Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 1 | Following are change highlights associated with official releases. Important |
Jason Evans | be09b81 | 2015-07-07 09:33:22 -0700 | [diff] [blame] | 2 | bug fixes are all mentioned, but some internal enhancements are omitted here for |
| 3 | brevity. Much more detail can be found in the git revision history: |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 4 | |
Jason Evans | b9ec5c9 | 2014-02-25 16:43:51 -0800 | [diff] [blame] | 5 | https://github.com/jemalloc/jemalloc |
| 6 | |
Jason Evans | 09f8585 | 2016-05-12 14:23:50 -0700 | [diff] [blame] | 7 | * 4.2.0 (May 12, 2016) |
Jason Evans | 62c217e | 2016-05-06 15:22:32 -0700 | [diff] [blame] | 8 | |
| 9 | New features: |
| 10 | - Add the arena.<i>.reset mallctl, which makes it possible to discard all of |
Jason Evans | dc7ff63 | 2016-05-12 15:06:50 -0700 | [diff] [blame] | 11 | an arena's allocations in a single operation. (@jasone) |
Jason Evans | 62c217e | 2016-05-06 15:22:32 -0700 | [diff] [blame] | 12 | - Add the stats.retained and stats.arenas.<i>.retained statistics. (@jasone) |
| 13 | - Add the --with-version configure option. (@jasone) |
| 14 | - Support --with-lg-page values larger than actual page size. (@jasone) |
| 15 | |
| 16 | Optimizations: |
| 17 | - Use pairing heaps rather than red-black trees for various hot data |
| 18 | structures. (@djwatson, @jasone) |
| 19 | - Streamline fast paths of rtree operations. (@jasone) |
| 20 | - Optimize the fast paths of calloc() and [m,d,sd]allocx(). (@jasone) |
| 21 | - Decommit unused virtual memory if the OS does not overcommit. (@jasone) |
| 22 | - Specify MAP_NORESERVE on Linux if [heuristic] overcommit is active, in order |
| 23 | to avoid unfortunate interactions during fork(2). (@jasone) |
| 24 | |
| 25 | Bug fixes: |
Jason Evans | 7790a0b | 2016-05-11 00:52:59 -0700 | [diff] [blame] | 26 | - Fix chunk accounting related to triggering gdump profiles. (@jasone) |
Jason Evans | 62c217e | 2016-05-06 15:22:32 -0700 | [diff] [blame] | 27 | - Link against librt for clock_gettime(2) if glibc < 2.17. (@jasone) |
| 28 | - Scale leak report summary according to sampling probability. (@jasone) |
| 29 | |
Jason Evans | 21cda0d | 2016-05-03 12:11:36 -0700 | [diff] [blame] | 30 | * 4.1.1 (May 3, 2016) |
| 31 | |
| 32 | This bugfix release resolves a variety of mostly minor issues, though the |
| 33 | bitmap fix is critical for 64-bit Windows. |
| 34 | |
| 35 | Bug fixes: |
| 36 | - Fix the linear scan version of bitmap_sfu() to shift by the proper amount |
| 37 | even when sizeof(long) is not the same as sizeof(void *), as on 64-bit |
| 38 | Windows. (@jasone) |
| 39 | - Fix hashing functions to avoid unaligned memory accesses (and resulting |
| 40 | crashes). This is relevant at least to some ARM-based platforms. |
| 41 | (@rkmisra) |
| 42 | - Fix fork()-related lock rank ordering reversals. These reversals were |
| 43 | unlikely to cause deadlocks in practice except when heap profiling was |
| 44 | enabled and active. (@jasone) |
| 45 | - Fix various chunk leaks in OOM code paths. (@jasone) |
| 46 | - Fix malloc_stats_print() to print opt.narenas correctly. (@jasone) |
Jason Evans | 7ba6e74 | 2016-05-03 17:46:07 -0700 | [diff] [blame] | 47 | - Fix MSVC-specific build/test issues. (@rustyx, @yuslepukhin) |
Jason Evans | 21cda0d | 2016-05-03 12:11:36 -0700 | [diff] [blame] | 48 | - Fix a variety of test failures that were due to test fragility rather than |
| 49 | core bugs. (@jasone) |
| 50 | |
Jason Evans | 3a34261 | 2016-02-28 14:52:17 -0800 | [diff] [blame] | 51 | * 4.1.0 (February 28, 2016) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 52 | |
| 53 | This release is primarily about optimizations, but it also incorporates a lot |
| 54 | of portability-motivated refactoring and enhancements. Many people worked on |
| 55 | this release, to an extent that even with the omission here of minor changes |
| 56 | (see git revision history), and of the people who reported and diagnosed |
| 57 | issues, so much of the work was contributed that starting with this release, |
| 58 | changes are annotated with author credits to help reflect the collaborative |
| 59 | effort involved. |
| 60 | |
| 61 | New features: |
| 62 | - Implement decay-based unused dirty page purging, a major optimization with |
| 63 | mallctl API impact. This is an alternative to the existing ratio-based |
| 64 | unused dirty page purging, and is intended to eventually become the sole |
| 65 | purging mechanism. New mallctls: |
| 66 | + opt.purge |
| 67 | + opt.decay_time |
| 68 | + arena.<i>.decay |
| 69 | + arena.<i>.decay_time |
| 70 | + arenas.decay_time |
| 71 | + stats.arenas.<i>.decay_time |
| 72 | (@jasone, @cevans87) |
| 73 | - Add --with-malloc-conf, which makes it possible to embed a default |
| 74 | options string during configuration. This was motivated by the desire to |
| 75 | specify --with-malloc-conf=purge:decay , since the default must remain |
| 76 | purge:ratio until the 5.0.0 release. (@jasone) |
Jason Evans | e025c51 | 2016-02-28 00:01:13 -0800 | [diff] [blame] | 77 | - Add MS Visual Studio 2015 support. (@rustyx, @yuslepukhin) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 78 | - Make *allocx() size class overflow behavior defined. The maximum |
| 79 | size class is now less than PTRDIFF_MAX to protect applications against |
| 80 | numerical overflow, and all allocation functions are guaranteed to indicate |
| 81 | errors rather than potentially crashing if the request size exceeds the |
| 82 | maximum size class. (@jasone) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 83 | - jeprof: |
| 84 | + Add raw heap profile support. (@jasone) |
| 85 | + Add --retain and --exclude for backtrace symbol filtering. (@jasone) |
| 86 | |
| 87 | Optimizations: |
| 88 | - Optimize the fast path to combine various bootstrapping and configuration |
| 89 | checks and execute more streamlined code in the common case. (@interwq) |
| 90 | - Use linear scan for small bitmaps (used for small object tracking). In |
| 91 | addition to speeding up bitmap operations on 64-bit systems, this reduces |
| 92 | allocator metadata overhead by approximately 0.2%. (@djwatson) |
| 93 | - Separate arena_avail trees, which substantially speeds up run tree |
| 94 | operations. (@djwatson) |
| 95 | - Use memoization (boot-time-computed table) for run quantization. Separate |
| 96 | arena_avail trees reduced the importance of this optimization. (@jasone) |
| 97 | - Attempt mmap-based in-place huge reallocation. This can dramatically speed |
| 98 | up incremental huge reallocation. (@jasone) |
| 99 | |
| 100 | Incompatible changes: |
| 101 | - Make opt.narenas unsigned rather than size_t. (@jasone) |
| 102 | |
| 103 | Bug fixes: |
Jason Evans | e025c51 | 2016-02-28 00:01:13 -0800 | [diff] [blame] | 104 | - Fix stats.cactive accounting regression. (@rustyx, @jasone) |
| 105 | - Handle unaligned keys in hash(). This caused problems for some ARM systems. |
Jason Evans | e3998c6 | 2016-03-07 17:55:55 -0800 | [diff] [blame] | 106 | (@jasone, @cferris1000) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 107 | - Refactor arenas array. In addition to fixing a fork-related deadlock, this |
| 108 | makes arena lookups faster and simpler. (@jasone) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 109 | - Move retained memory allocation out of the default chunk allocation |
| 110 | function, to a location that gets executed even if the application installs |
| 111 | a custom chunk allocation function. This resolves a virtual memory leak. |
| 112 | (@buchgr) |
Jason Evans | e3998c6 | 2016-03-07 17:55:55 -0800 | [diff] [blame] | 113 | - Fix a potential tsd cleanup leak. (@cferris1000, @jasone) |
Jason Evans | e025c51 | 2016-02-28 00:01:13 -0800 | [diff] [blame] | 114 | - Fix run quantization. In practice this bug had no impact unless |
| 115 | applications requested memory with alignment exceeding one page. |
| 116 | (@jasone, @djwatson) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 117 | - Fix LinuxThreads-specific bootstrapping deadlock. (Cosmin Paraschiv) |
| 118 | - jeprof: |
| 119 | + Don't discard curl options if timeout is not defined. (@djwatson) |
| 120 | + Detect failed profile fetches. (@djwatson) |
Jason Evans | e025c51 | 2016-02-28 00:01:13 -0800 | [diff] [blame] | 121 | - Fix stats.arenas.<i>.{dss,lg_dirty_mult,decay_time,pactive,pdirty} for |
| 122 | --disable-stats case. (@jasone) |
Jason Evans | 14be4a7 | 2016-02-26 21:00:02 -0800 | [diff] [blame] | 123 | |
Jason Evans | be41347 | 2015-10-24 07:53:25 -0700 | [diff] [blame] | 124 | * 4.0.4 (October 24, 2015) |
| 125 | |
| 126 | This bugfix release fixes another xallocx() regression. No other regressions |
| 127 | have come to light in over a month, so this is likely a good starting point |
| 128 | for people who prefer to wait for "dot one" releases with all the major issues |
| 129 | shaken out. |
Jason Evans | a784e41 | 2015-09-24 22:21:55 -0700 | [diff] [blame] | 130 | |
| 131 | Bug fixes: |
| 132 | - Fix xallocx(..., MALLOCX_ZERO to zero the last full trailing page of large |
| 133 | allocations that have been randomly assigned an offset of 0 when |
| 134 | --enable-cache-oblivious configure option is enabled. |
| 135 | |
Jason Evans | 0270968 | 2015-09-24 20:05:26 -0700 | [diff] [blame] | 136 | * 4.0.3 (September 24, 2015) |
| 137 | |
| 138 | This bugfix release continues the trend of xallocx() and heap profiling fixes. |
Jason Evans | fb64ec2 | 2015-09-21 18:37:18 -0700 | [diff] [blame] | 139 | |
| 140 | Bug fixes: |
Jason Evans | d260f44 | 2015-09-24 16:38:45 -0700 | [diff] [blame] | 141 | - Fix xallocx(..., MALLOCX_ZERO) to zero all trailing bytes of large |
| 142 | allocations when --enable-cache-oblivious configure option is enabled. |
| 143 | - Fix xallocx(..., MALLOCX_ZERO) to zero trailing bytes of huge allocations |
| 144 | when resizing from/to a size class that is not a multiple of the chunk size. |
Jason Evans | fb64ec2 | 2015-09-21 18:37:18 -0700 | [diff] [blame] | 145 | - Fix prof_tctx_dump_iter() to filter out nodes that were created after heap |
| 146 | profile dumping started. |
Jason Evans | d36c7eb | 2015-09-24 16:53:18 -0700 | [diff] [blame] | 147 | - Work around a potentially bad thread-specific data initialization |
| 148 | interaction with NPTL (glibc's pthreads implementation). |
Jason Evans | fb64ec2 | 2015-09-21 18:37:18 -0700 | [diff] [blame] | 149 | |
Jason Evans | b8e966f | 2015-09-21 10:19:37 -0700 | [diff] [blame] | 150 | * 4.0.2 (September 21, 2015) |
| 151 | |
| 152 | This bugfix release addresses a few bugs specific to heap profiling. |
Jason Evans | 38e2c8f | 2015-09-17 10:05:56 -0700 | [diff] [blame] | 153 | |
| 154 | Bug fixes: |
| 155 | - Fix ixallocx_prof_sample() to never modify nor create sampled small |
| 156 | allocations. xallocx() is in general incapable of moving small allocations, |
| 157 | so this fix removes buggy code without loss of generality. |
Jason Evans | 4be9c79 | 2015-09-17 10:17:55 -0700 | [diff] [blame] | 158 | - Fix irallocx_prof_sample() to always allocate large regions, even when |
| 159 | alignment is non-zero. |
Jason Evans | 3ca0cf6 | 2015-09-17 14:47:39 -0700 | [diff] [blame] | 160 | - Fix prof_alloc_rollback() to read tdata from thread-specific data rather |
| 161 | than dereferencing a potentially invalid tctx. |
Jason Evans | 38e2c8f | 2015-09-17 10:05:56 -0700 | [diff] [blame] | 162 | |
Jason Evans | 1d7540c | 2015-09-15 15:26:23 -0700 | [diff] [blame] | 163 | * 4.0.1 (September 15, 2015) |
| 164 | |
| 165 | This is a bugfix release that is somewhat high risk due to the amount of |
| 166 | refactoring required to address deep xallocx() problems. As a side effect of |
| 167 | these fixes, xallocx() now tries harder to partially fulfill requests for |
| 168 | optional extra space. Note that a couple of minor heap profiling |
| 169 | optimizations are included, but these are better thought of as performance |
| 170 | fixes that were integral to disovering most of the other bugs. |
| 171 | |
| 172 | Optimizations: |
| 173 | - Avoid a chunk metadata read in arena_prof_tctx_set(), since it is in the |
| 174 | fast path when heap profiling is enabled. Additionally, split a special |
| 175 | case out into arena_prof_tctx_reset(), which also avoids chunk metadata |
| 176 | reads. |
| 177 | - Optimize irallocx_prof() to optimistically update the sampler state. The |
| 178 | prior implementation appears to have been a holdover from when |
| 179 | rallocx()/xallocx() functionality was combined as rallocm(). |
Jason Evans | 5ef33a9 | 2015-08-19 14:12:05 -0700 | [diff] [blame] | 180 | |
| 181 | Bug fixes: |
Jason Evans | 1d7540c | 2015-09-15 15:26:23 -0700 | [diff] [blame] | 182 | - Fix TLS configuration such that it is enabled by default for platforms on |
| 183 | which it works correctly. |
Jason Evans | 30949da | 2015-08-25 16:13:59 -0700 | [diff] [blame] | 184 | - Fix arenas_cache_cleanup() and arena_get_hard() to handle |
| 185 | allocation/deallocation within the application's thread-specific data |
| 186 | cleanup functions even after arenas_cache is torn down. |
Jason Evans | 1d7540c | 2015-09-15 15:26:23 -0700 | [diff] [blame] | 187 | - Fix xallocx() bugs related to size+extra exceeding HUGE_MAXCLASS. |
Mike Hommey | 6d8075f | 2015-08-27 20:30:15 -0700 | [diff] [blame] | 188 | - Fix chunk purge hook calls for in-place huge shrinking reallocation to |
| 189 | specify the old chunk size rather than the new chunk size. This bug caused |
| 190 | no correctness issues for the default chunk purge function, but was |
| 191 | visible to custom functions set via the "arena.<i>.chunk_hooks" mallctl. |
Jason Evans | 1d7540c | 2015-09-15 15:26:23 -0700 | [diff] [blame] | 192 | - Fix heap profiling bugs: |
| 193 | + Fix heap profiling to distinguish among otherwise identical sample sites |
| 194 | with interposed resets (triggered via the "prof.reset" mallctl). This bug |
| 195 | could cause data structure corruption that would most likely result in a |
| 196 | segfault. |
| 197 | + Fix irealloc_prof() to prof_alloc_rollback() on OOM. |
| 198 | + Make one call to prof_active_get_unlocked() per allocation event, and use |
| 199 | the result throughout the relevant functions that handle an allocation |
| 200 | event. Also add a missing check in prof_realloc(). These fixes protect |
| 201 | allocation events against concurrent prof_active changes. |
| 202 | + Fix ixallocx_prof() to pass usize_max and zero to ixallocx_prof_sample() |
| 203 | in the correct order. |
| 204 | + Fix prof_realloc() to call prof_free_sampled_object() after calling |
| 205 | prof_malloc_sample_object(). Prior to this fix, if tctx and old_tctx were |
| 206 | the same, the tctx could have been prematurely destroyed. |
| 207 | - Fix portability bugs: |
| 208 | + Don't bitshift by negative amounts when encoding/decoding run sizes in |
| 209 | chunk header maps. This affected systems with page sizes greater than 8 |
| 210 | KiB. |
| 211 | + Rename index_t to szind_t to avoid an existing type on Solaris. |
| 212 | + Add JEMALLOC_CXX_THROW to the memalign() function prototype, in order to |
| 213 | match glibc and avoid compilation errors when including both |
| 214 | jemalloc/jemalloc.h and malloc.h in C++ code. |
| 215 | + Don't assume that /bin/sh is appropriate when running size_classes.sh |
| 216 | during configuration. |
| 217 | + Consider __sparcv9 a synonym for __sparc64__ when defining LG_QUANTUM. |
| 218 | + Link tests to librt if it contains clock_gettime(2). |
Jason Evans | 5ef33a9 | 2015-08-19 14:12:05 -0700 | [diff] [blame] | 219 | |
Jason Evans | 9b68f67 | 2015-08-17 13:21:08 -0700 | [diff] [blame] | 220 | * 4.0.0 (August 17, 2015) |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 221 | |
| 222 | This version contains many speed and space optimizations, both minor and |
| 223 | major. The major themes are generalization, unification, and simplification. |
| 224 | Although many of these optimizations cause no visible behavior change, their |
| 225 | cumulative effect is substantial. |
| 226 | |
| 227 | New features: |
| 228 | - Normalize size class spacing to be consistent across the complete size |
| 229 | range. By default there are four size classes per size doubling, but this |
| 230 | is now configurable via the --with-lg-size-class-group option. Also add the |
| 231 | --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and |
| 232 | --with-lg-tiny-min options, which can be used to tweak page and size class |
| 233 | settings. Impacts: |
| 234 | + Worst case performance for incrementally growing/shrinking reallocation |
| 235 | is improved because there are far fewer size classes, and therefore |
| 236 | copying happens less often. |
| 237 | + Internal fragmentation is limited to 20% for all but the smallest size |
| 238 | classes (those less than four times the quantum). (1B + 4 KiB) |
| 239 | and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation. |
| 240 | + Chunk fragmentation tends to be lower because there are fewer distinct run |
| 241 | sizes to pack. |
| 242 | - Add support for explicit tcaches. The "tcache.create", "tcache.flush", and |
| 243 | "tcache.destroy" mallctls control tcache lifetime and flushing, and the |
| 244 | MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API |
| 245 | control which tcache is used for each operation. |
| 246 | - Implement per thread heap profiling, as well as the ability to |
| 247 | enable/disable heap profiling on a per thread basis. Add the "prof.reset", |
| 248 | "prof.lg_sample", "thread.prof.name", "thread.prof.active", |
| 249 | "opt.prof_thread_active_init", "prof.thread_active_init", and |
| 250 | "thread.prof.active" mallctls. |
| 251 | - Add support for per arena application-specified chunk allocators, configured |
Jason Evans | b49a334 | 2015-07-28 11:28:19 -0400 | [diff] [blame] | 252 | via the "arena.<i>.chunk_hooks" mallctl. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 253 | - Refactor huge allocation to be managed by arenas, so that arenas now |
| 254 | function as general purpose independent allocators. This is important in |
| 255 | the context of user-specified chunk allocators, aside from the scalability |
| 256 | benefits. Related new statistics: |
| 257 | + The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc", |
| 258 | "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests" |
| 259 | mallctls provide high level per arena huge allocation statistics. |
Qinfan Wu | 8975035 | 2015-04-21 16:57:42 -0700 | [diff] [blame] | 260 | + The "arenas.nhchunks", "arenas.hchunk.<i>.size", |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 261 | "stats.arenas.<i>.hchunks.<j>.nmalloc", |
| 262 | "stats.arenas.<i>.hchunks.<j>.ndalloc", |
| 263 | "stats.arenas.<i>.hchunks.<j>.nrequests", and |
| 264 | "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class |
| 265 | statistics. |
| 266 | - Add the 'util' column to malloc_stats_print() output, which reports the |
| 267 | proportion of available regions that are currently in use for each small |
| 268 | size class. |
| 269 | - Add "alloc" and "free" modes for for junk filling (see the "opt.junk" |
| 270 | mallctl), so that it is possible to separately enable junk filling for |
| 271 | allocation versus deallocation. |
| 272 | - Add the jemalloc-config script, which provides information about how |
| 273 | jemalloc was configured, and how to integrate it into application builds. |
| 274 | - Add metadata statistics, which are accessible via the "stats.metadata", |
| 275 | "stats.arenas.<i>.metadata.mapped", and |
| 276 | "stats.arenas.<i>.metadata.allocated" mallctls. |
Jason Evans | 4acd75a | 2015-03-23 17:25:57 -0700 | [diff] [blame] | 277 | - Add the "stats.resident" mallctl, which reports the upper limit of |
| 278 | physically resident memory mapped by the allocator. |
Jason Evans | 562d266 | 2015-03-24 16:36:12 -0700 | [diff] [blame] | 279 | - Add per arena control over unused dirty page purging, via the |
| 280 | "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and |
| 281 | "stats.arenas.<i>.lg_dirty_mult" mallctls. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 282 | - Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump |
| 283 | feature on/off during program execution. |
| 284 | - Add sdallocx(), which implements sized deallocation. The primary |
| 285 | optimization over dallocx() is the removal of a metadata read, which often |
| 286 | suffers an L1 cache miss. |
| 287 | - Add missing header includes in jemalloc/jemalloc.h, so that applications |
| 288 | only have to #include <jemalloc/jemalloc.h>. |
| 289 | - Add support for additional platforms: |
| 290 | + Bitrig |
| 291 | + Cygwin |
| 292 | + DragonFlyBSD |
| 293 | + iOS |
| 294 | + OpenBSD |
| 295 | + OpenRISC/or1k |
| 296 | |
| 297 | Optimizations: |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 298 | - Maintain dirty runs in per arena LRUs rather than in per arena trees of |
| 299 | dirty-run-containing chunks. In practice this change significantly reduces |
| 300 | dirty page purging volume. |
| 301 | - Integrate whole chunks into the unused dirty page purging machinery. This |
| 302 | reduces the cost of repeated huge allocation/deallocation, because it |
| 303 | effectively introduces a cache of chunks. |
| 304 | - Split the arena chunk map into two separate arrays, in order to increase |
| 305 | cache locality for the frequently accessed bits. |
| 306 | - Move small run metadata out of runs, into arena chunk headers. This reduces |
| 307 | run fragmentation, smaller runs reduce external fragmentation for small size |
| 308 | classes, and packed (less uniformly aligned) metadata layout improves CPU |
| 309 | cache set distribution. |
Jason Evans | 8a03cf0 | 2015-05-04 09:58:36 -0700 | [diff] [blame] | 310 | - Randomly distribute large allocation base pointer alignment relative to page |
| 311 | boundaries in order to more uniformly utilize CPU cache sets. This can be |
Jason Evans | f2bc852 | 2015-07-17 16:38:25 -0700 | [diff] [blame] | 312 | disabled via the --disable-cache-oblivious configure option, and queried via |
| 313 | the "config.cache_oblivious" mallctl. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 314 | - Micro-optimize the fast paths for the public API functions. |
| 315 | - Refactor thread-specific data to reside in a single structure. This assures |
| 316 | that only a single TLS read is necessary per call into the public API. |
| 317 | - Implement in-place huge allocation growing and shrinking. |
| 318 | - Refactor rtree (radix tree for chunk lookups) to be lock-free, and make |
| 319 | additional optimizations that reduce maximum lookup depth to one or two |
| 320 | levels. This resolves what was a concurrency bottleneck for per arena huge |
| 321 | allocation, because a global data structure is critical for determining |
| 322 | which arenas own which huge allocations. |
| 323 | |
| 324 | Incompatible changes: |
| 325 | - Replace --enable-cc-silence with --disable-cc-silence to suppress spurious |
| 326 | warnings by default. |
| 327 | - Assure that the constness of malloc_usable_size()'s return type matches that |
| 328 | of the system implementation. |
| 329 | - Change the heap profile dump format to support per thread heap profiling, |
Jason Evans | 7041720 | 2015-05-01 12:31:12 -0700 | [diff] [blame] | 330 | rename pprof to jeprof, and enhance it with the --thread=<n> option. As a |
| 331 | result, the bundled jeprof must now be used rather than the upstream |
| 332 | (gperftools) pprof. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 333 | - Disable "opt.prof_final" by default, in order to avoid atexit(3), which can |
| 334 | internally deadlock on some platforms. |
| 335 | - Change the "arenas.nlruns" mallctl type from size_t to unsigned. |
| 336 | - Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with |
| 337 | "stats.arenas.<i>.bins.<j>.curregs". |
| 338 | - Ignore MALLOC_CONF in set{uid,gid,cap} binaries. |
| 339 | - Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the |
| 340 | MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage. |
| 341 | |
| 342 | Removed features: |
| 343 | - Remove the *allocm() API, which is superseded by the *allocx() API. |
| 344 | - Remove the --enable-dss options, and make dss non-optional on all platforms |
| 345 | which support sbrk(2). |
| 346 | - Remove the "arenas.purge" mallctl, which was obsoleted by the |
| 347 | "arena.<i>.purge" mallctl in 3.1.0. |
| 348 | - Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically |
| 349 | detects whether it is running inside Valgrind. |
| 350 | - Remove the "stats.huge.allocated", "stats.huge.nmalloc", and |
| 351 | "stats.huge.ndalloc" mallctls. |
| 352 | - Remove the --enable-mremap option. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 353 | - Remove the "stats.chunks.current", "stats.chunks.total", and |
| 354 | "stats.chunks.high" mallctls. |
| 355 | |
| 356 | Bug fixes: |
| 357 | - Fix the cactive statistic to decrease (rather than increase) when active |
| 358 | memory decreases. This regression was first released in 3.5.0. |
| 359 | - Fix OOM handling in memalign() and valloc(). A variant of this bug existed |
| 360 | in all releases since 2.0.0, which introduced these functions. |
Jason Evans | 32dca11 | 2015-07-09 11:34:13 -0700 | [diff] [blame] | 361 | - Fix an OOM-related regression in arena_tcache_fill_small(), which could |
| 362 | cause cache corruption on OOM. This regression was present in all releases |
| 363 | from 2.2.0 through 3.6.0. |
Jason Evans | 241abc6 | 2015-06-23 18:47:07 -0700 | [diff] [blame] | 364 | - Fix size class overflow handling for malloc(), posix_memalign(), memalign(), |
| 365 | calloc(), and realloc() when profiling is enabled. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 366 | - Fix the "arena.<i>.dss" mallctl to return an error if "primary" or |
| 367 | "secondary" precedence is specified, but sbrk(2) is not supported. |
| 368 | - Fix fallback lg_floor() implementations to handle extremely large inputs. |
| 369 | - Ensure the default purgeable zone is after the default zone on OS X. |
| 370 | - Fix latent bugs in atomic_*(). |
| 371 | - Fix the "arena.<i>.dss" mallctl to handle read-only calls. |
| 372 | - Fix tls_model configuration to enable the initial-exec model when possible. |
| 373 | - Mark malloc_conf as a weak symbol so that the application can override it. |
| 374 | - Correctly detect glibc's adaptive pthread mutexes. |
| 375 | - Fix the --without-export configure option. |
| 376 | |
Jason Evans | ff53631 | 2014-03-31 09:23:10 -0700 | [diff] [blame] | 377 | * 3.6.0 (March 31, 2014) |
| 378 | |
| 379 | This version contains a critical bug fix for a regression present in 3.5.0 and |
| 380 | 3.5.1. |
| 381 | |
| 382 | Bug fixes: |
| 383 | - Fix a regression in arena_chunk_alloc() that caused crashes during |
| 384 | small/large allocation if chunk allocation failed. In the absence of this |
| 385 | bug, chunk allocation failure would result in allocation failure, e.g. NULL |
| 386 | return from malloc(). This regression was introduced in 3.5.0. |
| 387 | - Fix backtracing for gcc intrinsics-based backtracing by specifying |
| 388 | -fno-omit-frame-pointer to gcc. Note that the application (and all the |
| 389 | libraries it links to) must also be compiled with this option for |
| 390 | backtracing to be reliable. |
| 391 | - Use dss allocation precedence for huge allocations as well as small/large |
| 392 | allocations. |
Jason Evans | 54673fd | 2015-02-23 22:28:43 -0800 | [diff] [blame] | 393 | - Fix test assertion failure message formatting. This bug did not manifest on |
Jason Evans | ff53631 | 2014-03-31 09:23:10 -0700 | [diff] [blame] | 394 | x86_64 systems because of implementation subtleties in va_list. |
| 395 | - Fix inconsequential test failures for hash and SFMT code. |
| 396 | |
| 397 | New features: |
| 398 | - Support heap profiling on FreeBSD. This feature depends on the proc |
| 399 | filesystem being mounted during heap profile dumping. |
| 400 | |
Jason Evans | b9ec5c9 | 2014-02-25 16:43:51 -0800 | [diff] [blame] | 401 | * 3.5.1 (February 25, 2014) |
| 402 | |
| 403 | This version primarily addresses minor bugs in test code. |
| 404 | |
| 405 | Bug fixes: |
| 406 | - Configure Solaris/Illumos to use MADV_FREE. |
| 407 | - Fix junk filling for mremap(2)-based huge reallocation. This is only |
| 408 | relevant if configuring with the --enable-mremap option specified. |
| 409 | - Avoid compilation failure if 'restrict' C99 keyword is not supported by the |
| 410 | compiler. |
| 411 | - Add a configure test for SSE2 rather than assuming it is usable on i686 |
| 412 | systems. This fixes test compilation errors, especially on 32-bit Linux |
| 413 | systems. |
| 414 | - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit |
| 415 | test. |
| 416 | - Fix/remove flawed alignment-related overflow tests. |
| 417 | - Prevent compiler optimizations that could change backtraces in the |
| 418 | prof_accum unit test. |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 419 | |
Jason Evans | 798a481 | 2014-01-22 11:09:50 -0800 | [diff] [blame] | 420 | * 3.5.0 (January 22, 2014) |
Jason Evans | 86abd0d | 2013-11-30 15:25:42 -0800 | [diff] [blame] | 421 | |
Jason Evans | 264dfd3 | 2014-01-17 17:01:23 -0800 | [diff] [blame] | 422 | This version focuses on refactoring and automated testing, though it also |
| 423 | includes some non-trivial heap profiling optimizations not mentioned below. |
| 424 | |
| 425 | New features: |
| 426 | - Add the *allocx() API, which is a successor to the experimental *allocm() |
| 427 | API. The *allocx() functions are slightly simpler to use because they have |
| 428 | fewer parameters, they directly return the results of primary interest, and |
| 429 | mallocx()/rallocx() avoid the strict aliasing pitfall that |
Jason Evans | 9c8baec | 2014-01-22 13:08:47 -0800 | [diff] [blame] | 430 | allocm()/rallocm() share with posix_memalign(). Note that *allocm() is |
Jason Evans | 264dfd3 | 2014-01-17 17:01:23 -0800 | [diff] [blame] | 431 | slated for removal in the next non-bugfix release. |
| 432 | - Add support for LinuxThreads. |
| 433 | |
Jason Evans | 86abd0d | 2013-11-30 15:25:42 -0800 | [diff] [blame] | 434 | Bug fixes: |
Jason Evans | d37d5ad | 2013-12-05 23:01:50 -0800 | [diff] [blame] | 435 | - Unless heap profiling is enabled, disable floating point code and don't link |
| 436 | with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64 |
| 437 | systems, makes it possible to completely disable floating point register |
| 438 | use. Some versions of glibc neglect to save/restore caller-saved floating |
| 439 | point registers during dynamic lazy symbol loading, and the symbol loading |
| 440 | code uses whatever malloc the application happens to have linked/loaded |
| 441 | with, the result being potential floating point register corruption. |
Jason Evans | 264dfd3 | 2014-01-17 17:01:23 -0800 | [diff] [blame] | 442 | - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling |
| 443 | backtrace creation in imemalign(). This bug impacted posix_memalign() and |
| 444 | aligned_alloc(). |
| 445 | - Fix a file descriptor leak in a prof_dump_maps() error path. |
| 446 | - Fix prof_dump() to close the dump file descriptor for all relevant error |
| 447 | paths. |
| 448 | - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for |
| 449 | allocation, not just deallocation. |
| 450 | - Fix a data race for large allocation stats counters. |
| 451 | - Fix a potential infinite loop during thread exit. This bug occurred on |
| 452 | Solaris, and could affect other platforms with similar pthreads TSD |
| 453 | implementations. |
| 454 | - Don't junk-fill reallocations unless usable size changes. This fixes a |
| 455 | violation of the *allocx()/*allocm() semantics. |
| 456 | - Fix growing large reallocation to junk fill new space. |
| 457 | - Fix huge deallocation to junk fill when munmap is disabled. |
Jason Evans | 86abd0d | 2013-11-30 15:25:42 -0800 | [diff] [blame] | 458 | - Change the default private namespace prefix from empty to je_, and change |
| 459 | --with-private-namespace-prefix so that it prepends an additional prefix |
| 460 | rather than replacing je_. This reduces the likelihood of applications |
| 461 | which statically link jemalloc experiencing symbol name collisions. |
Jason Evans | 264dfd3 | 2014-01-17 17:01:23 -0800 | [diff] [blame] | 462 | - Add missing private namespace mangling (relevant when |
| 463 | --with-private-namespace is specified). |
| 464 | - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as |
| 465 | static even for debug builds. |
| 466 | - Add a missing mutex unlock in a malloc_init_hard() error path. In practice |
| 467 | this error path is never executed. |
| 468 | - Fix numerous bugs in malloc_strotumax() error handling/reporting. These |
| 469 | bugs had no impact except for malformed inputs. |
| 470 | - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by |
| 471 | existing calls, so they had no impact. |
Jason Evans | 86abd0d | 2013-11-30 15:25:42 -0800 | [diff] [blame] | 472 | |
Jason Evans | 0f7ba3f | 2013-10-20 19:40:09 -0700 | [diff] [blame] | 473 | * 3.4.1 (October 20, 2013) |
Jason Evans | ff08ef7 | 2013-10-19 21:41:10 -0700 | [diff] [blame] | 474 | |
| 475 | Bug fixes: |
Jason Evans | 7b65180 | 2013-10-20 14:09:54 -0700 | [diff] [blame] | 476 | - Fix a race in the "arenas.extend" mallctl that could cause memory corruption |
| 477 | of internal data structures and subsequent crashes. |
Jason Evans | dda90f5 | 2013-10-19 23:48:40 -0700 | [diff] [blame] | 478 | - Fix Valgrind integration flaws that caused Valgrind warnings about reads of |
| 479 | uninitialized memory in: |
| 480 | + arena chunk headers |
| 481 | + internal zero-initialized data structures (relevant to tcache and prof |
| 482 | code) |
Jason Evans | ff08ef7 | 2013-10-19 21:41:10 -0700 | [diff] [blame] | 483 | - Preserve errno during the first allocation. A readlink(2) call during |
| 484 | initialization fails unless /etc/malloc.conf exists, so errno was typically |
| 485 | set during the first allocation prior to this fix. |
| 486 | - Fix compilation warnings reported by gcc 4.8.1. |
| 487 | |
Jason Evans | 765cc2b | 2013-06-02 20:58:00 -0700 | [diff] [blame] | 488 | * 3.4.0 (June 2, 2013) |
| 489 | |
| 490 | This version is essentially a small bugfix release, but the addition of |
| 491 | aarch64 support requires that the minor version be incremented. |
| 492 | |
| 493 | Bug fixes: |
| 494 | - Fix race-triggered deadlocks in chunk_record(). These deadlocks were |
| 495 | typically triggered by multiple threads concurrently deallocating huge |
| 496 | objects. |
| 497 | |
| 498 | New features: |
| 499 | - Add support for the aarch64 architecture. |
| 500 | |
Jason Evans | 2298835 | 2013-03-06 11:11:17 -0800 | [diff] [blame] | 501 | * 3.3.1 (March 6, 2013) |
| 502 | |
| 503 | This version fixes bugs that are typically encountered only when utilizing |
| 504 | custom run-time options. |
Jason Evans | bbe29d3 | 2013-01-30 15:03:11 -0800 | [diff] [blame] | 505 | |
| 506 | Bug fixes: |
Jason Evans | 88c222c | 2013-02-06 11:59:30 -0800 | [diff] [blame] | 507 | - Fix a locking order bug that could cause deadlock during fork if heap |
| 508 | profiling were enabled. |
Jason Evans | a7a28c3 | 2013-01-31 16:53:58 -0800 | [diff] [blame] | 509 | - Fix a chunk recycling bug that could cause the allocator to lose track of |
Jason Evans | 765cc2b | 2013-06-02 20:58:00 -0700 | [diff] [blame] | 510 | whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause |
Jason Evans | a7a28c3 | 2013-01-31 16:53:58 -0800 | [diff] [blame] | 511 | corruption if allocating via sbrk(2) (unlikely unless running with the |
| 512 | "dss:primary" option specified). This was completely harmless on Linux |
| 513 | unless using mlockall(2) (and unlikely even then, unless the |
| 514 | --disable-munmap configure option or the "dss:primary" option was |
| 515 | specified). This regression was introduced in 3.1.0 by the |
| 516 | mlockall(2)/madvise(2) interaction fix. |
Jason Evans | bbe29d3 | 2013-01-30 15:03:11 -0800 | [diff] [blame] | 517 | - Fix TLS-related memory corruption that could occur during thread exit if the |
| 518 | thread never allocated memory. Only the quarantine and prof facilities were |
| 519 | susceptible. |
Jason Evans | d0e942e | 2013-01-31 14:42:41 -0800 | [diff] [blame] | 520 | - Fix two quarantine bugs: |
| 521 | + Internal reallocation of the quarantined object array leaked the old |
| 522 | array. |
| 523 | + Reallocation failure for internal reallocation of the quarantined object |
| 524 | array (very unlikely) resulted in memory corruption. |
Jason Evans | 0691275 | 2013-01-31 17:02:53 -0800 | [diff] [blame] | 525 | - Fix Valgrind integration to annotate all internally allocated memory in a |
| 526 | way that keeps Valgrind happy about internal data structure access. |
Jason Evans | 2298835 | 2013-03-06 11:11:17 -0800 | [diff] [blame] | 527 | - Fix building for s390 systems. |
Jason Evans | bbe29d3 | 2013-01-30 15:03:11 -0800 | [diff] [blame] | 528 | |
Jason Evans | b5681fb | 2013-01-22 22:45:09 -0800 | [diff] [blame] | 529 | * 3.3.0 (January 23, 2013) |
| 530 | |
| 531 | This version includes a few minor performance improvements in addition to the |
| 532 | listed new features and bug fixes. |
| 533 | |
| 534 | New features: |
| 535 | - Add clipping support to lg_chunk option processing. |
| 536 | - Add the --enable-ivsalloc option. |
| 537 | - Add the --without-export option. |
| 538 | - Add the --disable-zone-allocator option. |
Jason Evans | 6eb84fb | 2012-11-29 22:13:04 -0800 | [diff] [blame] | 539 | |
| 540 | Bug fixes: |
| 541 | - Fix "arenas.extend" mallctl to output the number of arenas. |
Jason Evans | a33488d | 2013-10-03 14:38:39 -0700 | [diff] [blame] | 542 | - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory |
Jason Evans | 1271185 | 2012-12-12 10:12:18 -0800 | [diff] [blame] | 543 | is undefined. |
Jason Evans | b5681fb | 2013-01-22 22:45:09 -0800 | [diff] [blame] | 544 | - Fix build break on FreeBSD related to alloca.h. |
Jason Evans | 6eb84fb | 2012-11-29 22:13:04 -0800 | [diff] [blame] | 545 | |
Jason Evans | 556ddc7 | 2012-11-07 15:16:29 -0800 | [diff] [blame] | 546 | * 3.2.0 (November 9, 2012) |
| 547 | |
| 548 | In addition to a couple of bug fixes, this version modifies page run |
| 549 | allocation and dirty page purging algorithms in order to better control |
| 550 | page-level virtual memory fragmentation. |
Jason Evans | 12efefb | 2012-10-16 22:06:56 -0700 | [diff] [blame] | 551 | |
Jason Evans | e3d1306 | 2012-10-30 15:42:37 -0700 | [diff] [blame] | 552 | Incompatible changes: |
Jason Evans | 556ddc7 | 2012-11-07 15:16:29 -0800 | [diff] [blame] | 553 | - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1). |
Jason Evans | e3d1306 | 2012-10-30 15:42:37 -0700 | [diff] [blame] | 554 | |
Jason Evans | 12efefb | 2012-10-16 22:06:56 -0700 | [diff] [blame] | 555 | Bug fixes: |
| 556 | - Fix dss/mmap allocation precedence code to use recyclable mmap memory only |
| 557 | after primary dss allocation fails. |
Jason Evans | e3d1306 | 2012-10-30 15:42:37 -0700 | [diff] [blame] | 558 | - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced |
| 559 | in 3.1.0 by the addition of the "arena.<i>.purge" mallctl. |
Jason Evans | 12efefb | 2012-10-16 22:06:56 -0700 | [diff] [blame] | 560 | |
Jason Evans | 2b592b0 | 2012-10-16 10:12:40 -0700 | [diff] [blame] | 561 | * 3.1.0 (October 16, 2012) |
Jason Evans | 3860eac | 2012-05-15 13:53:21 -0700 | [diff] [blame] | 562 | |
Jason Evans | 781fe75 | 2012-05-15 14:48:14 -0700 | [diff] [blame] | 563 | New features: |
| 564 | - Auto-detect whether running inside Valgrind, thus removing the need to |
| 565 | manually specify MALLOC_CONF=valgrind:true. |
Jason Evans | 2b592b0 | 2012-10-16 10:12:40 -0700 | [diff] [blame] | 566 | - Add the "arenas.extend" mallctl, which allows applications to create |
| 567 | manually managed arenas. |
| 568 | - Add the ALLOCM_ARENA() flag for {,r,d}allocm(). |
| 569 | - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls, |
| 570 | which provide control over dss/mmap precedence. |
| 571 | - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge". |
| 572 | - Define LG_QUANTUM for hppa. |
Jason Evans | 781fe75 | 2012-05-15 14:48:14 -0700 | [diff] [blame] | 573 | |
Jason Evans | 174b70e | 2012-05-15 23:31:53 -0700 | [diff] [blame] | 574 | Incompatible changes: |
| 575 | - Disable tcache by default if running inside Valgrind, in order to avoid |
| 576 | making unallocated objects appear reachable to Valgrind. |
Jason Evans | 2b592b0 | 2012-10-16 10:12:40 -0700 | [diff] [blame] | 577 | - Drop const from malloc_usable_size() argument on Linux. |
Jason Evans | 174b70e | 2012-05-15 23:31:53 -0700 | [diff] [blame] | 578 | |
Jason Evans | 3860eac | 2012-05-15 13:53:21 -0700 | [diff] [blame] | 579 | Bug fixes: |
| 580 | - Fix heap profiling crash if sampled object is freed via realloc(p, 0). |
Jason Evans | 5c710ce | 2012-05-23 16:09:22 -0700 | [diff] [blame] | 581 | - Remove const from __*_hook variable declarations, so that glibc can modify |
| 582 | them during process forking. |
Jason Evans | 2b592b0 | 2012-10-16 10:12:40 -0700 | [diff] [blame] | 583 | - Fix mlockall(2)/madvise(2) interaction. |
| 584 | - Fix fork(2)-related deadlocks. |
| 585 | - Fix error return value for "thread.tcache.enabled" mallctl. |
Jason Evans | 3860eac | 2012-05-15 13:53:21 -0700 | [diff] [blame] | 586 | |
Jason Evans | cbb71ca | 2012-05-11 17:00:20 -0700 | [diff] [blame] | 587 | * 3.0.0 (May 11, 2012) |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 588 | |
| 589 | Although this version adds some major new features, the primary focus is on |
| 590 | internal code cleanup that facilitates maintainability and portability, most |
| 591 | of which is not reflected in the ChangeLog. This is the first release to |
| 592 | incorporate substantial contributions from numerous other developers, and the |
| 593 | result is a more broadly useful allocator (see the git revision history for |
| 594 | contribution details). Note that the license has been unified, thanks to |
| 595 | Facebook granting a license under the same terms as the other copyright |
| 596 | holders (see COPYING). |
| 597 | |
| 598 | New features: |
| 599 | - Implement Valgrind support, redzones, and quarantine. |
Jason Evans | 079687b | 2012-04-23 12:49:23 -0700 | [diff] [blame] | 600 | - Add support for additional platforms: |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 601 | + FreeBSD |
| 602 | + Mac OS X Lion |
Jason Evans | 40f514f | 2012-04-22 16:21:06 -0700 | [diff] [blame] | 603 | + MinGW |
Jason Evans | cbb71ca | 2012-05-11 17:00:20 -0700 | [diff] [blame] | 604 | + Windows (no support yet for replacing the system malloc) |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 605 | - Add support for additional architectures: |
| 606 | + MIPS |
| 607 | + SH4 |
| 608 | + Tilera |
| 609 | - Add support for cross compiling. |
| 610 | - Add nallocm(), which rounds a request size up to the nearest size class |
| 611 | without actually allocating. |
| 612 | - Implement aligned_alloc() (blame C11). |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 613 | - Add the "thread.tcache.enabled" mallctl. |
Jason Evans | 0b25fe7 | 2012-04-17 16:39:33 -0700 | [diff] [blame] | 614 | - Add the "opt.prof_final" mallctl. |
Jason Evans | 25a000e | 2012-04-17 15:49:30 -0700 | [diff] [blame] | 615 | - Update pprof (from gperftools 2.0). |
Jason Evans | cbb71ca | 2012-05-11 17:00:20 -0700 | [diff] [blame] | 616 | - Add the --with-mangling option. |
| 617 | - Add the --disable-experimental option. |
| 618 | - Add the --disable-munmap option, and make it the default on Linux. |
| 619 | - Add the --enable-mremap option, which disables use of mremap(2) by default. |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 620 | |
| 621 | Incompatible changes: |
| 622 | - Enable stats by default. |
| 623 | - Enable fill by default. |
| 624 | - Disable lazy locking by default. |
| 625 | - Rename the "tcache.flush" mallctl to "thread.tcache.flush". |
| 626 | - Rename the "arenas.pagesize" mallctl to "arenas.page". |
Jason Evans | 0b25fe7 | 2012-04-17 16:39:33 -0700 | [diff] [blame] | 627 | - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). |
| 628 | - Change the "opt.prof_accum" default from true to false. |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 629 | |
| 630 | Removed features: |
| 631 | - Remove the swap feature, including the "config.swap", "swap.avail", |
| 632 | "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls. |
| 633 | - Remove highruns statistics, including the |
| 634 | "stats.arenas.<i>.bins.<j>.highruns" and |
| 635 | "stats.arenas.<i>.lruns.<j>.highruns" mallctls. |
| 636 | - As part of small size class refactoring, remove the "opt.lg_[qc]space_max", |
| 637 | "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and |
| 638 | "arenas.[tqcs]bins" mallctls. |
| 639 | - Remove the "arenas.chunksize" mallctl. |
| 640 | - Remove the "opt.lg_prof_tcmax" option. |
| 641 | - Remove the "opt.lg_prof_bt_max" option. |
| 642 | - Remove the "opt.lg_tcache_gc_sweep" option. |
| 643 | - Remove the --disable-tiny option, including the "config.tiny" mallctl. |
| 644 | - Remove the --enable-dynamic-page-shift configure option. |
| 645 | - Remove the --enable-sysv configure option. |
| 646 | |
| 647 | Bug fixes: |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 648 | - Fix a statistics-related bug in the "thread.arena" mallctl that could cause |
| 649 | invalid statistics and crashes. |
Jason Evans | 079687b | 2012-04-23 12:49:23 -0700 | [diff] [blame] | 650 | - Work around TLS deallocation via free() on Linux. This bug could cause |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 651 | write-after-free memory corruption. |
Jason Evans | 52386b2 | 2012-04-22 16:00:11 -0700 | [diff] [blame] | 652 | - Fix a potential deadlock that could occur during interval- and |
| 653 | growth-triggered heap profile dumps. |
Jason Evans | d8ceef6 | 2012-05-10 20:59:39 -0700 | [diff] [blame] | 654 | - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags. |
Jason Evans | 8f0e0eb | 2012-04-21 13:33:48 -0700 | [diff] [blame] | 655 | - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could |
| 656 | cause memory corruption and crashes with --enable-dss specified. |
Jason Evans | 52386b2 | 2012-04-22 16:00:11 -0700 | [diff] [blame] | 657 | - Fix fork-related bugs that could cause deadlock in children between fork |
| 658 | and exec. |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 659 | - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter. |
| 660 | - Fix realloc(p, 0) to act like free(p). |
| 661 | - Do not enforce minimum alignment in memalign(). |
| 662 | - Check for NULL pointer in malloc_usable_size(). |
Jason Evans | 52386b2 | 2012-04-22 16:00:11 -0700 | [diff] [blame] | 663 | - Fix an off-by-one heap profile statistics bug that could be observed in |
| 664 | interval- and growth-triggered heap profiles. |
Jason Evans | 6b9ed67 | 2012-04-25 13:12:46 -0700 | [diff] [blame] | 665 | - Fix the "epoch" mallctl to update cached stats even if the passed in epoch |
| 666 | is 0. |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 667 | - Fix bin->runcur management to fix a layout policy bug. This bug did not |
| 668 | affect correctness. |
| 669 | - Fix a bug in choose_arena_hard() that potentially caused more arenas to be |
| 670 | initialized than necessary. |
| 671 | - Add missing "opt.lg_tcache_max" mallctl implementation. |
| 672 | - Use glibc allocator hooks to make mixed allocator usage less likely. |
| 673 | - Fix build issues for --disable-tcache. |
Jason Evans | 918d6e2 | 2012-04-20 13:42:21 -0700 | [diff] [blame] | 674 | - Don't mangle pthread_create() when --with-private-namespace is specified. |
Jason Evans | 9ef7f5d | 2012-04-16 18:16:48 -0700 | [diff] [blame] | 675 | |
Jason Evans | b3bd885 | 2011-11-14 17:12:45 -0800 | [diff] [blame] | 676 | * 2.2.5 (November 14, 2011) |
| 677 | |
| 678 | Bug fixes: |
| 679 | - Fix huge_ralloc() race when using mremap(2). This is a serious bug that |
| 680 | could cause memory corruption and/or crashes. |
| 681 | - Fix huge_ralloc() to maintain chunk statistics. |
| 682 | - Fix malloc_stats_print(..., "a") output. |
| 683 | |
Jason Evans | ca9ee1a | 2011-11-05 21:46:23 -0700 | [diff] [blame] | 684 | * 2.2.4 (November 5, 2011) |
| 685 | |
| 686 | Bug fixes: |
| 687 | - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as |
| 688 | well as for --disable-tls builds in earlier releases. |
| 689 | - Do not assume a 4 KiB page size in test/rallocm.c. |
| 690 | |
Jason Evans | c67e4fd | 2011-08-31 15:19:13 -0700 | [diff] [blame] | 691 | * 2.2.3 (August 31, 2011) |
| 692 | |
| 693 | This version fixes numerous bugs related to heap profiling. |
| 694 | |
| 695 | Bug fixes: |
| 696 | - Fix a prof-related race condition. This bug could cause memory corruption, |
| 697 | but only occurred in non-default configurations (prof_accum:false). |
| 698 | - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is |
| 699 | excluded from backtraces). |
| 700 | - Fix a prof-related bug in realloc() (only triggered by OOM errors). |
| 701 | - Fix prof-related bugs in allocm() and rallocm(). |
| 702 | - Fix prof_tdata_cleanup() for --disable-tls builds. |
| 703 | - Fix a relative include path, to fix objdir builds. |
| 704 | |
Jason Evans | 4c48481 | 2011-07-30 16:59:13 -0700 | [diff] [blame] | 705 | * 2.2.2 (July 30, 2011) |
| 706 | |
| 707 | Bug fixes: |
| 708 | - Fix a build error for --disable-tcache. |
| 709 | - Fix assertions in arena_purge() (for real this time). |
| 710 | - Add the --with-private-namespace option. This is a workaround for symbol |
| 711 | conflicts that can inadvertently arise when using static libraries. |
| 712 | |
Jason Evans | 7d9ebea | 2011-03-30 15:01:08 -0700 | [diff] [blame] | 713 | * 2.2.1 (March 30, 2011) |
| 714 | |
| 715 | Bug fixes: |
| 716 | - Implement atomic operations for x86/x64. This fixes compilation failures |
| 717 | for versions of gcc that are still in wide use. |
| 718 | - Fix an assertion in arena_purge(). |
| 719 | |
Jason Evans | 4bcd987 | 2011-03-22 15:30:22 -0700 | [diff] [blame] | 720 | * 2.2.0 (March 22, 2011) |
| 721 | |
| 722 | This version incorporates several improvements to algorithms and data |
| 723 | structures that tend to reduce fragmentation and increase speed. |
| 724 | |
| 725 | New features: |
| 726 | - Add the "stats.cactive" mallctl. |
| 727 | - Update pprof (from google-perftools 1.7). |
| 728 | - Improve backtracing-related configuration logic, and add the |
| 729 | --disable-prof-libgcc option. |
| 730 | |
| 731 | Bug fixes: |
| 732 | - Change default symbol visibility from "internal", to "hidden", which |
| 733 | decreases the overhead of library-internal function calls. |
| 734 | - Fix symbol visibility so that it is also set on OS X. |
| 735 | - Fix a build dependency regression caused by the introduction of the .pic.o |
| 736 | suffix for PIC object files. |
| 737 | - Add missing checks for mutex initialization failures. |
| 738 | - Don't use libgcc-based backtracing except on x64, where it is known to work. |
| 739 | - Fix deadlocks on OS X that were due to memory allocation in |
| 740 | pthread_mutex_lock(). |
| 741 | - Heap profiling-specific fixes: |
| 742 | + Fix memory corruption due to integer overflow in small region index |
| 743 | computation, when using a small enough sample interval that profiling |
| 744 | context pointers are stored in small run headers. |
| 745 | + Fix a bootstrap ordering bug that only occurred with TLS disabled. |
| 746 | + Fix a rallocm() rsize bug. |
| 747 | + Fix error detection bugs for aligned memory allocation. |
| 748 | |
Jason Evans | 0e4d0d1 | 2011-03-14 16:41:03 -0700 | [diff] [blame] | 749 | * 2.1.3 (March 14, 2011) |
| 750 | |
| 751 | Bug fixes: |
| 752 | - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix |
| 753 | for OS X in 2.1.2). |
| 754 | - Fix a "thread.arena" mallctl bug. |
| 755 | - Fix a thread cache stats merging bug. |
| 756 | |
je | 6e56e5e | 2011-03-02 11:23:41 -0800 | [diff] [blame] | 757 | * 2.1.2 (March 2, 2011) |
| 758 | |
| 759 | Bug fixes: |
| 760 | - Fix "thread.{de,}allocatedp" mallctl for OS X. |
| 761 | - Add missing jemalloc.a to build system. |
| 762 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 763 | * 2.1.1 (January 31, 2011) |
Jason Evans | ada55b2 | 2011-01-31 20:08:56 -0800 | [diff] [blame] | 764 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 765 | Bug fixes: |
Jason Evans | ada55b2 | 2011-01-31 20:08:56 -0800 | [diff] [blame] | 766 | - Fix aligned huge reallocation (affected allocm()). |
| 767 | - Fix the ALLOCM_LG_ALIGN macro definition. |
| 768 | - Fix a heap dumping deadlock. |
| 769 | - Fix a "thread.arena" mallctl bug. |
| 770 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 771 | * 2.1.0 (December 3, 2010) |
Jason Evans | 0e8d3d2 | 2010-12-03 17:02:16 -0800 | [diff] [blame] | 772 | |
| 773 | This version incorporates some optimizations that can't quite be considered |
| 774 | bug fixes. |
| 775 | |
| 776 | New features: |
| 777 | - Use Linux's mremap(2) for huge object reallocation when possible. |
| 778 | - Avoid locking in mallctl*() when possible. |
| 779 | - Add the "thread.[de]allocatedp" mallctl's. |
| 780 | - Convert the manual page source from roff to DocBook, and generate both roff |
| 781 | and HTML manuals. |
| 782 | |
| 783 | Bug fixes: |
| 784 | - Fix a crash due to incorrect bootstrap ordering. This only impacted |
| 785 | --enable-debug --enable-dss configurations. |
| 786 | - Fix a minor statistics bug for mallctl("swap.avail", ...). |
| 787 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 788 | * 2.0.1 (October 29, 2010) |
Jason Evans | 53806fe | 2010-10-29 20:16:39 -0700 | [diff] [blame] | 789 | |
| 790 | Bug fixes: |
| 791 | - Fix a race condition in heap profiling that could cause undefined behavior |
Jason Evans | ada55b2 | 2011-01-31 20:08:56 -0800 | [diff] [blame] | 792 | if "opt.prof_accum" were disabled. |
Jason Evans | 53806fe | 2010-10-29 20:16:39 -0700 | [diff] [blame] | 793 | - Add missing mutex unlocks for some OOM error paths in the heap profiling |
| 794 | code. |
| 795 | - Fix a compilation error for non-C99 builds. |
| 796 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 797 | * 2.0.0 (October 24, 2010) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 798 | |
| 799 | This version focuses on the experimental *allocm() API, and on improved |
| 800 | run-time configuration/introspection. Nonetheless, numerous performance |
| 801 | improvements are also included. |
| 802 | |
| 803 | New features: |
Jason Evans | b059a53 | 2010-10-24 16:54:40 -0700 | [diff] [blame] | 804 | - Implement the experimental {,r,s,d}allocm() API, which provides a superset |
| 805 | of the functionality available via malloc(), calloc(), posix_memalign(), |
| 806 | realloc(), malloc_usable_size(), and free(). These functions can be used to |
| 807 | allocate/reallocate aligned zeroed memory, ask for optional extra memory |
| 808 | during reallocation, prevent object movement during reallocation, etc. |
| 809 | - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is |
| 810 | more human-readable, and more flexible. For example: |
| 811 | JEMALLOC_OPTIONS=AJP |
| 812 | is now: |
| 813 | MALLOC_CONF=abort:true,fill:true,stats_print:true |
| 814 | - Port to Apple OS X. Sponsored by Mozilla. |
| 815 | - Make it possible for the application to control thread-->arena mappings via |
| 816 | the "thread.arena" mallctl. |
| 817 | - Add compile-time support for all TLS-related functionality via pthreads TSD. |
| 818 | This is mainly of interest for OS X, which does not support TLS, but has a |
| 819 | TSD implementation with similar performance. |
| 820 | - Override memalign() and valloc() if they are provided by the system. |
| 821 | - Add the "arenas.purge" mallctl, which can be used to synchronously purge all |
| 822 | dirty unused pages. |
| 823 | - Make cumulative heap profiling data optional, so that it is possible to |
| 824 | limit the amount of memory consumed by heap profiling data structures. |
| 825 | - Add per thread allocation counters that can be accessed via the |
| 826 | "thread.allocated" and "thread.deallocated" mallctls. |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 827 | |
| 828 | Incompatible changes: |
Jason Evans | b059a53 | 2010-10-24 16:54:40 -0700 | [diff] [blame] | 829 | - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above). |
| 830 | - Increase default backtrace depth from 4 to 128 for heap profiling. |
| 831 | - Disable interval-based profile dumps by default. |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 832 | |
| 833 | Bug fixes: |
| 834 | - Remove bad assertions in fork handler functions. These assertions could |
| 835 | cause aborts for some combinations of configure settings. |
| 836 | - Fix strerror_r() usage to deal with non-standard semantics in GNU libc. |
| 837 | - Fix leak context reporting. This bug tended to cause the number of contexts |
| 838 | to be underreported (though the reported number of objects and bytes were |
| 839 | correct). |
| 840 | - Fix a realloc() bug for large in-place growing reallocation. This bug could |
| 841 | cause memory corruption, but it was hard to trigger. |
| 842 | - Fix an allocation bug for small allocations that could be triggered if |
| 843 | multiple threads raced to create a new run of backing pages. |
| 844 | - Enhance the heap profiler to trigger samples based on usable size, rather |
| 845 | than request size. |
| 846 | - Fix a heap profiling bug due to sometimes losing track of requested object |
| 847 | size for sampled objects. |
| 848 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 849 | * 1.0.3 (August 12, 2010) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 850 | |
| 851 | Bug fixes: |
| 852 | - Fix the libunwind-based implementation of stack backtracing (used for heap |
| 853 | profiling). This bug could cause zero-length backtraces to be reported. |
| 854 | - Add a missing mutex unlock in library initialization code. If multiple |
| 855 | threads raced to initialize malloc, some of them could end up permanently |
| 856 | blocked. |
| 857 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 858 | * 1.0.2 (May 11, 2010) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 859 | |
| 860 | Bug fixes: |
| 861 | - Fix junk filling of large objects, which could cause memory corruption. |
| 862 | - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual |
| 863 | memory limits could cause swap file configuration to fail. Contributed by |
| 864 | Jordan DeLong. |
| 865 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 866 | * 1.0.1 (April 14, 2010) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 867 | |
| 868 | Bug fixes: |
| 869 | - Fix compilation when --enable-fill is specified. |
| 870 | - Fix threads-related profiling bugs that affected accuracy and caused memory |
| 871 | to be leaked during thread exit. |
| 872 | - Fix dirty page purging race conditions that could cause crashes. |
| 873 | - Fix crash in tcache flushing code during thread destruction. |
| 874 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 875 | * 1.0.0 (April 11, 2010) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 876 | |
| 877 | This release focuses on speed and run-time introspection. Numerous |
| 878 | algorithmic improvements make this release substantially faster than its |
| 879 | predecessors. |
| 880 | |
| 881 | New features: |
| 882 | - Implement autoconf-based configuration system. |
| 883 | - Add mallctl*(), for the purposes of introspection and run-time |
| 884 | configuration. |
| 885 | - Make it possible for the application to manually flush a thread's cache, via |
| 886 | the "tcache.flush" mallctl. |
| 887 | - Base maximum dirty page count on proportion of active memory. |
charsyam | ad6800f | 2015-07-04 01:06:06 +0900 | [diff] [blame] | 888 | - Compute various additional run-time statistics, including per size class |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 889 | statistics for large objects. |
| 890 | - Expose malloc_stats_print(), which can be called repeatedly by the |
| 891 | application. |
| 892 | - Simplify the malloc_message() signature to only take one string argument, |
| 893 | and incorporate an opaque data pointer argument for use by the application |
| 894 | in combination with malloc_stats_print(). |
| 895 | - Add support for allocation backed by one or more swap files, and allow the |
| 896 | application to disable over-commit if swap files are in use. |
| 897 | - Implement allocation profiling and leak checking. |
| 898 | |
| 899 | Removed features: |
| 900 | - Remove the dynamic arena rebalancing code, since thread-specific caching |
| 901 | reduces its utility. |
| 902 | |
| 903 | Bug fixes: |
| 904 | - Modify chunk allocation to work when address space layout randomization |
| 905 | (ASLR) is in use. |
| 906 | - Fix thread cleanup bugs related to TLS destruction. |
| 907 | - Handle 0-size allocation requests in posix_memalign(). |
| 908 | - Fix a chunk leak. The leaked chunks were never touched, so this impacted |
| 909 | virtual memory usage, but not physical memory usage. |
| 910 | |
Jason Evans | 6369286 | 2011-02-07 22:48:35 -0800 | [diff] [blame] | 911 | * linux_2008082[78]a (August 27/28, 2008) |
Jason Evans | 379f847 | 2010-10-24 16:18:29 -0700 | [diff] [blame] | 912 | |
| 913 | These snapshot releases are the simple result of incorporating Linux-specific |
| 914 | support into the FreeBSD malloc sources. |
| 915 | |
| 916 | -------------------------------------------------------------------------------- |
| 917 | vim:filetype=text:textwidth=80 |