable-tls builds. 36 - Fix a relative include path, to fix objdir builds. 37 38* 2.2.2 (July 30, 2011) 39 40 Bug fixes: 41 - Fix a build error for --disable-tcache. 42 - Fix assertions in arena_purge() (for real this time). 43 - Add the --with-private-namespace option. This is a workaround for symbol 44 conflicts that can inadvertently arise when using static libraries. 45 46* 2.2.1 (March 30, 2011) 47 48 Bug fixes: 49 - Implement atomic operations for x86/x64. This fixes compilation failures 50 for versions of gcc that are still in wide use. 51 - Fix an assertion in arena_purge(). 52 53* 2.2.0 (March 22, 2011) 54 55 This version incorporates several improvements to algorithms and data 56 structures that tend to reduce fragmentation and increase speed. 57 58 New features: 59 - Add the "stats.cactive" mallctl. 60 - Update pprof (from google-perftools 1.7). 61 - Improve backtracing-related configuration logic, and add the 62 --disable-prof-libgcc option. 63 64 Bug fixes: 65 - Change default symbol visibility from "internal", to "hidden", which 66 decreases the overhead of library-internal function calls. 67 - Fix symbol visibility so that it is also set on OS X. 68 - Fix a build dependency regression caused by the introduction of the .pic.o 69 suffix for PIC object files. 70 - Add missing checks for mutex initialization failures. 71 - Don't use libgcc-based backtracing except on x64, where it is known to work. 72 - Fix deadlocks on OS X that were due to memory allocation in 73 pthread_mutex_lock(). 74 - Heap profiling-specific fixes: 75 + Fix memory corruption due to integer overflow in small region index 76 computation, when using a small enough sample interval that profiling 77 context pointers are stored in small run headers. 78 + Fix a bootstrap ordering bug that only occurred with TLS disabled. 79 + Fix a rallocm() rsize bug. 80 + Fix error detection bugs for aligned memory allocation. 81 82* 2.1.3 (March 14, 2011) 83 84 Bug fixes: 85 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix 86 for OS X in 2.1.2). 87 - Fix a "thread.arena" mallctl bug. 88 - Fix a thread cache stats merging bug. 89 90* 2.1.2 (March 2, 2011) 91 92 Bug fixes: 93 - Fix "thread.{de,}allocatedp" mallctl for OS X. 94 - Add missing jemalloc.a to build system. 95 96* 2.1.1 (January 31, 2011) 97 98 Bug fixes: 99 - Fix aligned huge reallocation (affected allocm()). 100 - Fix the ALLOCM_LG_ALIGN macro definition. 101 - Fix a heap dumping deadlock. 102 - Fix a "thread.arena" mallctl bug. 103 104* 2.1.0 (December 3, 2010) 105 106 This version incorporates some optimizations that can't quite be considered 107 bug fixes. 108 109 New features: 110 - Use Linux's mremap(2) for huge object reallocation when possible. 111 - Avoid locking in mallctl*() when possible. 112 - Add the "thread.[de]allocatedp" mallctl's. 113 - Convert the manual page source from roff to DocBook, and generate both roff 114 and HTML manuals. 115 116 Bug fixes: 117 - Fix a crash due to incorrect bootstrap ordering. This only impacted 118 --enable-debug --enable-dss configurations. 119 - Fix a minor statistics bug for mallctl("swap.avail", ...). 120 121* 2.0.1 (October 29, 2010) 122 123 Bug fixes: 124 - Fix a race condition in heap profiling that could cause undefined behavior 125 if "opt.prof_accum" were disabled. 126 - Add missing mutex unlocks for some OOM error paths in the heap profiling 127 code. 128 - Fix a compilation error for non-C99 builds. 129 130* 2.0.0 (October 24, 2010) 131 132 This version focuses on the experimental *allocm() API, and on improved 133 run-time configuration/introspection. Nonetheless, numerous performance 134 improvements are also included. 135 136 New features: 137 - Implement the experimental {,r,s,d}allocm() API, which provides a superset 138 of the functionality available via malloc(), calloc(), posix_memalign(), 139 realloc(), malloc_usable_size(), and free(). These functions can be used to 140 allocate/reallocate aligned zeroed memory, ask for optional extra memory 141 during reallocation, prevent object movement during reallocation, etc. 142 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is 143 more human-readable, and more flexible. For example: 144 JEMALLOC_OPTIONS=AJP 145 is now: 146 MALLOC_CONF=abort:true,fill:true,stats_print:true 147 - Port to Apple OS X. Sponsored by Mozilla. 148 - Make it possible for the application to control thread-->arena mappings via 149 the "thread.arena" mallctl. 150 - Add compile-time support for all TLS-related functionality via pthreads TSD. 151 This is mainly of interest for OS X, which does not support TLS, but has a 152 TSD implementation with similar performance. 153 - Override memalign() and valloc() if they are provided by the system. 154 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all 155 dirty unused pages. 156 - Make cumulative heap profiling data optional, so that it is possible to 157 limit the amount of memory consumed by heap profiling data structures. 158 - Add per thread allocation counters that can be accessed via the 159 "thread.allocated" and "thread.deallocated" mallctls. 160 161 Incompatible changes: 162 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above). 163 - Increase default backtrace depth from 4 to 128 for heap profiling. 164 - Disable interval-based profile dumps by default. 165 166 Bug fixes: 167 - Remove bad assertions in fork handler functions. These assertions could 168 cause aborts for some combinations of configure settings. 169 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc. 170 - Fix leak context reporting. This bug tended to cause the number of contexts 171 to be underreported (though the reported number of objects and bytes were 172 correct). 173 - Fix a realloc() bug for large in-place growing reallocation. This bug could 174 cause memory corruption, but it was hard to trigger. 175 - Fix an allocation bug for small allocations that could be triggered if 176 multiple threads raced to create a new run of backing pages. 177 - Enhance the heap profiler to trigger samples based on usable size, rather 178 than request size. 179 - Fix a heap profiling bug due to sometimes losing track of requested object 180 size for sampled objects. 181 182* 1.0.3 (August 12, 2010) 183 184 Bug fixes: 185 - Fix the libunwind-based implementation of stack backtracing (used for heap 186 profiling). This bug could cause zero-length backtraces to be reported. 187 - Add a missing mutex unlock in library initialization code. If multiple 188 threads raced to initialize malloc, some of them could end up permanently 189 blocked. 190 191* 1.0.2 (May 11, 2010) 192 193 Bug fixes: 194 - Fix junk filling of large objects, which could cause memory corruption. 195 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual 196 memory limits could cause swap file configuration to fail. Contributed by 197 Jordan DeLong. 198 199* 1.0.1 (April 14, 2010) 200 201 Bug fixes: 202 - Fix compilation when --enable-fill is specified. 203 - Fix threads-related profiling bugs that affected accuracy and caused memory 204 to be leaked during thread exit. 205 - Fix dirty page purging race conditions that could cause crashes. 206 - Fix crash in tcache flushing code during thread destruction. 207 208* 1.0.0 (April 11, 2010) 209 210 This release focuses on speed and run-time introspection. Numerous 211 algorithmic improvements make this release substantially faster than its 212 predecessors. 213 214 New features: 215 - Implement autoconf-based configuration system. 216 - Add mallctl*(), for the purposes of introspection and run-time 217 configuration. 218 - Make it possible for the application to manually flush a thread's cache, via 219 the "tcache.flush" mallctl. 220 - Base maximum dirty page count on proportion of active memory. 221 - Compute various addtional run-time statistics, including per size class 222 statistics for large objects. 223 - Expose malloc_stats_print(), which can be called repeatedly by the 224 application. 225 - Simplify the malloc_message() signature to only take one string argument, 226 and incorporate an opaque data pointer argument for use by the application 227 in combination with malloc_stats_print(). 228 - Add support for allocation backed by one or more swap files, and allow the 229 application to disable over-commit if swap files are in use. 230 - Implement allocation profiling and leak checking. 231 232 Removed features: 233 - Remove the dynamic arena rebalancing code, since thread-specific caching 234 reduces its utility. 235 236 Bug fixes: 237 - Modify chunk allocation to work when address space layout randomization 238 (ASLR) is in use. 239 - Fix thread cleanup bugs related to TLS destruction. 240 - Handle 0-size allocation requests in posix_memalign(). 241 - Fix a chunk leak. The leaked chunks were never touched, so this impacted 242 virtual memory usage, but not physical memory usage. 243 244* linux_2008082[78]a (August 27/28, 2008) 245 246 These snapshot releases are the simple result of incorporating Linux-specific 247 support into the FreeBSD malloc sources. 248 249-------------------------------------------------------------------------------- 250vim:filetype=text:textwidth=80 251