4.2.1

Bug fixes:

• Fix bootstrapping issues for configurations that require allocation during tsd initialization (e.g. --disable-tls). (@cferris1000, @jasone)
• Fix gettimeofday() version of nstime_update(). (@ronawho)
• Fix Valgrind regressions in calloc() and chunk_alloc_wrapper(). (@ronawho)
• Fix potential VM map fragmentation regression. (@jasone)
• Fix opt_zero-triggered in-place huge reallocation zeroing. (@jasone)
• Fix heap profiling context leaks in reallocation edge cases. (@jasone)

4.2.0

New features:

• Add the arena.<i>.reset mallctl, which makes it possible to discard all of an arena's allocations in a single operation. (@jasone)
• Add the stats.retained and stats.arenas.<i>.retained statistics. (@jasone)
• Add the --with-version configure option. (@jasone)
• Support --with-lg-page values larger than actual page size. (@jasone)

Optimizations:

• Use pairing heaps rather than red-black trees for various hot data structures. (@djwatson, @jasone)
• Streamline fast paths of rtree operations. (@jasone)
• Optimize the fast paths of calloc() and [m,d,sd]allocx(). (@jasone)
• Decommit unused virtual memory if the OS does not overcommit. (@jasone)
• Specify MAP_NORESERVE on Linux if [heuristic] overcommit is active, in order to avoid unfortunate interactions during fork(2). (@jasone)

Bug fixes:

• Fix chunk accounting related to triggering gdump profiles. (@jasone)
• Link against librt for clock_gettime(2) if glibc < 2.17. (@jasone)
• Scale leak report summary according to sampling probability. (@jasone)

4.1.1

This bugfix release resolves a variety of mostly minor issues, though the bitmap fix is critical for 64-bit Windows.

Bug fixes:

• Fix the linear scan version of bitmap_sfu() to shift by the proper amount even when sizeof(long) is not the same as sizeof(void *), as on 64-bit Windows. (@jasone)
• Fix hashing functions to avoid unaligned memory accesses (and resulting crashes). This is relevant at least to some ARM-based platforms. (@rkmisra)
• Fix fork()-related lock rank ordering reversals. These reversals were unlikely to cause deadlocks in practice except when heap profiling was enabled and active. (@jasone)
• Fix various chunk leaks in OOM code paths. (@jasone)
• Fix malloc_stats_print() to print opt.narenas correctly. (@jasone)
• Fix MSVC-specific build/test issues. (@rustyx, @yuslepukhin)
• Fix a variety of test failures that were due to test fragility rather than core bugs. (@jasone)

4.1.0

This release is primarily about optimizations, but it also incorporates a lot of portability-motivated refactoring and enhancements. Many people worked on this release, to an extent that even with the omission here of minor changes (see git revision history), and of the people who reported and diagnosed issues, so much of the work was contributed that starting with this release, changes are annotated with author credits to help reflect the collaborative effort involved.

New features:

• Implement decay-based unused dirty page purging, a major optimization with mallctl API impact. This is an alternative to the existing ratio-based unused dirty page purging, and is intended to eventually become the sole purging mechanism. New mallctls:

  • opt.purge
  • opt.decay_time
  • arena.<i>.decay
  • arena.<i>.decay_time
  • arenas.decay_time
  • stats.arenas.<i>.decay_time

  (@jasone, @cevans87)

• Add --with-malloc-conf, which makes it possible to embed a default options string during configuration. This was motivated by the desire to specify --with-malloc-conf=purge:decay , since the default must remain purge:ratio until the 5.0.0 release. (@jasone)

• Add MS Visual Studio 2015 support. (@rustyx, @yuslepukhin)

• Make *allocx() size class overflow behavior defined. The maximum size class is now less than PTRDIFF_MAX to protect applications against numerical overflow, and all allocation functions are guaranteed to indicate errors rather than potentially crashing if the request size exceeds the maximum size class. (@jasone)

• jeprof:

  • Add raw heap profile support. (@jasone)
  • Add --retain and --exclude for backtrace symbol filtering. (@jasone)

Optimizations:

• Optimize the fast path to combine various bootstrapping and configuration checks and execute more streamlined code in the common case. (@interwq)
• Use linear scan for small bitmaps (used for small object tracking). In addition to speeding up bitmap operations on 64-bit systems, this reduces allocator metadata overhead by approximately 0.2%. (@djwatson)
• Separate arena_avail trees, which substantially speeds up run tree operations. (@djwatson)
• Use memoization (boot-time-computed table) for run quantization. Separate arena_avail trees reduced the importance of this optimization. (@jasone)
• Attempt mmap-based in-place huge reallocation. This can dramatically speed up incremental huge reallocation. (@jasone)

Incompatible changes:

• Make opt.narenas unsigned rather than size_t. (@jasone)

Bug fixes:

• Fix stats.cactive accounting regression. (@rustyx, @jasone)
• Handle unaligned keys in hash(). This caused problems for some ARM systems. (@jasone, Christopher Ferris)
• Refactor arenas array. In addition to fixing a fork-related deadlock, this makes arena lookups faster and simpler. (@jasone)
• Move retained memory allocation out of the default chunk allocation function, to a location that gets executed even if the application installs a custom chunk allocation function. This resolves a virtual memory leak. (@buchgr)
• Fix a potential tsd cleanup leak. (Christopher Ferris, @jasone)
• Fix run quantization. In practice this bug had no impact unless applications requested memory with alignment exceeding one page. (@jasone, @djwatson)
• Fix LinuxThreads-specific bootstrapping deadlock. (Cosmin Paraschiv)
• jeprof:
  • Don't discard curl options if timeout is not defined. (@djwatson)
  • Detect failed profile fetches. (@djwatson)
• Fix stats.arenas.<i>.{dss,lg_dirty_mult,decay_time,pactive,pdirty} for --disable-stats case. (@jasone)

4.0.4

This bugfix release fixes another xallocx() regression. No other regressions have come to light in over a month, so this is likely a good starting point for people who prefer to wait for "dot one" releases with all the major issues shaken out.

Bug fixes:

• Fix xallocx(..., MALLOCX_ZERO to zero the last full trailing page of large allocations that have been randomly assigned an offset of 0 when --enable-cache-oblivious configure option is enabled.

4.0.3

This bugfix release continues the trend of xallocx() and heap profiling fixes.

Bug fixes:

• Fix xallocx(..., MALLOCX_ZERO) to zero all trailing bytes of large allocations when --enable-cache-oblivious configure option is enabled.
• Fix xallocx(..., MALLOCX_ZERO) to zero trailing bytes of huge allocations when resizing from/to a size class that is not a multiple of the chunk size.
• Fix prof_tctx_dump_iter() to filter out nodes that were created after heap profile dumping started.
• Work around a potentially bad thread-specific data initialization interaction with NPTL (glibc's pthreads implementation).

4.0.2

This bugfix release addresses a few bugs specific to heap profiling.

Bug fixes:

• Fix ixallocx_prof_sample() to never modify nor create sampled small allocations. xallocx() is in general incapable of moving small allocations, so this fix removes buggy code without loss of generality.
• Fix irallocx_prof_sample() to always allocate large regions, even when alignment is non-zero.
• Fix prof_alloc_rollback() to read tdata from thread-specific data rather than dereferencing a potentially invalid tctx.

4.0.1

This is a bugfix release that is somewhat high risk due to the amount of refactoring required to address deep xallocx() problems. As a side effect of these fixes, xallocx() now tries harder to partially fulfill requests for optional extra space. Note that a couple of minor heap profiling optimizations are included, but these are better thought of as performance fixes that were integral to discovering most of the other bugs.

Optimizations:

• Avoid a chunk metadata read in arena_prof_tctx_set(), since it is in the fast path when heap profiling is enabled. Additionally, split a special case out into arena_prof_tctx_reset(), which also avoids chunk metadata reads.
• Optimize irallocx_prof() to optimistically update the sampler state. The prior implementation appears to have been a holdover from when rallocx()/xallocx() functionality was combined as rallocm().

Bug fixes:

• Fix TLS configuration such that it is enabled by default for platforms on which it works correctly.
• Fix arenas_cache_cleanup() and arena_get_hard() to handle allocation/deallocation within the application's thread-specific data cleanup functions even after arenas_cache is torn down.
• Fix xallocx() bugs related to size+extra exceeding HUGE_MAXCLASS.
• Fix chunk purge hook calls for in-place huge shrinking reallocation to specify the old chunk size rather than the new chunk size. This bug caused no correctness issues for the default chunk purge function, but was visible to custom functions set via the "arena.<i>.chunk_hooks" mallctl.
• Fix heap profiling bugs:
  • Fix heap profiling to distinguish among otherwise identical sample sites with interposed resets (triggered via the "prof.reset" mallctl). This bug could cause data structure corruption that would most likely result in a segfault.
  • Fix irealloc_prof() to prof_alloc_rollback() on OOM.
  • Make one call to prof_active_get_unlocked() per allocation event, and use the result throughout the relevant functions that handle an allocation event. Also add a missing check in prof_realloc(). These fixes protect allocation events against concurrent prof_active changes.
  • Fix ixallocx_prof() to pass usize_max and zero to ixallocx_prof_sample() in the correct order.
  • Fix prof_realloc() to call prof_free_sampled_object() after calling prof_malloc_sample_object(). Prior to this fix, if tctx and old_tctx were the same, the tctx could have been prematurely destroyed.
• Fix portability bugs:
  • Don't bitshift by negative amounts when encoding/decoding run sizes in chunk header maps. This affected systems with page sizes greater than 8 KiB.
  • Rename index_t to szind_t to avoid an existing type on Solaris.
  • Add JEMALLOC_CXX_THROW to the memalign() function prototype, in order to match glibc and avoid compilation errors when including both jemalloc/jemalloc.h and malloc.h in C++ code.
  • Don't assume that /bin/sh is appropriate when running size_classes.sh during configuration.
  • Consider __sparcv9 a synonym for sparc64 when defining LG_QUANTUM.
  • Link tests to librt if it contains clock_gettime(2).

4.0.0

This version contains many speed and space optimizations, both minor and major. The major themes are generalization, unification, and simplification. Although many of these optimizations cause no visible behavior change, their cumulative effect is substantial.

New features:

• Normalize size class spacing to be consistent across the complete size range. By default there are four size classes per size doubling, but this is now configurable via the --with-lg-size-class-group option. Also add the --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and --with-lg-tiny-min options, which can be used to tweak page and size class settings. Impacts:
  • Worst case performance for incrementally growing/shrinking reallocation is improved because there are far fewer size classes, and therefore copying happens less often.
  • Internal fragmentation is limited to 20% for all but the smallest size classes (those less than four times the quantum). (1B + 4 KiB) and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation.
  • Chunk fragmentation tends to be lower because there are fewer distinct run sizes to pack.
• Add support for explicit tcaches. The "tcache.create", "tcache.flush", and "tcache.destroy" mallctls control tcache lifetime and flushing, and the MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API control which tcache is used for each operation.
• Implement per thread heap profiling, as well as the ability to enable/disable heap profiling on a per thread basis. Add the "prof.reset", "prof.lg_sample", "thread.prof.name", "thread.prof.active", "opt.prof_thread_active_init", "prof.thread_active_init", and "thread.prof.active" mallctls.
• Add support for per arena application-specified chunk allocators, configured via the "arena.<i>.chunk_hooks" mallctl.
• Refactor huge allocation to be managed by arenas, so that arenas now function as general purpose independent allocators. This is important in the context of user-specified chunk allocators, aside from the scalability benefits. Related new statistics:
  • The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc", "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests" mallctls provide high level per arena huge allocation statistics.
  • The "arenas.nhchunks", "arenas.hchunk.<i>.size", "stats.arenas.<i>.hchunks.<j>.nmalloc", "stats.arenas.<i>.hchunks.<j>.ndalloc", "stats.arenas.<i>.hchunks.<j>.nrequests", and "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class statistics.
• Add the 'util' column to malloc_stats_print() output, which reports the proportion of available regions that are currently in use for each small size class.
• Add "alloc" and "free" modes for for junk filling (see the "opt.junk" mallctl), so that it is possible to separately enable junk filling for allocation versus deallocation.
• Add the jemalloc-config script, which provides information about how jemalloc was configured, and how to integrate it into application builds.
• Add metadata statistics, which are accessible via the "stats.metadata", "stats.arenas.<i>.metadata.mapped", and "stats.arenas.<i>.metadata.allocated" mallctls.
• Add the "stats.resident" mallctl, which reports the upper limit of physically resident memory mapped by the allocator.
• Add per arena control over unused dirty page purging, via the "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and "stats.arenas.<i>.lg_dirty_mult" mallctls.
• Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump feature on/off during program execution.
• Add sdallocx(), which implements sized deallocation. The primary optimization over dallocx() is the removal of a metadata read, which often suffers an L1 cache miss.
• Add missing header includes in jemalloc/jemalloc.h, so that applications only have to #include <jemalloc/jemalloc.h>.
• Add support for additional platforms:
  • Bitrig
  • Cygwin
  • DragonFlyBSD
  • iOS
  • OpenBSD
  • OpenRISC/or1k

Optimizations:

• Maintain dirty runs in per arena LRUs rather than in per arena trees of dirty-run-containing chunks. In practice this change significantly reduces dirty page purging volume.
• Integrate whole chunks into the unused dirty page purging machinery. This reduces the cost of repeated huge allocation/deallocation, because it effectively introduces a cache of chunks.
• Split the arena chunk map into two separate arrays, in order to increase cache locality for the frequently accessed bits.
• Move small run metadata out of runs, into arena chunk headers. This reduces run fragmentation, smaller runs reduce external fragmentation for small size classes, and packed (less uniformly aligned) metadata layout improves CPU cache set distribution.
• Randomly distribute large allocation base pointer alignment relative to page boundaries in order to more uniformly utilize CPU cache sets. This can be disabled via the --disable-cache-oblivious configure option, and queried via the "config.cache_oblivious" mallctl.
• Micro-optimize the fast paths for the public API functions.
• Refactor thread-specific data to reside in a single structure. This assures that only a single TLS read is necessary per call into the public API.
• Implement in-place huge allocation growing and shrinking.
• Refactor rtree (radix tree for chunk lookups) to be lock-free, and make additional optimizations that reduce maximum lookup depth to one or two levels. This resolves what was a concurrency bottleneck for per arena huge allocation, because a global data structure is critical for determining which arenas own which huge allocations.

Incompatible changes:

• Replace --enable-cc-silence with --disable-cc-silence to suppress spurious warnings by default.
• Assure that the constness of malloc_usable_size()'s return type matches that of the system implementation.
• Change the heap profile dump format to support per thread heap profiling, rename pprof to jeprof, and enhance it with the --thread=<n> option. As a result, the bundled jeprof must now be used rather than the upstream (gperftools) pprof.
• Disable "opt.prof_final" by default, in order to avoid atexit(3), which can internally deadlock on some platforms.
• Change the "arenas.nlruns" mallctl type from size_t to unsigned.
• Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with "stats.arenas.<i>.bins.<j>.curregs".
• Ignore MALLOC_CONF in set{uid,gid,cap} binaries.
• Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage.

Removed features:

• Remove the *allocm() API, which is superseded by the *allocx() API.
• Remove the --enable-dss options, and make dss non-optional on all platforms which support sbrk(2).
• Remove the "arenas.purge" mallctl, which was obsoleted by the "arena.<i>.purge" mallctl in 3.1.0.
• Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically detects whether it is running inside Valgrind.
• Remove the "stats.huge.allocated", "stats.huge.nmalloc", and "stats.huge.ndalloc" mallctls.
• Remove the --enable-mremap option.
• Remove the "stats.chunks.current", "stats.chunks.total", and "stats.chunks.high" mallctls.

Bug fixes:

• Fix the cactive statistic to decrease (rather than increase) when active memory decreases. This regression was first released in 3.5.0.
• Fix OOM handling in memalign() and valloc(). A variant of this bug existed in all releases since 2.0.0, which introduced these functions.
• Fix an OOM-related regression in arena_tcache_fill_small(), which could cause cache corruption on OOM. This regression was present in all releases from 2.2.0 through 3.6.0.
• Fix size class overflow handling for malloc(), posix_memalign(), memalign(), calloc(), and realloc() when profiling is enabled.
• Fix the "arena.<i>.dss" mallctl to return an error if "primary" or "secondary" precedence is specified, but sbrk(2) is not supported.
• Fix fallback lg_floor() implementations to handle extremely large inputs.
• Ensure the default purgeable zone is after the default zone on OS X.
• Fix latent bugs in atomic_*().
• Fix the "arena.<i>.dss" mallctl to handle read-only calls.
• Fix tls_model configuration to enable the initial-exec model when possible.
• Mark malloc_conf as a weak symbol so that the application can override it.
• Correctly detect glibc's adaptive pthread mutexes.
• Fix the --without-export configure option.

3.6.0

This version contains a critical bug fix for a regression present in 3.5.0 and 3.5.1.

Bug fixes:

• Fix a regression in arena_chunk_alloc() that caused crashes during
  small/large allocation if chunk allocation failed. In the absence of this
  bug, chunk allocation failure would result in allocation failure, e.g. NULL
  return from malloc(). This regression was introduced in 3.5.0.
• Fix backtracing for gcc intrinsics-based backtracing by specifying
  -fno-omit-frame-pointer to gcc. Note that the application (and all the
  libraries it links to) must also be compiled with this option for
  backtracing to be reliable.
• Use dss allocation precedence for huge allocations as well as small/large
  allocations.
• Fix test assertion failure message formatting. This bug did not manifest on
  x86_64 systems because of implementation subtleties in va_list.
• Fix inconsequential test failures for hash and SFMT code.

New features:

• Support heap profiling on FreeBSD. This feature depends on the proc
  filesystem being mounted during heap profile dumping.