Changes in cache_bin.c http://nxr.netbsd.org/rss/src/external/bsd/jemalloc/dist/test/unit/cache_bin.c en Copyright 2005 Java branches: 1.1.1;<br/>Initial revision /src/external/bsd/jemalloc/dist/test/unit/cache_bin.c - 1.1 Mon Sep 23 14:59:44 UTC 2024 christos Import jemalloc-5.3.1 (previous was 5.3.0)<br/><br/>This release includes over 390 commits spanning bug fixes, new features,<br/>performance optimizations, and portability improvements. Multiple percent<br/>of system-level metric improvements were measured in tested production<br/>workloads. The release has gone through large-scale production testing<br/>at Meta.<br/><br/>New features:<br/><br/>Support pvalloc. (@Lapenkov: 5b1f2cc)<br/>Add double free detection for the debug build. (@izaitsevfb:<br/>36366f3, @guangli-dai: 42daa1a, @divanorama: 1897f18)<br/>Add compile-time option --enable-pageid to enable memory mapping<br/>annotation. (@devnexen: 4fc5c4f)<br/>Add runtime option prof_bt_max to control the max stack depth for<br/>profiling. (@guangli-dai: a0734fd)<br/>Add compile-time option --enable-force-getenv to use getenv instead<br/>of secure_getenv. (@interwq: 481bbfc)<br/>Add compile-time option --disable-dss to disable the usage of<br/>sbrk(2). (@Svetlitski: ea5b7be)<br/>Add runtime option tcache_ncached_max to control the number of items<br/>in each size bin in the thread cache. (@guangli-dai: 8a22d10)<br/>Add runtime option calloc_madvise_threshold to determine if kernel or<br/>memset is used to zero the allocations for calloc. (@nullptr0-0:<br/>5081c16)<br/>Add compile-time option --disable-user-config to disable reading the<br/>runtime configurations from /etc/malloc.conf or environment variable<br/>MALLOC_CONF. (@roblabla: c17bf8b)<br/>Add runtime option disable_large_size_classes to guard the new usable<br/>size calculation, which minimizes the memory overhead for large<br/>allocations, i.e., >= 4 * PAGE. (@guangli-dai: c067a55, 8347f10)<br/>Enable process_madvise usage, add runtime option<br/>process_madvise_max_batch to control the max # of regions in each<br/>madvise batch. (@interwq: 22440a0, @spredolac: 4246475)<br/>Add mallctl interfaces:<br/>opt.prof_bt_max (@guangli-dai: a0734fd)<br/>arena.<i>.name to set and get arena names. (@guangli-dai: ba19d2c)<br/>thread.tcache.max to set and get the tcache_max of the current<br/>thread. (@guangli-dai: a442d9b)<br/>thread.tcache.ncached_max.write and<br/>thread.tcache.ncached_max.read_sizeclass to set and get the<br/>ncached_max setup of the current thread. (@guangli-dai: 630f7de,<br/>6b197fd)<br/>arenas.hugepage to return the hugepage size used, also exported to<br/>malloc stats. (@ilvokhin: 90c627e)<br/>approximate_stats.active to return an estimate of the current active<br/>bytes, which should not be compared with other stats retrieved.<br/>(@guangli-dai: 0988583)<br/>Bug fixes:<br/><br/>Prevent potential deadlocks in decaying during reentrancy. (@interwq:<br/>434a68e)<br/>Fix segfault in extent coalescing. (@Svetlitski: 12311fe)<br/>Add null pointer detections in mallctl calls. (@Svetlitski: dc0a184,<br/>0288126)<br/>Make mallctl arenas.lookup triable without crashing on invalid<br/>pointers. (@auxten: 019cccc, 5bac384)<br/>Demote sampled allocations for proper deallocations during<br/>arena_reset. (@Svetlitski: 62648c8)<br/>Fix jemalloc's read(2) and write(2). (@Svetlitski: d2c9ed3, @lexprfuncall:<br/>9fdc116)<br/>Fix the pkg-config metadata file. (@BtbN: ed7e6fe, ce8ce99)<br/>Fix the autogen.sh so that it accepts quoted extra options.<br/>(@honggyukim: f6fe6ab)<br/>Fix rallocx() to set errno to ENOMEM upon OOMing. (@arter97: 38056fe,<br/>@interwq: 83b0757)<br/>Avoid stack overflow for internal variable array usage. (@nullptr0-0:<br/>47c9bcd, 48f66cf, @xinydev: 9169e92)<br/>Fix background thread initialization race. (@puzpuzpuz: 4d0ffa0)<br/>Guard os_page_id against a NULL address. (@lexprfuncall: 79cc7dc)<br/>Handle tcache init failures gracefully. (@lexprfuncall: a056c20)<br/>Fix missing release of acquired neighbor edata in<br/>extent_try_coalesce_impl. (@spredolac: 675ab07)<br/>Fix memory leak of old curr_reg on san_bump_grow_locked failure.<br/>(@spredolac: 5904a42)<br/>Fix large alloc nrequests under-counting on cache misses. (@spredolac:<br/>3cc56d3)<br/>Portability improvements:<br/><br/>Fix the build in C99. (@abaelhe: 56ddbea)<br/>Add pthread_setaffinity_np detection for non Linux/BSD platforms.<br/>(@devnexen: 4c95c95)<br/>Make VARIABLE_ARRAY compatible with compilers not supporting VLA,<br/>i.e., Visual Studio C compiler in C11 or C17 modes. (@madscientist:<br/>be65438)<br/>Fix the build on Linux using musl library. (@marv: aba1645, 45249cf)<br/>Reduce the memory overhead in small allocation sampling for systems<br/>with larger page sizes, e.g., ARM. (@Svetlitski: 5a858c6)<br/>Add C23's free_sized and free_aligned_sized. (@Svetlitski:<br/>cdb2c0e)<br/>Enable heap profiling on MacOS. (@nullptr0-0: 4b555c1)<br/>Fix incorrect printing on 32bit. (@sundb: 630434b)<br/>Make JEMALLOC_CXX_THROW compatible with C++ versions newer than<br/>C++17. (@r-barnes, @guangli-dai: 21bcc0a)<br/>Fix mmap tag conflicts on MacOS. (@kdrag0n: c893fcd)<br/>Fix monotonic timer assumption for win32. (@burtonli: 8dc97b1)<br/>Fix VM over-reservation on systems with larger pages, e.g., aarch64.<br/>(@interwq: cd05b19)<br/>Remove unreachable() macro conditionally to prevent definition<br/>conflicts for C23+. (@appujee: d8486b2, 4b88bdd)<br/>Fix dlsym failure observed on FreeBSD. (@rhelmot: 86bbaba)<br/>Change the default page size to 64KB on aarch64 Linux. (@lexprfuncall:<br/>9442300)<br/>Update config.guess and config.sub to the latest version.<br/>(@lexprfuncall: c51949e)<br/>Determine the page size on Android from NDK header files.<br/>(@lexprfuncall: c51abba)<br/>Improve the portability of grep patterns in configure.ac.<br/>(@lexprfuncall: 365747b)<br/>Add compile-time option --with-cxx-stdlib to specify the C++ standard<br/>library. (@yuxuanchen1997: a10ef3e)<br/>Optimizations and refactors:<br/><br/>Enable tcache for deallocation-only threads. (@interwq: 143e9c4)<br/>Inline to accelerate operator delete. (@guangli-dai: e8f9f13)<br/>Optimize pairing heap's performance. (@deadalnix: 5266152, be6da4f,<br/>543e2d6, 10d7131, 92aa52c, @Svetlitski: 36ca0c1)<br/>Inline the storage for thread name in the profiling data. (@interwq:<br/>ce0b7ab, e62aa47)<br/>Optimize a hot function edata_cmp_summary_comp to accelerate it.<br/>(@Svetlitski: 6841110, @guangli-dai: 0181aaa)<br/>Allocate thread cache using the base allocator, which enables thread<br/>cache to use thp when metadata_thp is turned on. (@interwq:<br/>72cfdce)<br/>Allow oversize arena not to purge immediately when background threads<br/>are enabled, although the default decay time is 0 to be back compatible.<br/>(@interwq: d131331)<br/>Optimize thread-local storage implementation on Windows. (@mcfi:<br/>9e123a8, 3a0d9cd)<br/>Optimize fast path to allow static size class computation. (@interwq:<br/>323ed2e)<br/>Redesign tcache GC to regulate the frequency and make it<br/>locality-aware. The new design is default on, guarded by option<br/>experimental_tcache_gc. (@nullptr0-0: 0c88be9, e2c9f3a,<br/>14d5dc1, @deadalnix: 5afff2e)<br/>Reduce the arena switching overhead by avoiding forced purging when<br/>background thread is enabled. (@interwq: a3910b9)<br/>Improve the reuse efficiency by limiting the maximum coalesced size for<br/>large extents. (@jiebinn: 3c14707)<br/>Refactor thread events to allow registration of users' thread events<br/>and remove prof_threshold as the built-in event. (@spredolac: e6864c6,<br/>015b017, 34ace91)<br/>Documentation:<br/><br/>Update Windows building instructions. (@Lapenkov: 3713932)<br/>Add vcpkg installation instructions. (@LilyWangLL: c0c9783)<br/>Update profiling internals with an example. (@jordalgo: b04e766) /src/external/bsd/jemalloc/dist/test/unit/cache_bin.c - 1.1.1.2 Sun Apr 19 16:20:55 UTC 2026 christos branches: 1.1.1.1.4;<br/>Import jemalloc-5.3.0 (previous was 5.1.0)<br/><br/>* 5.3.0 (May 6, 2022)<br/><br/> This release contains many speed and space optimizations, from micro<br/> optimizations on common paths to rework of internal data structures and<br/> locking schemes, and many more too detailed to list below. Multiple percent<br/> of system level metric improvements were measured in tested production<br/> workloads. The release has gone through large-scale production testing.<br/><br/> New features:<br/> - Add the thread.idle mallctl which hints that the calling thread will be<br/> idle for a nontrivial period of time. (@davidtgoldblatt)<br/> - Allow small size classes to be the maximum size class to cache in the<br/> thread-specific cache, through the opt.[lg_]tcache_max option. (@interwq,<br/> @jordalgo)<br/> - Make the behavior of realloc(ptr, 0) configurable with opt.zero_realloc.<br/> (@davidtgoldblatt)<br/> - Add 'make uninstall' support. (@sangshuduo, @Lapenkov)<br/> - Support C++17 over-aligned allocation. (@marksantaniello)<br/> - Add the thread.peak mallctl for approximate per-thread peak memory tracking.<br/> (@davidtgoldblatt)<br/> - Add interval-based stats output opt.stats_interval. (@interwq)<br/> - Add prof.prefix to override filename prefixes for dumps. (@zhxchen17)<br/> - Add high resolution timestamp support for profiling. (@tyroguru)<br/> - Add the --collapsed flag to jeprof for flamegraph generation.<br/> (@igorwwwwwwwwwwwwwwwwwwww)<br/> - Add the --debug-syms-by-id option to jeprof for debug symbols discovery.<br/> (@DeannaGelbart)<br/> - Add the opt.prof_leak_error option to exit with error code when leak is<br/> detected using opt.prof_final. (@yunxuo)<br/> - Add opt.cache_oblivious as an runtime alternative to config.cache_oblivious.<br/> (@interwq)<br/> - Add mallctl interfaces:<br/> + opt.zero_realloc (@davidtgoldblatt)<br/> + opt.cache_oblivious (@interwq)<br/> + opt.prof_leak_error (@yunxuo)<br/> + opt.stats_interval (@interwq)<br/> + opt.stats_interval_opts (@interwq)<br/> + opt.tcache_max (@interwq)<br/> + opt.trust_madvise (@azat)<br/> + prof.prefix (@zhxchen17)<br/> + stats.zero_reallocs (@davidtgoldblatt)<br/> + thread.idle (@davidtgoldblatt)<br/> + thread.peak.{read,reset} (@davidtgoldblatt)<br/><br/> Bug fixes:<br/> - Fix the synchronization around explicit tcache creation which could cause<br/> invalid tcache identifiers. This regression was first released in 5.0.0.<br/> (@yoshinorim, @davidtgoldblatt)<br/> - Fix a profiling biasing issue which could cause incorrect heap usage and<br/> object counts. This issue existed in all previous releases with the heap<br/> profiling feature. (@davidtgoldblatt)<br/> - Fix the order of stats counter updating on large realloc which could cause<br/> failed assertions. This regression was first released in 5.0.0. (@azat)<br/> - Fix the locking on the arena destroy mallctl, which could cause concurrent<br/> arena creations to fail. This functionality was first introduced in 5.0.0.<br/> (@interwq)<br/><br/> Portability improvements:<br/> - Remove nothrow from system function declarations on macOS and FreeBSD.<br/> (@davidtgoldblatt, @fredemmott, @leres)<br/> - Improve overcommit and page alignment settings on NetBSD. (@zoulasc)<br/> - Improve CPU affinity support on BSD platforms. (@devnexen)<br/> - Improve utrace detection and support. (@devnexen)<br/> - Improve QEMU support with MADV_DONTNEED zeroed pages detection. (@azat)<br/> - Add memcntl support on Solaris / illumos. (@devnexen)<br/> - Improve CPU_SPINWAIT on ARM. (@AWSjswinney)<br/> - Improve TSD cleanup on FreeBSD. (@Lapenkov)<br/> - Disable percpu_arena if the CPU count cannot be reliably detected. (@azat)<br/> - Add malloc_size(3) override support. (@devnexen)<br/> - Add mmap VM_MAKE_TAG support. (@devnexen)<br/> - Add support for MADV_[NO]CORE. (@devnexen)<br/> - Add support for DragonFlyBSD. (@devnexen)<br/> - Fix the QUANTUM setting on MIPS64. (@brooksdavis)<br/> - Add the QUANTUM setting for ARC. (@vineetgarc)<br/> - Add the QUANTUM setting for LoongArch. (@wangjl-uos)<br/> - Add QNX support. (@jqian-aurora)<br/> - Avoid atexit(3) calls unless the relevant profiling features are enabled.<br/> (@BusyJay, @laiwei-rice, @interwq)<br/> - Fix unknown option detection when using Clang. (@Lapenkov)<br/> - Fix symbol conflict with musl libc. (@georgthegreat)<br/> - Add -Wimplicit-fallthrough checks. (@nickdesaulniers)<br/> - Add __forceinline support on MSVC. (@santagada)<br/> - Improve FreeBSD and Windows CI support. (@Lapenkov)<br/> - Add CI support for PPC64LE architecture. (@ezeeyahoo)<br/><br/> Incompatible changes:<br/> - Maximum size class allowed in tcache (opt.[lg_]tcache_max) now has an upper<br/> bound of 8MiB. (@interwq)<br/><br/> Optimizations and refactors (@davidtgoldblatt, @Lapenkov, @interwq):<br/> - Optimize the common cases of the thread cache operations.<br/> - Optimize internal data structures, including RB tree and pairing heap.<br/> - Optimize the internal locking on extent management.<br/> - Extract and refactor the internal page allocator and interface modules.<br/><br/> Documentation:<br/> - Fix doc build with --with-install-suffix. (@lawmurray, @interwq)<br/> - Add PROFILING_INTERNALS.md. (@davidtgoldblatt)<br/> - Ensure the proper order of doc building and installation. (@Mingli-Yu)<br/><br/>* 5.2.1 (August 5, 2019)<br/><br/> This release is primarily about Windows. A critical virtual memory leak is<br/> resolved on all Windows platforms. The regression was present in all releases<br/> since 5.0.0.<br/><br/> Bug fixes:<br/> - Fix a severe virtual memory leak on Windows. This regression was first<br/> released in 5.0.0. (@Ignition, @j0t, @frederik-h, @davidtgoldblatt,<br/> @interwq)<br/> - Fix size 0 handling in posix_memalign(). This regression was first released<br/> in 5.2.0. (@interwq)<br/> - Fix the prof_log unit test which may observe unexpected backtraces from<br/> compiler optimizations. The test was first added in 5.2.0. (@marxin,<br/> @gnzlbg, @interwq)<br/> - Fix the declaration of the extent_avail tree. This regression was first<br/> released in 5.1.0. (@zoulasc)<br/> - Fix an incorrect reference in jeprof. This functionality was first released<br/> in 3.0.0. (@prehistoric-penguin)<br/> - Fix an assertion on the deallocation fast-path. This regression was first<br/> released in 5.2.0. (@yinan1048576)<br/> - Fix the TLS_MODEL attribute in headers. This regression was first released<br/> in 5.0.0. (@zoulasc, @interwq)<br/><br/> Optimizations and refactors:<br/> - Implement opt.retain on Windows and enable by default on 64-bit. (@interwq,<br/> @davidtgoldblatt)<br/> - Optimize away a branch on the operator delete[] path. (@mgrice)<br/> - Add format annotation to the format generator function. (@zoulasc)<br/> - Refactor and improve the size class header generation. (@yinan1048576)<br/> - Remove best fit. (@djwatson)<br/> - Avoid blocking on background thread locks for stats. (@oranagra, @interwq)<br/><br/>* 5.2.0 (April 2, 2019)<br/><br/> This release includes a few notable improvements, which are summarized below:<br/> 1) improved fast-path performance from the optimizations by @djwatson; 2)<br/> reduced virtual memory fragmentation and metadata usage; and 3) bug fixes on<br/> setting the number of background threads. In addition, peak / spike memory<br/> usage is improved with certain allocation patterns. As usual, the release and<br/> prior dev versions have gone through large-scale production testing.<br/><br/> New features:<br/> - Implement oversize_threshold, which uses a dedicated arena for allocations<br/> crossing the specified threshold to reduce fragmentation. (@interwq)<br/> - Add extents usage information to stats. (@tyleretzel)<br/> - Log time information for sampled allocations. (@tyleretzel)<br/> - Support 0 size in sdallocx. (@djwatson)<br/> - Output rate for certain counters in malloc_stats. (@zinoale)<br/> - Add configure option --enable-readlinkat, which allows the use of readlinkat<br/> over readlink. (@davidtgoldblatt)<br/> - Add configure options --{enable,disable}-{static,shared} to allow not<br/> building unwanted libraries. (@Ericson2314)<br/> - Add configure option --disable-libdl to enable fully static builds.<br/> (@interwq)<br/> - Add mallctl interfaces:<br/> + opt.oversize_threshold (@interwq)<br/> + stats.arenas.<i>.extent_avail (@tyleretzel)<br/> + stats.arenas.<i>.extents.<j>.n{dirty,muzzy,retained} (@tyleretzel)<br/> + stats.arenas.<i>.extents.<j>.{dirty,muzzy,retained}_bytes<br/> (@tyleretzel)<br/><br/> Portability improvements:<br/> - Update MSVC builds. (@maksqwe, @rustyx)<br/> - Workaround a compiler optimizer bug on s390x. (@rkmisra)<br/> - Make use of pthread_set_name_np(3) on FreeBSD. (@trasz)<br/> - Implement malloc_getcpu() to enable percpu_arena for windows. (@santagada)<br/> - Link against -pthread instead of -lpthread. (@paravoid)<br/> - Make background_thread not dependent on libdl. (@interwq)<br/> - Add stringify to fix a linker directive issue on MSVC. (@daverigby)<br/> - Detect and fall back when 8-bit atomics are unavailable. (@interwq)<br/> - Fall back to the default pthread_create if dlsym(3) fails. (@interwq)<br/><br/> Optimizations and refactors:<br/> - Refactor the TSD module. (@davidtgoldblatt)<br/> - Avoid taking extents_muzzy mutex when muzzy is disabled. (@interwq)<br/> - Avoid taking large_mtx for auto arenas on the tcache flush path. (@interwq)<br/> - Optimize ixalloc by avoiding a size lookup. (@interwq)<br/> - Implement opt.oversize_threshold which uses a dedicated arena for requests<br/> crossing the threshold, also eagerly purges the oversize extents. Default<br/> the threshold to 8 MiB. (@interwq)<br/> - Clean compilation with -Wextra. (@gnzlbg, @jasone)<br/> - Refactor the size class module. (@davidtgoldblatt)<br/> - Refactor the stats emitter. (@tyleretzel)<br/> - Optimize pow2_ceil. (@rkmisra)<br/> - Avoid runtime detection of lazy purging on FreeBSD. (@trasz)<br/> - Optimize mmap(2) alignment handling on FreeBSD. (@trasz)<br/> - Improve error handling for THP state initialization. (@jsteemann)<br/> - Rework the malloc() fast path. (@djwatson)<br/> - Rework the free() fast path. (@djwatson)<br/> - Refactor and optimize the tcache fill / flush paths. (@djwatson)<br/> - Optimize sync / lwsync on PowerPC. (@chmeeedalf)<br/> - Bypass extent_dalloc() when retain is enabled. (@interwq)<br/> - Optimize the locking on large deallocation. (@interwq)<br/> - Reduce the number of pages committed from sanity checking in debug build.<br/> (@trasz, @interwq)<br/> - Deprecate OSSpinLock. (@interwq)<br/> - Lower the default number of background threads to 4 (when the feature<br/> is enabled). (@interwq)<br/> - Optimize the trylock spin wait. (@djwatson)<br/> - Use arena index for arena-matching checks. (@interwq)<br/> - Avoid forced decay on thread termination when using background threads.<br/> (@interwq)<br/> - Disable muzzy decay by default. (@djwatson, @interwq)<br/> - Only initialize libgcc unwinder when profiling is enabled. (@paravoid,<br/> @interwq)<br/><br/> Bug fixes (all only relevant to jemalloc 5.x):<br/> - Fix background thread index issues with max_background_threads. (@djwatson,<br/> @interwq)<br/> - Fix stats output for opt.lg_extent_max_active_fit. (@interwq)<br/> - Fix opt.prof_prefix initialization. (@davidtgoldblatt)<br/> - Properly trigger decay on tcache destroy. (@interwq, @amosbird)<br/> - Fix tcache.flush. (@interwq)<br/> - Detect whether explicit extent zero out is necessary with huge pages or<br/> custom extent hooks, which may change the purge semantics. (@interwq)<br/> - Fix a side effect caused by extent_max_active_fit combined with decay-based<br/> purging, where freed extents can accumulate and not be reused for an<br/> extended period of time. (@interwq, @mpghf)<br/> - Fix a missing unlock on extent register error handling. (@zoulasc)<br/><br/> Testing:<br/> - Simplify the Travis script output. (@gnzlbg)<br/> - Update the test scripts for FreeBSD. (@devnexen)<br/> - Add unit tests for the producer-consumer pattern. (@interwq)<br/> - Add Cirrus-CI config for FreeBSD builds. (@jasone)<br/> - Add size-matching sanity checks on tcache flush. (@davidtgoldblatt,<br/> @interwq)<br/><br/> Incompatible changes:<br/> - Remove --with-lg-page-sizes. (@davidtgoldblatt)<br/><br/> Documentation:<br/> - Attempt to build docs by default, however skip doc building when xsltproc<br/> is missing. (@interwq, @cmuellner) /src/external/bsd/jemalloc/dist/test/unit/cache_bin.c - 1.1.1.1 Mon Sep 23 14:59:44 UTC 2024 christos Sync with HEAD /src/external/bsd/jemalloc/dist/test/unit/cache_bin.c - 1.1.1.1.4.2 Sat Aug 02 05:21:23 UTC 2025 perseant file cache_bin.c was added on branch perseant-exfatfs on 2025-08-02 05:21:23 +0000 /src/external/bsd/jemalloc/dist/test/unit/cache_bin.c - 1.1.1.1.4.1 Mon Sep 23 14:59:44 UTC 2024 perseant