From: tfheen Date: Wed, 24 Sep 2008 12:41:27 +0000 (+0000) Subject: Import jemalloc X-Git-Url: https://err.no/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=609c38f4c792c094de44a8c653cabed0a11dc4cd;p=varnish Import jemalloc git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@3215 d4fa192b-c00b-0410-8231-f00ffab90ce4 --- diff --git a/varnish-cache/lib/libjemalloc/Makefile b/varnish-cache/lib/libjemalloc/Makefile new file mode 100644 index 00000000..49f4197d --- /dev/null +++ b/varnish-cache/lib/libjemalloc/Makefile @@ -0,0 +1,23 @@ +CFLAGS := -O3 -g +# See source code comments to avoid memory leaks when enabling MALLOC_MAG. +#CPPFLAGS := -DMALLOC_PRODUCTION -DMALLOC_MAG +CPPFLAGS := -DMALLOC_PRODUCTION + +all: libjemalloc.so.0 libjemalloc_mt.so.0 + +jemalloc_linux_mt.o: jemalloc_linux.c + gcc $(CFLAGS) -c -DPIC -fPIC $(CPPFLAGS) -D__isthreaded=true -o $@ $+ + +jemalloc_linux.o: jemalloc_linux.c + gcc $(CFLAGS) -c -DPIC -fPIC $(CPPFLAGS) -D__isthreaded=false -o $@ $+ + +libjemalloc_mt.so.0: jemalloc_linux_mt.o + gcc -shared -lpthread -o $@ $+ + ln -sf $@ libjemalloc_mt.so + +libjemalloc.so.0: jemalloc_linux.o + gcc -shared -lpthread -o $@ $+ + ln -sf $@ libjemalloc.so + +clean: + rm -f *.o *.so.0 *.so diff --git a/varnish-cache/lib/libjemalloc/README b/varnish-cache/lib/libjemalloc/README new file mode 100644 index 00000000..5b809976 --- /dev/null +++ b/varnish-cache/lib/libjemalloc/README @@ -0,0 +1,55 @@ +This is a minimal-effort stand-alone jemalloc distribution for Linux. The main +rough spots are: + +* __isthreaded must be hard-coded, since the pthreads library really needs to + be involved in order to toggle it at run time. Therefore, this distribution + builds two separate libraries: + + + libjemalloc_mt.so.0 : Use for multi-threaded applications. + + libjemalloc.so.0 : Use for single-threaded applications. + + Both libraries link against libpthread, though with a bit more code hacking, + this dependency could be removed for the single-threaded version. + +* MALLOC_MAG (thread-specific caching, using magazines) is disabled, because + special effort is required to avoid memory leaks when it is enabled. To make + cleanup automatic, we would need help from the pthreads library. If you + enable MALLOC_MAG, be sure to call _malloc_thread_cleanup() in each thread + just before it exits. + +* The code that determines the number of CPUs is sketchy. The trouble is that + we must avoid any memory allocation during early initialization. + +In order to build: + + make + +This generates two shared libraries, which you can either link against, or +pre-load. + +Linking and running, where /path/to is the path to libjemalloc (-lpthread +required even for libjemalloc.so): + + gcc app.o -o app -L/path/to -ljemalloc_mt -lpthread + LD_LIBRARY_PATH=/path/to app + +Pre-loading: + + LD_PRELOAD=/path/to/libjemalloc_mt.so.0 app + +jemalloc has a lot of run-time tuning options. See the man page for details: + + nroff -man malloc.3 | less + +In particular, take a look at the B, F, and N options. If you enable +MALLOC_MAG, look at the G and R options. + +If your application is crashing, or performance seems to be lacking, enable +assertions and statistics gathering by removing MALLOC_PRODUCTION from CPPFLAGS +in the Makefile. In order to print a statistics summary at program exit, run +your application like: + + LD_PRELOAD=/path/to/libjemalloc_mt.so.0 MALLOC_OPTIONS=P app + +Please contact Jason Evans with questions, comments, bug +reports, etc. diff --git a/varnish-cache/lib/libjemalloc/jemalloc_linux.c b/varnish-cache/lib/libjemalloc/jemalloc_linux.c new file mode 100644 index 00000000..3b5d2655 --- /dev/null +++ b/varnish-cache/lib/libjemalloc/jemalloc_linux.c @@ -0,0 +1,5670 @@ +/*- + * Copyright (C) 2006-2008 Jason Evans . + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice(s), this list of conditions and the following disclaimer as + * the first lines of this file unmodified other than the possible + * addition of one or more copyright notices. + * 2. Redistributions in binary form must reproduce the above copyright + * notice(s), this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE + * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR + * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, + * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE + * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, + * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + ******************************************************************************* + * + * This allocator implementation is designed to provide scalable performance + * for multi-threaded programs on multi-processor systems. The following + * features are included for this purpose: + * + * + Multiple arenas are used if there are multiple CPUs, which reduces lock + * contention and cache sloshing. + * + * + Thread-specific caching is used if there are multiple threads, which + * reduces the amount of locking. + * + * + Cache line sharing between arenas is avoided for internal data + * structures. + * + * + Memory is managed in chunks and runs (chunks can be split into runs), + * rather than as individual pages. This provides a constant-time + * mechanism for associating allocations with particular arenas. + * + * Allocation requests are rounded up to the nearest size class, and no record + * of the original request size is maintained. Allocations are broken into + * categories according to size class. Assuming runtime defaults, 4 kB pages + * and a 16 byte quantum on a 32-bit system, the size classes in each category + * are as follows: + * + * |=======================================| + * | Category | Subcategory | Size | + * |=======================================| + * | Small | Tiny | 2 | + * | | | 4 | + * | | | 8 | + * | |------------------+---------| + * | | Quantum-spaced | 16 | + * | | | 32 | + * | | | 48 | + * | | | ... | + * | | | 96 | + * | | | 112 | + * | | | 128 | + * | |------------------+---------| + * | | Cacheline-spaced | 192 | + * | | | 256 | + * | | | 320 | + * | | | 384 | + * | | | 448 | + * | | | 512 | + * | |------------------+---------| + * | | Sub-page | 760 | + * | | | 1024 | + * | | | 1280 | + * | | | ... | + * | | | 3328 | + * | | | 3584 | + * | | | 3840 | + * |=======================================| + * | Large | 4 kB | + * | | 8 kB | + * | | 12 kB | + * | | ... | + * | | 1012 kB | + * | | 1016 kB | + * | | 1020 kB | + * |=======================================| + * | Huge | 1 MB | + * | | 2 MB | + * | | 3 MB | + * | | ... | + * |=======================================| + * + * A different mechanism is used for each category: + * + * Small : Each size class is segregated into its own set of runs. Each run + * maintains a bitmap of which regions are free/allocated. + * + * Large : Each allocation is backed by a dedicated run. Metadata are stored + * in the associated arena chunk header maps. + * + * Huge : Each allocation is backed by a dedicated contiguous set of chunks. + * Metadata are stored in a separate red-black tree. + * + ******************************************************************************* + */ + +/* + * Set to false if single-threaded. Even better, rip out all of the code that + * doesn't get used if __isthreaded is false, so that libpthread isn't + * necessary. + */ +#ifndef __isthreaded +# define __isthreaded true +#endif + +/* + * MALLOC_PRODUCTION disables assertions and statistics gathering. It also + * defaults the A and J runtime options to off. These settings are appropriate + * for production systems. + */ +/* #define MALLOC_PRODUCTION */ + +#ifndef MALLOC_PRODUCTION + /* + * MALLOC_DEBUG enables assertions and other sanity checks, and disables + * inline functions. + */ +# define MALLOC_DEBUG + + /* MALLOC_STATS enables statistics calculation. */ +# define MALLOC_STATS +#endif + +/* + * MALLOC_TINY enables support for tiny objects, which are smaller than one + * quantum. + */ +#define MALLOC_TINY + +/* + * MALLOC_MAG enables a magazine-based thread-specific caching layer for small + * objects. This makes it possible to allocate/deallocate objects without any + * locking when the cache is in the steady state. + * + * If MALLOC_MAG is enabled, make sure that _malloc_thread_cleanup() is called + * by each thread just before it exits. + */ +/* #define MALLOC_MAG */ + +/* + * MALLOC_BALANCE enables monitoring of arena lock contention and dynamically + * re-balances arena load if exponentially averaged contention exceeds a + * certain threshold. + */ +#define MALLOC_BALANCE + +/* + * MALLOC_DSS enables use of sbrk(2) to allocate chunks from the data storage + * segment (DSS). In an ideal world, this functionality would be completely + * unnecessary, but we are burdened by history and the lack of resource limits + * for anonymous mapped memory. + */ +/* #define MALLOC_DSS */ + +#define _GNU_SOURCE /* For mremap(2). */ +#define issetugid() 0 +#define __DECONST(type, var) ((type)(uintptr_t)(const void *)(var)) + +/* __FBSDID("$FreeBSD: head/lib/libc/stdlib/malloc.c 182225 2008-08-27 02:00:53Z jasone $"); */ + +#include +#include +#include +#include +#include +#include + +#include +#include +#ifndef SIZE_T_MAX +# define SIZE_T_MAX SIZE_MAX +#endif +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "rb.h" + +#ifdef MALLOC_DEBUG + /* Disable inlining to make debugging easier. */ +# define inline +#endif + +/* Size of stack-allocated buffer passed to strerror_r(). */ +#define STRERROR_BUF 64 + +/* + * The const_size2bin table is sized according to PAGESIZE_2POW, but for + * correctness reasons, we never assume that + * (pagesize == (1U << * PAGESIZE_2POW)). + * + * Minimum alignment of allocations is 2^QUANTUM_2POW bytes. + */ +#ifdef __i386__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 2 +# define CPU_SPINWAIT __asm__ volatile("pause") +#endif +#ifdef __ia64__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +#endif +#ifdef __alpha__ +# define PAGESIZE_2POW 13 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define NO_TLS +#endif +#ifdef __sparc64__ +# define PAGESIZE_2POW 13 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define NO_TLS +#endif +#ifdef __amd64__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define CPU_SPINWAIT __asm__ volatile("pause") +#endif +#ifdef __arm__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 3 +# define SIZEOF_PTR_2POW 2 +# define NO_TLS +#endif +#ifdef __mips__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 3 +# define SIZEOF_PTR_2POW 2 +# define NO_TLS +#endif +#ifdef __powerpc__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 2 +#endif + +#define QUANTUM ((size_t)(1U << QUANTUM_2POW)) +#define QUANTUM_MASK (QUANTUM - 1) + +#define SIZEOF_PTR (1U << SIZEOF_PTR_2POW) + +/* sizeof(int) == (1U << SIZEOF_INT_2POW). */ +#ifndef SIZEOF_INT_2POW +# define SIZEOF_INT_2POW 2 +#endif + +/* We can't use TLS in non-PIC programs, since TLS relies on loader magic. */ +#if (!defined(PIC) && !defined(NO_TLS)) +# define NO_TLS +#endif + +#ifdef NO_TLS + /* MALLOC_MAG requires TLS. */ +# ifdef MALLOC_MAG +# undef MALLOC_MAG +# endif + /* MALLOC_BALANCE requires TLS. */ +# ifdef MALLOC_BALANCE +# undef MALLOC_BALANCE +# endif +#endif + +/* + * Size and alignment of memory chunks that are allocated by the OS's virtual + * memory system. + */ +#define CHUNK_2POW_DEFAULT 20 + +/* Maximum number of dirty pages per arena. */ +#define DIRTY_MAX_DEFAULT (1U << 9) + +/* + * Maximum size of L1 cache line. This is used to avoid cache line aliasing. + * In addition, this controls the spacing of cacheline-spaced size classes. + */ +#define CACHELINE_2POW 6 +#define CACHELINE ((size_t)(1U << CACHELINE_2POW)) +#define CACHELINE_MASK (CACHELINE - 1) + +/* + * Subpages are an artificially designated partitioning of pages. Their only + * purpose is to support subpage-spaced size classes. + * + * There must be at least 4 subpages per page, due to the way size classes are + * handled. + */ +#define SUBPAGE_2POW 8 +#define SUBPAGE ((size_t)(1U << SUBPAGE_2POW)) +#define SUBPAGE_MASK (SUBPAGE - 1) + +#ifdef MALLOC_TINY + /* Smallest size class to support. */ +# define TINY_MIN_2POW 1 +#endif + +/* + * Maximum size class that is a multiple of the quantum, but not (necessarily) + * a power of 2. Above this size, allocations are rounded up to the nearest + * power of 2. + */ +#define QSPACE_MAX_2POW_DEFAULT 7 + +/* + * Maximum size class that is a multiple of the cacheline, but not (necessarily) + * a power of 2. Above this size, allocations are rounded up to the nearest + * power of 2. + */ +#define CSPACE_MAX_2POW_DEFAULT 9 + +/* + * RUN_MAX_OVRHD indicates maximum desired run header overhead. Runs are sized + * as small as possible such that this setting is still honored, without + * violating other constraints. The goal is to make runs as small as possible + * without exceeding a per run external fragmentation threshold. + * + * We use binary fixed point math for overhead computations, where the binary + * point is implicitly RUN_BFP bits to the left. + * + * Note that it is possible to set RUN_MAX_OVRHD low enough that it cannot be + * honored for some/all object sizes, since there is one bit of header overhead + * per object (plus a constant). This constraint is relaxed (ignored) for runs + * that are so small that the per-region overhead is greater than: + * + * (RUN_MAX_OVRHD / (reg_size << (3+RUN_BFP)) + */ +#define RUN_BFP 12 +/* \/ Implicit binary fixed point. */ +#define RUN_MAX_OVRHD 0x0000003dU +#define RUN_MAX_OVRHD_RELAX 0x00001800U + +/* Put a cap on small object run size. This overrides RUN_MAX_OVRHD. */ +#define RUN_MAX_SMALL (12 * pagesize) + +/* + * Hyper-threaded CPUs may need a special instruction inside spin loops in + * order to yield to another virtual CPU. If no such instruction is defined + * above, make CPU_SPINWAIT a no-op. + */ +#ifndef CPU_SPINWAIT +# define CPU_SPINWAIT +#endif + +/* + * Adaptive spinning must eventually switch to blocking, in order to avoid the + * potential for priority inversion deadlock. Backing off past a certain point + * can actually waste time. + */ +#define SPIN_LIMIT_2POW 11 + +/* + * Conversion from spinning to blocking is expensive; we use (1U << + * BLOCK_COST_2POW) to estimate how many more times costly blocking is than + * worst-case spinning. + */ +#define BLOCK_COST_2POW 4 + +#ifdef MALLOC_MAG + /* + * Default magazine size, in bytes. max_rounds is calculated to make + * optimal use of the space, leaving just enough room for the magazine + * header. + */ +# define MAG_SIZE_2POW_DEFAULT 9 +#endif + +#ifdef MALLOC_BALANCE + /* + * We use an exponential moving average to track recent lock contention, + * where the size of the history window is N, and alpha=2/(N+1). + * + * Due to integer math rounding, very small values here can cause + * substantial degradation in accuracy, thus making the moving average decay + * faster than it would with precise calculation. + */ +# define BALANCE_ALPHA_INV_2POW 9 + + /* + * Threshold value for the exponential moving contention average at which to + * re-assign a thread. + */ +# define BALANCE_THRESHOLD_DEFAULT (1U << (SPIN_LIMIT_2POW-4)) +#endif + +/******************************************************************************/ + +typedef pthread_mutex_t malloc_mutex_t; +typedef pthread_mutex_t malloc_spinlock_t; + +/* Set to true once the allocator has been initialized. */ +static bool malloc_initialized = false; + +/* Used to avoid initialization races. */ +static malloc_mutex_t init_lock = PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP; + +/******************************************************************************/ +/* + * Statistics data structures. + */ + +#ifdef MALLOC_STATS + +typedef struct malloc_bin_stats_s malloc_bin_stats_t; +struct malloc_bin_stats_s { + /* + * Number of allocation requests that corresponded to the size of this + * bin. + */ + uint64_t nrequests; + +#ifdef MALLOC_MAG + /* Number of magazine reloads from this bin. */ + uint64_t nmags; +#endif + + /* Total number of runs created for this bin's size class. */ + uint64_t nruns; + + /* + * Total number of runs reused by extracting them from the runs tree for + * this bin's size class. + */ + uint64_t reruns; + + /* High-water mark for this bin. */ + unsigned long highruns; + + /* Current number of runs in this bin. */ + unsigned long curruns; +}; + +typedef struct arena_stats_s arena_stats_t; +struct arena_stats_s { + /* Number of bytes currently mapped. */ + size_t mapped; + + /* + * Total number of purge sweeps, total number of madvise calls made, + * and total pages purged in order to keep dirty unused memory under + * control. + */ + uint64_t npurge; + uint64_t nmadvise; + uint64_t purged; + + /* Per-size-category statistics. */ + size_t allocated_small; + uint64_t nmalloc_small; + uint64_t ndalloc_small; + + size_t allocated_large; + uint64_t nmalloc_large; + uint64_t ndalloc_large; + +#ifdef MALLOC_BALANCE + /* Number of times this arena reassigned a thread due to contention. */ + uint64_t nbalance; +#endif +}; + +typedef struct chunk_stats_s chunk_stats_t; +struct chunk_stats_s { + /* Number of chunks that were allocated. */ + uint64_t nchunks; + + /* High-water mark for number of chunks allocated. */ + unsigned long highchunks; + + /* + * Current number of chunks allocated. This value isn't maintained for + * any other purpose, so keep track of it in order to be able to set + * highchunks. + */ + unsigned long curchunks; +}; + +#endif /* #ifdef MALLOC_STATS */ + +/******************************************************************************/ +/* + * Extent data structures. + */ + +/* Tree of extents. */ +typedef struct extent_node_s extent_node_t; +struct extent_node_s { +#ifdef MALLOC_DSS + /* Linkage for the size/address-ordered tree. */ + rb_node(extent_node_t) link_szad; +#endif + + /* Linkage for the address-ordered tree. */ + rb_node(extent_node_t) link_ad; + + /* Pointer to the extent that this tree node is responsible for. */ + void *addr; + + /* Total region size. */ + size_t size; +}; +typedef rb_tree(extent_node_t) extent_tree_t; + +/******************************************************************************/ +/* + * Arena data structures. + */ + +typedef struct arena_s arena_t; +typedef struct arena_bin_s arena_bin_t; + +/* Each element of the chunk map corresponds to one page within the chunk. */ +typedef struct arena_chunk_map_s arena_chunk_map_t; +struct arena_chunk_map_s { + /* + * Linkage for run trees. There are two disjoint uses: + * + * 1) arena_t's runs_avail tree. + * 2) arena_run_t conceptually uses this linkage for in-use non-full + * runs, rather than directly embedding linkage. + */ + rb_node(arena_chunk_map_t) link; + + /* + * Run address (or size) and various flags are stored together. The bit + * layout looks like (assuming 32-bit system): + * + * ???????? ???????? ????---- ---kdzla + * + * ? : Unallocated: Run address for first/last pages, unset for internal + * pages. + * Small: Run address. + * Large: Run size for first page, unset for trailing pages. + * - : Unused. + * k : key? + * d : dirty? + * z : zeroed? + * l : large? + * a : allocated? + * + * Following are example bit patterns for the three types of runs. + * + * r : run address + * s : run size + * x : don't care + * - : 0 + * [dzla] : bit set + * + * Unallocated: + * ssssssss ssssssss ssss---- -------- + * xxxxxxxx xxxxxxxx xxxx---- ----d--- + * ssssssss ssssssss ssss---- -----z-- + * + * Small: + * rrrrrrrr rrrrrrrr rrrr---- -------a + * rrrrrrrr rrrrrrrr rrrr---- -------a + * rrrrrrrr rrrrrrrr rrrr---- -------a + * + * Large: + * ssssssss ssssssss ssss---- ------la + * -------- -------- -------- ------la + * -------- -------- -------- ------la + */ + size_t bits; +#define CHUNK_MAP_KEY ((size_t)0x10U) +#define CHUNK_MAP_DIRTY ((size_t)0x08U) +#define CHUNK_MAP_ZEROED ((size_t)0x04U) +#define CHUNK_MAP_LARGE ((size_t)0x02U) +#define CHUNK_MAP_ALLOCATED ((size_t)0x01U) +}; +typedef rb_tree(arena_chunk_map_t) arena_avail_tree_t; +typedef rb_tree(arena_chunk_map_t) arena_run_tree_t; + +/* Arena chunk header. */ +typedef struct arena_chunk_s arena_chunk_t; +struct arena_chunk_s { + /* Arena that owns the chunk. */ + arena_t *arena; + + /* Linkage for the arena's chunks_dirty tree. */ + rb_node(arena_chunk_t) link_dirty; + + /* Number of dirty pages. */ + size_t ndirty; + + /* Map of pages within chunk that keeps track of free/large/small. */ + arena_chunk_map_t map[1]; /* Dynamically sized. */ +}; +typedef rb_tree(arena_chunk_t) arena_chunk_tree_t; + +typedef struct arena_run_s arena_run_t; +struct arena_run_s { +#ifdef MALLOC_DEBUG + uint32_t magic; +# define ARENA_RUN_MAGIC 0x384adf93 +#endif + + /* Bin this run is associated with. */ + arena_bin_t *bin; + + /* Index of first element that might have a free region. */ + unsigned regs_minelm; + + /* Number of free regions in run. */ + unsigned nfree; + + /* Bitmask of in-use regions (0: in use, 1: free). */ + unsigned regs_mask[1]; /* Dynamically sized. */ +}; + +struct arena_bin_s { + /* + * Current run being used to service allocations of this bin's size + * class. + */ + arena_run_t *runcur; + + /* + * Tree of non-full runs. This tree is used when looking for an + * existing run when runcur is no longer usable. We choose the + * non-full run that is lowest in memory; this policy tends to keep + * objects packed well, and it can also help reduce the number of + * almost-empty chunks. + */ + arena_run_tree_t runs; + + /* Size of regions in a run for this bin's size class. */ + size_t reg_size; + + /* Total size of a run for this bin's size class. */ + size_t run_size; + + /* Total number of regions in a run for this bin's size class. */ + uint32_t nregs; + + /* Number of elements in a run's regs_mask for this bin's size class. */ + uint32_t regs_mask_nelms; + + /* Offset of first region in a run for this bin's size class. */ + uint32_t reg0_offset; + +#ifdef MALLOC_STATS + /* Bin statistics. */ + malloc_bin_stats_t stats; +#endif +}; + +struct arena_s { +#ifdef MALLOC_DEBUG + uint32_t magic; +# define ARENA_MAGIC 0x947d3d24 +#endif + + /* All operations on this arena require that lock be locked. */ + pthread_mutex_t lock; + +#ifdef MALLOC_STATS + arena_stats_t stats; +#endif + + /* Tree of dirty-page-containing chunks this arena manages. */ + arena_chunk_tree_t chunks_dirty; + + /* + * In order to avoid rapid chunk allocation/deallocation when an arena + * oscillates right on the cusp of needing a new chunk, cache the most + * recently freed chunk. The spare is left in the arena's chunk trees + * until it is deleted. + * + * There is one spare chunk per arena, rather than one spare total, in + * order to avoid interactions between multiple threads that could make + * a single spare inadequate. + */ + arena_chunk_t *spare; + + /* + * Current count of pages within unused runs that are potentially + * dirty, and for which madvise(... MADV_DONTNEED) has not been called. + * By tracking this, we can institute a limit on how much dirty unused + * memory is mapped for each arena. + */ + size_t ndirty; + + /* + * Size/address-ordered tree of this arena's available runs. This tree + * is used for first-best-fit run allocation. + */ + arena_avail_tree_t runs_avail; + +#ifdef MALLOC_BALANCE + /* + * The arena load balancing machinery needs to keep track of how much + * lock contention there is. This value is exponentially averaged. + */ + uint32_t contention; +#endif + + /* + * bins is used to store rings of free regions of the following sizes, + * assuming a 16-byte quantum, 4kB pagesize, and default MALLOC_OPTIONS. + * + * bins[i] | size | + * --------+------+ + * 0 | 2 | + * 1 | 4 | + * 2 | 8 | + * --------+------+ + * 3 | 16 | + * 4 | 32 | + * 5 | 48 | + * 6 | 64 | + * : : + * : : + * 33 | 496 | + * 34 | 512 | + * --------+------+ + * 35 | 1024 | + * 36 | 2048 | + * --------+------+ + */ + arena_bin_t bins[1]; /* Dynamically sized. */ +}; + +/******************************************************************************/ +/* + * Magazine data structures. + */ + +#ifdef MALLOC_MAG +typedef struct mag_s mag_t; +struct mag_s { + size_t binind; /* Index of associated bin. */ + size_t nrounds; + void *rounds[1]; /* Dynamically sized. */ +}; + +/* + * Magazines are lazily allocated, but once created, they remain until the + * associated mag_rack is destroyed. + */ +typedef struct bin_mags_s bin_mags_t; +struct bin_mags_s { + mag_t *curmag; + mag_t *sparemag; +}; + +typedef struct mag_rack_s mag_rack_t; +struct mag_rack_s { + bin_mags_t bin_mags[1]; /* Dynamically sized. */ +}; +#endif + +/******************************************************************************/ +/* + * Data. + */ + +/* Number of CPUs. */ +static unsigned ncpus; + +/* VM page size. */ +static size_t pagesize; +static size_t pagesize_mask; +static size_t pagesize_2pow; + +/* Various bin-related settings. */ +#ifdef MALLOC_TINY /* Number of (2^n)-spaced tiny bins. */ +# define ntbins ((unsigned)(QUANTUM_2POW - TINY_MIN_2POW)) +#else +# define ntbins 0 +#endif +static unsigned nqbins; /* Number of quantum-spaced bins. */ +static unsigned ncbins; /* Number of cacheline-spaced bins. */ +static unsigned nsbins; /* Number of subpage-spaced bins. */ +static unsigned nbins; +#ifdef MALLOC_TINY +# define tspace_max ((size_t)(QUANTUM >> 1)) +#endif +#define qspace_min QUANTUM +static size_t qspace_max; +static size_t cspace_min; +static size_t cspace_max; +static size_t sspace_min; +static size_t sspace_max; +#define bin_maxclass sspace_max + +static uint8_t const *size2bin; +/* + * const_size2bin is a static constant lookup table that in the common case can + * be used as-is for size2bin. For dynamically linked programs, this avoids + * a page of memory overhead per process. + */ +#define S2B_1(i) i, +#define S2B_2(i) S2B_1(i) S2B_1(i) +#define S2B_4(i) S2B_2(i) S2B_2(i) +#define S2B_8(i) S2B_4(i) S2B_4(i) +#define S2B_16(i) S2B_8(i) S2B_8(i) +#define S2B_32(i) S2B_16(i) S2B_16(i) +#define S2B_64(i) S2B_32(i) S2B_32(i) +#define S2B_128(i) S2B_64(i) S2B_64(i) +#define S2B_256(i) S2B_128(i) S2B_128(i) +static const uint8_t const_size2bin[(1U << PAGESIZE_2POW) - 255] = { + S2B_1(0xffU) /* 0 */ +#if (QUANTUM_2POW == 4) +/* 64-bit system ************************/ +# ifdef MALLOC_TINY + S2B_2(0) /* 2 */ + S2B_2(1) /* 4 */ + S2B_4(2) /* 8 */ + S2B_8(3) /* 16 */ +# define S2B_QMIN 3 +# else + S2B_16(0) /* 16 */ +# define S2B_QMIN 0 +# endif + S2B_16(S2B_QMIN + 1) /* 32 */ + S2B_16(S2B_QMIN + 2) /* 48 */ + S2B_16(S2B_QMIN + 3) /* 64 */ + S2B_16(S2B_QMIN + 4) /* 80 */ + S2B_16(S2B_QMIN + 5) /* 96 */ + S2B_16(S2B_QMIN + 6) /* 112 */ + S2B_16(S2B_QMIN + 7) /* 128 */ +# define S2B_CMIN (S2B_QMIN + 8) +#else +/* 32-bit system ************************/ +# ifdef MALLOC_TINY + S2B_2(0) /* 2 */ + S2B_2(1) /* 4 */ + S2B_4(2) /* 8 */ +# define S2B_QMIN 2 +# else + S2B_8(0) /* 8 */ +# define S2B_QMIN 0 +# endif + S2B_8(S2B_QMIN + 1) /* 16 */ + S2B_8(S2B_QMIN + 2) /* 24 */ + S2B_8(S2B_QMIN + 3) /* 32 */ + S2B_8(S2B_QMIN + 4) /* 40 */ + S2B_8(S2B_QMIN + 5) /* 48 */ + S2B_8(S2B_QMIN + 6) /* 56 */ + S2B_8(S2B_QMIN + 7) /* 64 */ + S2B_8(S2B_QMIN + 8) /* 72 */ + S2B_8(S2B_QMIN + 9) /* 80 */ + S2B_8(S2B_QMIN + 10) /* 88 */ + S2B_8(S2B_QMIN + 11) /* 96 */ + S2B_8(S2B_QMIN + 12) /* 104 */ + S2B_8(S2B_QMIN + 13) /* 112 */ + S2B_8(S2B_QMIN + 14) /* 120 */ + S2B_8(S2B_QMIN + 15) /* 128 */ +# define S2B_CMIN (S2B_QMIN + 16) +#endif +/****************************************/ + S2B_64(S2B_CMIN + 0) /* 192 */ + S2B_64(S2B_CMIN + 1) /* 256 */ + S2B_64(S2B_CMIN + 2) /* 320 */ + S2B_64(S2B_CMIN + 3) /* 384 */ + S2B_64(S2B_CMIN + 4) /* 448 */ + S2B_64(S2B_CMIN + 5) /* 512 */ +# define S2B_SMIN (S2B_CMIN + 6) + S2B_256(S2B_SMIN + 0) /* 768 */ + S2B_256(S2B_SMIN + 1) /* 1024 */ + S2B_256(S2B_SMIN + 2) /* 1280 */ + S2B_256(S2B_SMIN + 3) /* 1536 */ + S2B_256(S2B_SMIN + 4) /* 1792 */ + S2B_256(S2B_SMIN + 5) /* 2048 */ + S2B_256(S2B_SMIN + 6) /* 2304 */ + S2B_256(S2B_SMIN + 7) /* 2560 */ + S2B_256(S2B_SMIN + 8) /* 2816 */ + S2B_256(S2B_SMIN + 9) /* 3072 */ + S2B_256(S2B_SMIN + 10) /* 3328 */ + S2B_256(S2B_SMIN + 11) /* 3584 */ + S2B_256(S2B_SMIN + 12) /* 3840 */ +#if (PAGESIZE_2POW == 13) + S2B_256(S2B_SMIN + 13) /* 4096 */ + S2B_256(S2B_SMIN + 14) /* 4352 */ + S2B_256(S2B_SMIN + 15) /* 4608 */ + S2B_256(S2B_SMIN + 16) /* 4864 */ + S2B_256(S2B_SMIN + 17) /* 5120 */ + S2B_256(S2B_SMIN + 18) /* 5376 */ + S2B_256(S2B_SMIN + 19) /* 5632 */ + S2B_256(S2B_SMIN + 20) /* 5888 */ + S2B_256(S2B_SMIN + 21) /* 6144 */ + S2B_256(S2B_SMIN + 22) /* 6400 */ + S2B_256(S2B_SMIN + 23) /* 6656 */ + S2B_256(S2B_SMIN + 24) /* 6912 */ + S2B_256(S2B_SMIN + 25) /* 7168 */ + S2B_256(S2B_SMIN + 26) /* 7424 */ + S2B_256(S2B_SMIN + 27) /* 7680 */ + S2B_256(S2B_SMIN + 28) /* 7936 */ +#endif +}; +#undef S2B_1 +#undef S2B_2 +#undef S2B_4 +#undef S2B_8 +#undef S2B_16 +#undef S2B_32 +#undef S2B_64 +#undef S2B_128 +#undef S2B_256 +#undef S2B_QMIN +#undef S2B_CMIN +#undef S2B_SMIN + +#ifdef MALLOC_MAG +static size_t max_rounds; +#endif + +/* Various chunk-related settings. */ +static size_t chunksize; +static size_t chunksize_mask; /* (chunksize - 1). */ +static size_t chunk_npages; +static size_t arena_chunk_header_npages; +static size_t arena_maxclass; /* Max size class for arenas. */ + +/********/ +/* + * Chunks. + */ + +/* Protects chunk-related data structures. */ +static malloc_mutex_t huge_mtx; + +/* Tree of chunks that are stand-alone huge allocations. */ +static extent_tree_t huge; + +#ifdef MALLOC_DSS +/* + * Protects sbrk() calls. This avoids malloc races among threads, though it + * does not protect against races with threads that call sbrk() directly. + */ +static malloc_mutex_t dss_mtx; +/* Base address of the DSS. */ +static void *dss_base; +/* Current end of the DSS, or ((void *)-1) if the DSS is exhausted. */ +static void *dss_prev; +/* Current upper limit on DSS addresses. */ +static void *dss_max; + +/* + * Trees of chunks that were previously allocated (trees differ only in node + * ordering). These are used when allocating chunks, in an attempt to re-use + * address space. Depending on function, different tree orderings are needed, + * which is why there are two trees with the same contents. + */ +static extent_tree_t dss_chunks_szad; +static extent_tree_t dss_chunks_ad; +#endif + +#ifdef MALLOC_STATS +/* Huge allocation statistics. */ +static uint64_t huge_nmalloc; +static uint64_t huge_ndalloc; +static size_t huge_allocated; +#endif + +/****************************/ +/* + * base (internal allocation). + */ + +/* + * Current pages that are being used for internal memory allocations. These + * pages are carved up in cacheline-size quanta, so that there is no chance of + * false cache line sharing. + */ +static void *base_pages; +static void *base_next_addr; +static void *base_past_addr; /* Addr immediately past base_pages. */ +static extent_node_t *base_nodes; +static malloc_mutex_t base_mtx; +#ifdef MALLOC_STATS +static size_t base_mapped; +#endif + +/********/ +/* + * Arenas. + */ + +/* + * Arenas that are used to service external requests. Not all elements of the + * arenas array are necessarily used; arenas are created lazily as needed. + */ +static arena_t **arenas; +static unsigned narenas; +#ifndef NO_TLS +# ifdef MALLOC_BALANCE +static unsigned narenas_2pow; +# else +static unsigned next_arena; +# endif +#endif +static pthread_mutex_t arenas_lock; /* Protects arenas initialization. */ + +#ifndef NO_TLS +/* + * Map of pthread_self() --> arenas[???], used for selecting an arena to use + * for allocations. + */ +static __thread arena_t *arenas_map; +#endif + +#ifdef MALLOC_MAG +/* + * Map of thread-specific magazine racks, used for thread-specific object + * caching. + */ +static __thread mag_rack_t *mag_rack; +#endif + +#ifdef MALLOC_STATS +/* Chunk statistics. */ +static chunk_stats_t stats_chunks; +#endif + +/*******************************/ +/* + * Runtime configuration options. + */ +const char *_malloc_options; + +#ifndef MALLOC_PRODUCTION +static bool opt_abort = true; +static bool opt_junk = true; +#else +static bool opt_abort = false; +static bool opt_junk = false; +#endif +#ifdef MALLOC_DSS +static bool opt_dss = true; +static bool opt_mmap = true; +#endif +#ifdef MALLOC_MAG +static bool opt_mag = true; +static size_t opt_mag_size_2pow = MAG_SIZE_2POW_DEFAULT; +#endif +static size_t opt_dirty_max = DIRTY_MAX_DEFAULT; +#ifdef MALLOC_BALANCE +static uint64_t opt_balance_threshold = BALANCE_THRESHOLD_DEFAULT; +#endif +static bool opt_print_stats = false; +static size_t opt_qspace_max_2pow = QSPACE_MAX_2POW_DEFAULT; +static size_t opt_cspace_max_2pow = CSPACE_MAX_2POW_DEFAULT; +static size_t opt_chunk_2pow = CHUNK_2POW_DEFAULT; +static bool opt_utrace = false; +static bool opt_sysv = false; +static bool opt_xmalloc = false; +static bool opt_zero = false; +static int opt_narenas_lshift = 0; + +typedef struct { + void *p; + size_t s; + void *r; +} malloc_utrace_t; + +#ifdef MALLOC_STATS +#define UTRACE(a, b, c) \ + if (opt_utrace) { \ + malloc_utrace_t ut; \ + ut.p = (a); \ + ut.s = (b); \ + ut.r = (c); \ + utrace(&ut, sizeof(ut)); \ + } +#else +#define UTRACE(a, b, c) +#endif + +/******************************************************************************/ +/* + * Begin function prototypes for non-inline static functions. + */ + +static bool malloc_mutex_init(malloc_mutex_t *mutex); +static bool malloc_spin_init(pthread_mutex_t *lock); +static void wrtmessage(const char *p1, const char *p2, const char *p3, + const char *p4); +#ifdef MALLOC_STATS +static void malloc_printf(const char *format, ...); +#endif +static char *umax2s(uintmax_t x, char *s); +#ifdef MALLOC_DSS +static bool base_pages_alloc_dss(size_t minsize); +#endif +static bool base_pages_alloc_mmap(size_t minsize); +static bool base_pages_alloc(size_t minsize); +static void *base_alloc(size_t size); +static void *base_calloc(size_t number, size_t size); +static extent_node_t *base_node_alloc(void); +static void base_node_dealloc(extent_node_t *node); +#ifdef MALLOC_STATS +static void stats_print(arena_t *arena); +#endif +static void *pages_map(void *addr, size_t size); +static void pages_unmap(void *addr, size_t size); +#ifdef MALLOC_DSS +static void *chunk_alloc_dss(size_t size); +static void *chunk_recycle_dss(size_t size, bool zero); +#endif +static void *chunk_alloc_mmap(size_t size); +static void *chunk_alloc(size_t size, bool zero); +#ifdef MALLOC_DSS +static extent_node_t *chunk_dealloc_dss_record(void *chunk, size_t size); +static bool chunk_dealloc_dss(void *chunk, size_t size); +#endif +static void chunk_dealloc_mmap(void *chunk, size_t size); +static void chunk_dealloc(void *chunk, size_t size); +#ifndef NO_TLS +static arena_t *choose_arena_hard(void); +#endif +static void arena_run_split(arena_t *arena, arena_run_t *run, size_t size, + bool large, bool zero); +static arena_chunk_t *arena_chunk_alloc(arena_t *arena); +static void arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk); +static arena_run_t *arena_run_alloc(arena_t *arena, size_t size, bool large, + bool zero); +static void arena_purge(arena_t *arena); +static void arena_run_dalloc(arena_t *arena, arena_run_t *run, bool dirty); +static void arena_run_trim_head(arena_t *arena, arena_chunk_t *chunk, + arena_run_t *run, size_t oldsize, size_t newsize); +static void arena_run_trim_tail(arena_t *arena, arena_chunk_t *chunk, + arena_run_t *run, size_t oldsize, size_t newsize, bool dirty); +static arena_run_t *arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin); +static void *arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin); +static size_t arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size); +#ifdef MALLOC_BALANCE +static void arena_lock_balance_hard(arena_t *arena); +#endif +#ifdef MALLOC_MAG +static void mag_load(mag_t *mag); +#endif +static void *arena_malloc_large(arena_t *arena, size_t size, bool zero); +static void *arena_palloc(arena_t *arena, size_t alignment, size_t size, + size_t alloc_size); +static size_t arena_salloc(const void *ptr); +#ifdef MALLOC_MAG +static void mag_unload(mag_t *mag); +#endif +static void arena_dalloc_large(arena_t *arena, arena_chunk_t *chunk, + void *ptr); +static void arena_ralloc_large_shrink(arena_t *arena, arena_chunk_t *chunk, + void *ptr, size_t size, size_t oldsize); +static bool arena_ralloc_large_grow(arena_t *arena, arena_chunk_t *chunk, + void *ptr, size_t size, size_t oldsize); +static bool arena_ralloc_large(void *ptr, size_t size, size_t oldsize); +static void *arena_ralloc(void *ptr, size_t size, size_t oldsize); +static bool arena_new(arena_t *arena); +static arena_t *arenas_extend(unsigned ind); +#ifdef MALLOC_MAG +static mag_t *mag_create(arena_t *arena, size_t binind); +static void mag_destroy(mag_t *mag); +static mag_rack_t *mag_rack_create(arena_t *arena); +static void mag_rack_destroy(mag_rack_t *rack); +#endif +static void *huge_malloc(size_t size, bool zero); +static void *huge_palloc(size_t alignment, size_t size); +static void *huge_ralloc(void *ptr, size_t size, size_t oldsize); +static void huge_dalloc(void *ptr); +static void malloc_print_stats(void); +#ifdef MALLOC_DEBUG +static void size2bin_validate(void); +#endif +static bool size2bin_init(void); +static bool size2bin_init_hard(void); +static unsigned malloc_ncpus(void); +static bool malloc_init_hard(void); +void _malloc_prefork(void); +void _malloc_postfork(void); + +/* + * End function prototypes. + */ +/******************************************************************************/ + +static void +wrtmessage(const char *p1, const char *p2, const char *p3, const char *p4) +{ + + write(STDERR_FILENO, p1, strlen(p1)); + write(STDERR_FILENO, p2, strlen(p2)); + write(STDERR_FILENO, p3, strlen(p3)); + write(STDERR_FILENO, p4, strlen(p4)); +} + +#define _malloc_message malloc_message +void (*_malloc_message)(const char *p1, const char *p2, const char *p3, + const char *p4) = wrtmessage; + +/* + * We don't want to depend on vsnprintf() for production builds, since that can + * cause unnecessary bloat for static binaries. umax2s() provides minimal + * integer printing functionality, so that malloc_printf() use can be limited to + * MALLOC_STATS code. + */ +#define UMAX2S_BUFSIZE 21 +static char * +umax2s(uintmax_t x, char *s) +{ + unsigned i; + + i = UMAX2S_BUFSIZE - 1; + s[i] = '\0'; + do { + i--; + s[i] = "0123456789"[x % 10]; + x /= 10; + } while (x > 0); + + return (&s[i]); +} + +/* + * Define a custom assert() in order to reduce the chances of deadlock during + * assertion failure. + */ +#ifdef MALLOC_DEBUG +# define assert(e) do { \ + if (!(e)) { \ + char line_buf[UMAX2S_BUFSIZE]; \ + _malloc_message(__FILE__, ":", umax2s(__LINE__, \ + line_buf), ": Failed assertion: "); \ + _malloc_message("\"", #e, "\"\n", ""); \ + abort(); \ + } \ +} while (0) +#else +#define assert(e) +#endif + +#ifdef MALLOC_STATS +static int +utrace(const void *addr, size_t len) +{ + malloc_utrace_t *ut = (malloc_utrace_t *)addr; + + assert(len == sizeof(malloc_utrace_t)); + + if (ut->p == NULL && ut->s == 0 && ut->r == NULL) + malloc_printf("%d x USER malloc_init()\n", getpid()); + else if (ut->p == NULL && ut->r != NULL) { + malloc_printf("%d x USER %p = malloc(%zu)\n", getpid(), ut->r, + ut->s); + } else if (ut->p != NULL && ut->r != NULL) { + malloc_printf("%d x USER %p = realloc(%p, %zu)\n", getpid(), + ut->r, ut->p, ut->s); + } else + malloc_printf("%d x USER free(%p)\n", getpid(), ut->p); + + return (0); +} +#endif + +static inline const char * +_getprogname(void) +{ + + return (""); +} + +#ifdef MALLOC_STATS +/* + * Print to stderr in such a way as to (hopefully) avoid memory allocation. + */ +static void +malloc_printf(const char *format, ...) +{ + char buf[4096]; + va_list ap; + + va_start(ap, format); + vsnprintf(buf, sizeof(buf), format, ap); + va_end(ap); + _malloc_message(buf, "", "", ""); +} +#endif + +/******************************************************************************/ +/* + * Begin mutex. + */ + +static bool +malloc_mutex_init(malloc_mutex_t *mutex) +{ + pthread_mutexattr_t attr; + + if (pthread_mutexattr_init(&attr) != 0) + return (true); + pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ADAPTIVE_NP); + if (pthread_mutex_init(mutex, &attr) != 0) { + pthread_mutexattr_destroy(&attr); + return (true); + } + pthread_mutexattr_destroy(&attr); + + return (false); +} + +static inline void +malloc_mutex_lock(malloc_mutex_t *mutex) +{ + + if (__isthreaded) + pthread_mutex_lock(mutex); +} + +static inline void +malloc_mutex_unlock(malloc_mutex_t *mutex) +{ + + if (__isthreaded) + pthread_mutex_unlock(mutex); +} + +/* + * End mutex. + */ +/******************************************************************************/ +/* + * Begin spin lock. Spin locks here are actually adaptive mutexes that block + * after a period of spinning, because unbounded spinning would allow for + * priority inversion. + */ + +static bool +malloc_spin_init(pthread_mutex_t *lock) +{ + + if (pthread_mutex_init(lock, NULL) != 0) + return (true); + + return (false); +} + +static inline unsigned +malloc_spin_lock(pthread_mutex_t *lock) +{ + unsigned ret = 0; + + if (__isthreaded) { + if (pthread_mutex_trylock(lock) != 0) { + unsigned i; + volatile unsigned j; + + /* Exponentially back off. */ + for (i = 1; i <= SPIN_LIMIT_2POW; i++) { + for (j = 0; j < (1U << i); j++) { + ret++; + CPU_SPINWAIT; + } + + if (pthread_mutex_trylock(lock) == 0) + return (ret); + } + + /* + * Spinning failed. Block until the lock becomes + * available, in order to avoid indefinite priority + * inversion. + */ + pthread_mutex_lock(lock); + assert((ret << BLOCK_COST_2POW) != 0); + return (ret << BLOCK_COST_2POW); + } + } + + return (ret); +} + +static inline void +malloc_spin_unlock(pthread_mutex_t *lock) +{ + + if (__isthreaded) + pthread_mutex_unlock(lock); +} + +/* + * End spin lock. + */ +/******************************************************************************/ +/* + * Begin Utility functions/macros. + */ + +/* Return the chunk address for allocation address a. */ +#define CHUNK_ADDR2BASE(a) \ + ((void *)((uintptr_t)(a) & ~chunksize_mask)) + +/* Return the chunk offset of address a. */ +#define CHUNK_ADDR2OFFSET(a) \ + ((size_t)((uintptr_t)(a) & chunksize_mask)) + +/* Return the smallest chunk multiple that is >= s. */ +#define CHUNK_CEILING(s) \ + (((s) + chunksize_mask) & ~chunksize_mask) + +/* Return the smallest quantum multiple that is >= a. */ +#define QUANTUM_CEILING(a) \ + (((a) + QUANTUM_MASK) & ~QUANTUM_MASK) + +/* Return the smallest cacheline multiple that is >= s. */ +#define CACHELINE_CEILING(s) \ + (((s) + CACHELINE_MASK) & ~CACHELINE_MASK) + +/* Return the smallest subpage multiple that is >= s. */ +#define SUBPAGE_CEILING(s) \ + (((s) + SUBPAGE_MASK) & ~SUBPAGE_MASK) + +/* Return the smallest pagesize multiple that is >= s. */ +#define PAGE_CEILING(s) \ + (((s) + pagesize_mask) & ~pagesize_mask) + +#ifdef MALLOC_TINY +/* Compute the smallest power of 2 that is >= x. */ +static inline size_t +pow2_ceil(size_t x) +{ + + x--; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; +#if (SIZEOF_PTR == 8) + x |= x >> 32; +#endif + x++; + return (x); +} +#endif + +#ifdef MALLOC_BALANCE +/* + * Use a simple linear congruential pseudo-random number generator: + * + * prn(y) = (a*x + c) % m + * + * where the following constants ensure maximal period: + * + * a == Odd number (relatively prime to 2^n), and (a-1) is a multiple of 4. + * c == Odd number (relatively prime to 2^n). + * m == 2^32 + * + * See Knuth's TAOCP 3rd Ed., Vol. 2, pg. 17 for details on these constraints. + * + * This choice of m has the disadvantage that the quality of the bits is + * proportional to bit position. For example. the lowest bit has a cycle of 2, + * the next has a cycle of 4, etc. For this reason, we prefer to use the upper + * bits. + */ +# define PRN_DEFINE(suffix, var, a, c) \ +static inline void \ +sprn_##suffix(uint32_t seed) \ +{ \ + var = seed; \ +} \ + \ +static inline uint32_t \ +prn_##suffix(uint32_t lg_range) \ +{ \ + uint32_t ret, x; \ + \ + assert(lg_range > 0); \ + assert(lg_range <= 32); \ + \ + x = (var * (a)) + (c); \ + var = x; \ + ret = x >> (32 - lg_range); \ + \ + return (ret); \ +} +# define SPRN(suffix, seed) sprn_##suffix(seed) +# define PRN(suffix, lg_range) prn_##suffix(lg_range) +#endif + +#ifdef MALLOC_BALANCE +/* Define the PRNG used for arena assignment. */ +static __thread uint32_t balance_x; +PRN_DEFINE(balance, balance_x, 1297, 1301) +#endif + +/******************************************************************************/ + +#ifdef MALLOC_DSS +static bool +base_pages_alloc_dss(size_t minsize) +{ + + /* + * Do special DSS allocation here, since base allocations don't need to + * be chunk-aligned. + */ + malloc_mutex_lock(&dss_mtx); + if (dss_prev != (void *)-1) { + intptr_t incr; + size_t csize = CHUNK_CEILING(minsize); + + do { + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Calculate how much padding is necessary to + * chunk-align the end of the DSS. Don't worry about + * dss_max not being chunk-aligned though. + */ + incr = (intptr_t)chunksize + - (intptr_t)CHUNK_ADDR2OFFSET(dss_max); + assert(incr >= 0); + if ((size_t)incr < minsize) + incr += csize; + + dss_prev = sbrk(incr); + if (dss_prev == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev + incr); + base_pages = dss_prev; + base_next_addr = base_pages; + base_past_addr = dss_max; +#ifdef MALLOC_STATS + base_mapped += incr; +#endif + malloc_mutex_unlock(&dss_mtx); + return (false); + } + } while (dss_prev != (void *)-1); + } + malloc_mutex_unlock(&dss_mtx); + + return (true); +} +#endif + +static bool +base_pages_alloc_mmap(size_t minsize) +{ + size_t csize; + + assert(minsize != 0); + csize = PAGE_CEILING(minsize); + base_pages = pages_map(NULL, csize); + if (base_pages == NULL) + return (true); + base_next_addr = base_pages; + base_past_addr = (void *)((uintptr_t)base_pages + csize); +#ifdef MALLOC_STATS + base_mapped += csize; +#endif + + return (false); +} + +static bool +base_pages_alloc(size_t minsize) +{ + +#ifdef MALLOC_DSS + if (opt_dss) { + if (base_pages_alloc_dss(minsize) == false) + return (false); + } + + if (opt_mmap && minsize != 0) +#endif + { + if (base_pages_alloc_mmap(minsize) == false) + return (false); + } + + return (true); +} + +static void * +base_alloc(size_t size) +{ + void *ret; + size_t csize; + + /* Round size up to nearest multiple of the cacheline size. */ + csize = CACHELINE_CEILING(size); + + malloc_mutex_lock(&base_mtx); + /* Make sure there's enough space for the allocation. */ + if ((uintptr_t)base_next_addr + csize > (uintptr_t)base_past_addr) { + if (base_pages_alloc(csize)) { + malloc_mutex_unlock(&base_mtx); + return (NULL); + } + } + /* Allocate. */ + ret = base_next_addr; + base_next_addr = (void *)((uintptr_t)base_next_addr + csize); + malloc_mutex_unlock(&base_mtx); + + return (ret); +} + +static void * +base_calloc(size_t number, size_t size) +{ + void *ret; + + ret = base_alloc(number * size); + memset(ret, 0, number * size); + + return (ret); +} + +static extent_node_t * +base_node_alloc(void) +{ + extent_node_t *ret; + + malloc_mutex_lock(&base_mtx); + if (base_nodes != NULL) { + ret = base_nodes; + base_nodes = *(extent_node_t **)ret; + malloc_mutex_unlock(&base_mtx); + } else { + malloc_mutex_unlock(&base_mtx); + ret = (extent_node_t *)base_alloc(sizeof(extent_node_t)); + } + + return (ret); +} + +static void +base_node_dealloc(extent_node_t *node) +{ + + malloc_mutex_lock(&base_mtx); + *(extent_node_t **)node = base_nodes; + base_nodes = node; + malloc_mutex_unlock(&base_mtx); +} + +/******************************************************************************/ + +#ifdef MALLOC_STATS +static void +stats_print(arena_t *arena) +{ + unsigned i, gap_start; + + malloc_printf("dirty: %zu page%s dirty, %llu sweep%s," + " %llu madvise%s, %llu page%s purged\n", + arena->ndirty, arena->ndirty == 1 ? "" : "s", + arena->stats.npurge, arena->stats.npurge == 1 ? "" : "s", + arena->stats.nmadvise, arena->stats.nmadvise == 1 ? "" : "s", + arena->stats.purged, arena->stats.purged == 1 ? "" : "s"); + + malloc_printf(" allocated nmalloc ndalloc\n"); + malloc_printf("small: %12zu %12llu %12llu\n", + arena->stats.allocated_small, arena->stats.nmalloc_small, + arena->stats.ndalloc_small); + malloc_printf("large: %12zu %12llu %12llu\n", + arena->stats.allocated_large, arena->stats.nmalloc_large, + arena->stats.ndalloc_large); + malloc_printf("total: %12zu %12llu %12llu\n", + arena->stats.allocated_small + arena->stats.allocated_large, + arena->stats.nmalloc_small + arena->stats.nmalloc_large, + arena->stats.ndalloc_small + arena->stats.ndalloc_large); + malloc_printf("mapped: %12zu\n", arena->stats.mapped); + +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + malloc_printf("bins: bin size regs pgs mags " + "newruns reruns maxruns curruns\n"); + } else { +#endif + malloc_printf("bins: bin size regs pgs requests " + "newruns reruns maxruns curruns\n"); +#ifdef MALLOC_MAG + } +#endif + for (i = 0, gap_start = UINT_MAX; i < nbins; i++) { + if (arena->bins[i].stats.nruns == 0) { + if (gap_start == UINT_MAX) + gap_start = i; + } else { + if (gap_start != UINT_MAX) { + if (i > gap_start + 1) { + /* Gap of more than one size class. */ + malloc_printf("[%u..%u]\n", + gap_start, i - 1); + } else { + /* Gap of one size class. */ + malloc_printf("[%u]\n", gap_start); + } + gap_start = UINT_MAX; + } + malloc_printf( + "%13u %1s %4u %4u %3u %9llu %9llu" + " %9llu %7lu %7lu\n", + i, + i < ntbins ? "T" : i < ntbins + nqbins ? "Q" : + i < ntbins + nqbins + ncbins ? "C" : "S", + arena->bins[i].reg_size, + arena->bins[i].nregs, + arena->bins[i].run_size >> pagesize_2pow, +#ifdef MALLOC_MAG + (__isthreaded && opt_mag) ? + arena->bins[i].stats.nmags : +#endif + arena->bins[i].stats.nrequests, + arena->bins[i].stats.nruns, + arena->bins[i].stats.reruns, + arena->bins[i].stats.highruns, + arena->bins[i].stats.curruns); + } + } + if (gap_start != UINT_MAX) { + if (i > gap_start + 1) { + /* Gap of more than one size class. */ + malloc_printf("[%u..%u]\n", gap_start, i - 1); + } else { + /* Gap of one size class. */ + malloc_printf("[%u]\n", gap_start); + } + } +} +#endif + +/* + * End Utility functions/macros. + */ +/******************************************************************************/ +/* + * Begin extent tree code. + */ + +#ifdef MALLOC_DSS +static inline int +extent_szad_comp(extent_node_t *a, extent_node_t *b) +{ + int ret; + size_t a_size = a->size; + size_t b_size = b->size; + + ret = (a_size > b_size) - (a_size < b_size); + if (ret == 0) { + uintptr_t a_addr = (uintptr_t)a->addr; + uintptr_t b_addr = (uintptr_t)b->addr; + + ret = (a_addr > b_addr) - (a_addr < b_addr); + } + + return (ret); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(static, extent_tree_szad_, extent_tree_t, extent_node_t, + link_szad, extent_szad_comp) +#endif + +static inline int +extent_ad_comp(extent_node_t *a, extent_node_t *b) +{ + uintptr_t a_addr = (uintptr_t)a->addr; + uintptr_t b_addr = (uintptr_t)b->addr; + + return ((a_addr > b_addr) - (a_addr < b_addr)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(static, extent_tree_ad_, extent_tree_t, extent_node_t, link_ad, + extent_ad_comp) + +/* + * End extent tree code. + */ +/******************************************************************************/ +/* + * Begin chunk management functions. + */ + +static void * +pages_map(void *addr, size_t size) +{ + void *ret; + + /* + * We don't use MAP_FIXED here, because it can cause the *replacement* + * of existing mappings, and we only want to create new mappings. + */ + ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, + -1, 0); + assert(ret != NULL); + + if (ret == MAP_FAILED) + ret = NULL; + else if (addr != NULL && ret != addr) { + /* + * We succeeded in mapping memory, but not in the right place. + */ + if (munmap(ret, size) == -1) { + char buf[STRERROR_BUF]; + + strerror_r(errno, buf, sizeof(buf)); + _malloc_message(_getprogname(), + ": (malloc) Error in munmap(): ", buf, "\n"); + if (opt_abort) + abort(); + } + ret = NULL; + } + + assert(ret == NULL || (addr == NULL && ret != addr) + || (addr != NULL && ret == addr)); + return (ret); +} + +static void +pages_unmap(void *addr, size_t size) +{ + + if (munmap(addr, size) == -1) { + char buf[STRERROR_BUF]; + + strerror_r(errno, buf, sizeof(buf)); + _malloc_message(_getprogname(), + ": (malloc) Error in munmap(): ", buf, "\n"); + if (opt_abort) + abort(); + } +} + +#ifdef MALLOC_DSS +static void * +chunk_alloc_dss(size_t size) +{ + + /* + * sbrk() uses a signed increment argument, so take care not to + * interpret a huge allocation request as a negative increment. + */ + if ((intptr_t)size < 0) + return (NULL); + + malloc_mutex_lock(&dss_mtx); + if (dss_prev != (void *)-1) { + intptr_t incr; + + /* + * The loop is necessary to recover from races with other + * threads that are using the DSS for something other than + * malloc. + */ + do { + void *ret; + + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Calculate how much padding is necessary to + * chunk-align the end of the DSS. + */ + incr = (intptr_t)size + - (intptr_t)CHUNK_ADDR2OFFSET(dss_max); + if (incr == (intptr_t)size) + ret = dss_max; + else { + ret = (void *)((intptr_t)dss_max + incr); + incr += size; + } + + dss_prev = sbrk(incr); + if (dss_prev == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev + incr); + malloc_mutex_unlock(&dss_mtx); + return (ret); + } + } while (dss_prev != (void *)-1); + } + malloc_mutex_unlock(&dss_mtx); + + return (NULL); +} + +static void * +chunk_recycle_dss(size_t size, bool zero) +{ + extent_node_t *node, key; + + key.addr = NULL; + key.size = size; + malloc_mutex_lock(&dss_mtx); + node = extent_tree_szad_nsearch(&dss_chunks_szad, &key); + if (node != NULL) { + void *ret = node->addr; + + /* Remove node from the tree. */ + extent_tree_szad_remove(&dss_chunks_szad, node); + if (node->size == size) { + extent_tree_ad_remove(&dss_chunks_ad, node); + base_node_dealloc(node); + } else { + /* + * Insert the remainder of node's address range as a + * smaller chunk. Its position within dss_chunks_ad + * does not change. + */ + assert(node->size > size); + node->addr = (void *)((uintptr_t)node->addr + size); + node->size -= size; + extent_tree_szad_insert(&dss_chunks_szad, node); + } + malloc_mutex_unlock(&dss_mtx); + + if (zero) + memset(ret, 0, size); + return (ret); + } + malloc_mutex_unlock(&dss_mtx); + + return (NULL); +} +#endif + +static void * +chunk_alloc_mmap(size_t size) +{ + void *ret; + size_t offset; + + /* + * Ideally, there would be a way to specify alignment to mmap() (like + * NetBSD has), but in the absence of such a feature, we have to work + * hard to efficiently create aligned mappings. The reliable, but + * expensive method is to create a mapping that is over-sized, then + * trim the excess. However, that always results in at least one call + * to pages_unmap(). + * + * A more optimistic approach is to try mapping precisely the right + * amount, then try to append another mapping if alignment is off. In + * practice, this works out well as long as the application is not + * interleaving mappings via direct mmap() calls. If we do run into a + * situation where there is an interleaved mapping and we are unable to + * extend an unaligned mapping, our best option is to momentarily + * revert to the reliable-but-expensive method. This will tend to + * leave a gap in the memory map that is too small to cause later + * problems for the optimistic method. + */ + + ret = pages_map(NULL, size); + if (ret == NULL) + return (NULL); + + offset = CHUNK_ADDR2OFFSET(ret); + if (offset != 0) { + /* Try to extend chunk boundary. */ + if (pages_map((void *)((uintptr_t)ret + size), + chunksize - offset) == NULL) { + /* + * Extension failed. Clean up, then revert to the + * reliable-but-expensive method. + */ + pages_unmap(ret, size); + + /* Beware size_t wrap-around. */ + if (size + chunksize <= size) + return NULL; + + ret = pages_map(NULL, size + chunksize); + if (ret == NULL) + return (NULL); + + /* Clean up unneeded leading/trailing space. */ + offset = CHUNK_ADDR2OFFSET(ret); + if (offset != 0) { + /* Leading space. */ + pages_unmap(ret, chunksize - offset); + + ret = (void *)((uintptr_t)ret + + (chunksize - offset)); + + /* Trailing space. */ + pages_unmap((void *)((uintptr_t)ret + size), + offset); + } else { + /* Trailing space only. */ + pages_unmap((void *)((uintptr_t)ret + size), + chunksize); + } + } else { + /* Clean up unneeded leading space. */ + pages_unmap(ret, chunksize - offset); + ret = (void *)((uintptr_t)ret + (chunksize - offset)); + } + } + + return (ret); +} + +static void * +chunk_alloc(size_t size, bool zero) +{ + void *ret; + + assert(size != 0); + assert((size & chunksize_mask) == 0); + +#ifdef MALLOC_DSS + if (opt_dss) { + ret = chunk_recycle_dss(size, zero); + if (ret != NULL) { + goto RETURN; + } + + ret = chunk_alloc_dss(size); + if (ret != NULL) + goto RETURN; + } + + if (opt_mmap) +#endif + { + ret = chunk_alloc_mmap(size); + if (ret != NULL) + goto RETURN; + } + + /* All strategies for allocation failed. */ + ret = NULL; +RETURN: +#ifdef MALLOC_STATS + if (ret != NULL) { + stats_chunks.nchunks += (size / chunksize); + stats_chunks.curchunks += (size / chunksize); + } + if (stats_chunks.curchunks > stats_chunks.highchunks) + stats_chunks.highchunks = stats_chunks.curchunks; +#endif + + assert(CHUNK_ADDR2BASE(ret) == ret); + return (ret); +} + +#ifdef MALLOC_DSS +static extent_node_t * +chunk_dealloc_dss_record(void *chunk, size_t size) +{ + extent_node_t *node, *prev, key; + + key.addr = (void *)((uintptr_t)chunk + size); + node = extent_tree_ad_nsearch(&dss_chunks_ad, &key); + /* Try to coalesce forward. */ + if (node != NULL && node->addr == key.addr) { + /* + * Coalesce chunk with the following address range. This does + * not change the position within dss_chunks_ad, so only + * remove/insert from/into dss_chunks_szad. + */ + extent_tree_szad_remove(&dss_chunks_szad, node); + node->addr = chunk; + node->size += size; + extent_tree_szad_insert(&dss_chunks_szad, node); + } else { + /* + * Coalescing forward failed, so insert a new node. Drop + * dss_mtx during node allocation, since it is possible that a + * new base chunk will be allocated. + */ + malloc_mutex_unlock(&dss_mtx); + node = base_node_alloc(); + malloc_mutex_lock(&dss_mtx); + if (node == NULL) + return (NULL); + node->addr = chunk; + node->size = size; + extent_tree_ad_insert(&dss_chunks_ad, node); + extent_tree_szad_insert(&dss_chunks_szad, node); + } + + /* Try to coalesce backward. */ + prev = extent_tree_ad_prev(&dss_chunks_ad, node); + if (prev != NULL && (void *)((uintptr_t)prev->addr + prev->size) == + chunk) { + /* + * Coalesce chunk with the previous address range. This does + * not change the position within dss_chunks_ad, so only + * remove/insert node from/into dss_chunks_szad. + */ + extent_tree_szad_remove(&dss_chunks_szad, prev); + extent_tree_ad_remove(&dss_chunks_ad, prev); + + extent_tree_szad_remove(&dss_chunks_szad, node); + node->addr = prev->addr; + node->size += prev->size; + extent_tree_szad_insert(&dss_chunks_szad, node); + + base_node_dealloc(prev); + } + + return (node); +} + +static bool +chunk_dealloc_dss(void *chunk, size_t size) +{ + + malloc_mutex_lock(&dss_mtx); + if ((uintptr_t)chunk >= (uintptr_t)dss_base + && (uintptr_t)chunk < (uintptr_t)dss_max) { + extent_node_t *node; + + /* Try to coalesce with other unused chunks. */ + node = chunk_dealloc_dss_record(chunk, size); + if (node != NULL) { + chunk = node->addr; + size = node->size; + } + + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Try to shrink the DSS if this chunk is at the end of the + * DSS. The sbrk() call here is subject to a race condition + * with threads that use brk(2) or sbrk(2) directly, but the + * alternative would be to leak memory for the sake of poorly + * designed multi-threaded programs. + */ + if ((void *)((uintptr_t)chunk + size) == dss_max + && (dss_prev = sbrk(-(intptr_t)size)) == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev - (intptr_t)size); + + if (node != NULL) { + extent_tree_szad_remove(&dss_chunks_szad, node); + extent_tree_ad_remove(&dss_chunks_ad, node); + base_node_dealloc(node); + } + malloc_mutex_unlock(&dss_mtx); + } else { + malloc_mutex_unlock(&dss_mtx); + madvise(chunk, size, MADV_DONTNEED); + } + + return (false); + } + malloc_mutex_unlock(&dss_mtx); + + return (true); +} +#endif + +static void +chunk_dealloc_mmap(void *chunk, size_t size) +{ + + pages_unmap(chunk, size); +} + +static void +chunk_dealloc(void *chunk, size_t size) +{ + + assert(chunk != NULL); + assert(CHUNK_ADDR2BASE(chunk) == chunk); + assert(size != 0); + assert((size & chunksize_mask) == 0); + +#ifdef MALLOC_STATS + stats_chunks.curchunks -= (size / chunksize); +#endif + +#ifdef MALLOC_DSS + if (opt_dss) { + if (chunk_dealloc_dss(chunk, size) == false) + return; + } + + if (opt_mmap) +#endif + chunk_dealloc_mmap(chunk, size); +} + +/* + * End chunk management functions. + */ +/******************************************************************************/ +/* + * Begin arena. + */ + +/* + * Choose an arena based on a per-thread value (fast-path code, calls slow-path + * code if necessary). + */ +static inline arena_t * +choose_arena(void) +{ + arena_t *ret; + + /* + * We can only use TLS if this is a PIC library, since for the static + * library version, libc's malloc is used by TLS allocation, which + * introduces a bootstrapping issue. + */ +#ifndef NO_TLS + if (__isthreaded == false) { + /* Avoid the overhead of TLS for single-threaded operation. */ + return (arenas[0]); + } + + ret = arenas_map; + if (ret == NULL) { + ret = choose_arena_hard(); + assert(ret != NULL); + } +#else + if (__isthreaded && narenas > 1) { + unsigned long ind; + + /* + * Hash pthread_self() to one of the arenas. There is a prime + * number of arenas, so this has a reasonable chance of + * working. Even so, the hashing can be easily thwarted by + * inconvenient pthread_self() values. Without specific + * knowledge of how pthread_self() calculates values, we can't + * easily do much better than this. + */ + ind = (unsigned long) pthread_self() % narenas; + + /* + * Optimistially assume that arenas[ind] has been initialized. + * At worst, we find out that some other thread has already + * done so, after acquiring the lock in preparation. Note that + * this lazy locking also has the effect of lazily forcing + * cache coherency; without the lock acquisition, there's no + * guarantee that modification of arenas[ind] by another thread + * would be seen on this CPU for an arbitrary amount of time. + * + * In general, this approach to modifying a synchronized value + * isn't a good idea, but in this case we only ever modify the + * value once, so things work out well. + */ + ret = arenas[ind]; + if (ret == NULL) { + /* + * Avoid races with another thread that may have already + * initialized arenas[ind]. + */ + malloc_spin_lock(&arenas_lock); + if (arenas[ind] == NULL) + ret = arenas_extend((unsigned)ind); + else + ret = arenas[ind]; + malloc_spin_unlock(&arenas_lock); + } + } else + ret = arenas[0]; +#endif + + assert(ret != NULL); + return (ret); +} + +#ifndef NO_TLS +/* + * Choose an arena based on a per-thread value (slow-path code only, called + * only by choose_arena()). + */ +static arena_t * +choose_arena_hard(void) +{ + arena_t *ret; + + assert(__isthreaded); + +#ifdef MALLOC_BALANCE + /* Seed the PRNG used for arena load balancing. */ + SPRN(balance, (uint32_t)(uintptr_t)(pthread_self())); +#endif + + if (narenas > 1) { +#ifdef MALLOC_BALANCE + unsigned ind; + + ind = PRN(balance, narenas_2pow); + if ((ret = arenas[ind]) == NULL) { + malloc_spin_lock(&arenas_lock); + if ((ret = arenas[ind]) == NULL) + ret = arenas_extend(ind); + malloc_spin_unlock(&arenas_lock); + } +#else + malloc_spin_lock(&arenas_lock); + if ((ret = arenas[next_arena]) == NULL) + ret = arenas_extend(next_arena); + next_arena = (next_arena + 1) % narenas; + malloc_spin_unlock(&arenas_lock); +#endif + } else + ret = arenas[0]; + + arenas_map = ret; + + return (ret); +} +#endif + +static inline int +arena_chunk_comp(arena_chunk_t *a, arena_chunk_t *b) +{ + uintptr_t a_chunk = (uintptr_t)a; + uintptr_t b_chunk = (uintptr_t)b; + + assert(a != NULL); + assert(b != NULL); + + return ((a_chunk > b_chunk) - (a_chunk < b_chunk)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(static, arena_chunk_tree_dirty_, arena_chunk_tree_t, + arena_chunk_t, link_dirty, arena_chunk_comp) + +static inline int +arena_run_comp(arena_chunk_map_t *a, arena_chunk_map_t *b) +{ + uintptr_t a_mapelm = (uintptr_t)a; + uintptr_t b_mapelm = (uintptr_t)b; + + assert(a != NULL); + assert(b != NULL); + + return ((a_mapelm > b_mapelm) - (a_mapelm < b_mapelm)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(static, arena_run_tree_, arena_run_tree_t, arena_chunk_map_t, + link, arena_run_comp) + +static inline int +arena_avail_comp(arena_chunk_map_t *a, arena_chunk_map_t *b) +{ + int ret; + size_t a_size = a->bits & ~pagesize_mask; + size_t b_size = b->bits & ~pagesize_mask; + + ret = (a_size > b_size) - (a_size < b_size); + if (ret == 0) { + uintptr_t a_mapelm, b_mapelm; + + if ((a->bits & CHUNK_MAP_KEY) == 0) + a_mapelm = (uintptr_t)a; + else { + /* + * Treat keys as though they are lower than anything + * else. + */ + a_mapelm = 0; + } + b_mapelm = (uintptr_t)b; + + ret = (a_mapelm > b_mapelm) - (a_mapelm < b_mapelm); + } + + return (ret); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(static, arena_avail_tree_, arena_avail_tree_t, + arena_chunk_map_t, link, arena_avail_comp) + +static inline void * +arena_run_reg_alloc(arena_run_t *run, arena_bin_t *bin) +{ + void *ret; + unsigned i, mask, bit, regind; + + assert(run->magic == ARENA_RUN_MAGIC); + assert(run->regs_minelm < bin->regs_mask_nelms); + + /* + * Move the first check outside the loop, so that run->regs_minelm can + * be updated unconditionally, without the possibility of updating it + * multiple times. + */ + i = run->regs_minelm; + mask = run->regs_mask[i]; + if (mask != 0) { + /* Usable allocation found. */ + bit = ffs((int)mask) - 1; + + regind = ((i << (SIZEOF_INT_2POW + 3)) + bit); + assert(regind < bin->nregs); + ret = (void *)(((uintptr_t)run) + bin->reg0_offset + + (bin->reg_size * regind)); + + /* Clear bit. */ + mask ^= (1U << bit); + run->regs_mask[i] = mask; + + return (ret); + } + + for (i++; i < bin->regs_mask_nelms; i++) { + mask = run->regs_mask[i]; + if (mask != 0) { + /* Usable allocation found. */ + bit = ffs((int)mask) - 1; + + regind = ((i << (SIZEOF_INT_2POW + 3)) + bit); + assert(regind < bin->nregs); + ret = (void *)(((uintptr_t)run) + bin->reg0_offset + + (bin->reg_size * regind)); + + /* Clear bit. */ + mask ^= (1U << bit); + run->regs_mask[i] = mask; + + /* + * Make a note that nothing before this element + * contains a free region. + */ + run->regs_minelm = i; /* Low payoff: + (mask == 0); */ + + return (ret); + } + } + /* Not reached. */ + assert(0); + return (NULL); +} + +static inline void +arena_run_reg_dalloc(arena_run_t *run, arena_bin_t *bin, void *ptr, size_t size) +{ + unsigned diff, regind, elm, bit; + + assert(run->magic == ARENA_RUN_MAGIC); + + /* + * Avoid doing division with a variable divisor if possible. Using + * actual division here can reduce allocator throughput by over 20%! + */ + diff = (unsigned)((uintptr_t)ptr - (uintptr_t)run - bin->reg0_offset); + if ((size & (size - 1)) == 0) { + /* + * log2_table allows fast division of a power of two in the + * [1..128] range. + * + * (x / divisor) becomes (x >> log2_table[divisor - 1]). + */ + static const unsigned char log2_table[] = { + 0, 1, 0, 2, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 4, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7 + }; + + if (size <= 128) + regind = (diff >> log2_table[size - 1]); + else if (size <= 32768) + regind = diff >> (8 + log2_table[(size >> 8) - 1]); + else + regind = diff / size; + } else if (size < qspace_max) { + /* + * To divide by a number D that is not a power of two we + * multiply by (2^21 / D) and then right shift by 21 positions. + * + * X / D + * + * becomes + * + * (X * qsize_invs[(D >> QUANTUM_2POW) - 3]) + * >> SIZE_INV_SHIFT + * + * We can omit the first three elements, because we never + * divide by 0, and QUANTUM and 2*QUANTUM are both powers of + * two, which are handled above. + */ +#define SIZE_INV_SHIFT 21 +#define QSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << QUANTUM_2POW)) + 1) + static const unsigned qsize_invs[] = { + QSIZE_INV(3), + QSIZE_INV(4), QSIZE_INV(5), QSIZE_INV(6), QSIZE_INV(7) +#if (QUANTUM_2POW < 4) + , + QSIZE_INV(8), QSIZE_INV(9), QSIZE_INV(10), QSIZE_INV(11), + QSIZE_INV(12),QSIZE_INV(13), QSIZE_INV(14), QSIZE_INV(15) +#endif + }; + assert(QUANTUM * (((sizeof(qsize_invs)) / sizeof(unsigned)) + 3) + >= (1U << QSPACE_MAX_2POW_DEFAULT)); + + if (size <= (((sizeof(qsize_invs) / sizeof(unsigned)) + 2) << + QUANTUM_2POW)) { + regind = qsize_invs[(size >> QUANTUM_2POW) - 3] * diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef QSIZE_INV + } else if (size < cspace_max) { +#define CSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << CACHELINE_2POW)) + 1) + static const unsigned csize_invs[] = { + CSIZE_INV(3), + CSIZE_INV(4), CSIZE_INV(5), CSIZE_INV(6), CSIZE_INV(7) + }; + assert(CACHELINE * (((sizeof(csize_invs)) / sizeof(unsigned)) + + 3) >= (1U << CSPACE_MAX_2POW_DEFAULT)); + + if (size <= (((sizeof(csize_invs) / sizeof(unsigned)) + 2) << + CACHELINE_2POW)) { + regind = csize_invs[(size >> CACHELINE_2POW) - 3] * + diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef CSIZE_INV + } else { +#define SSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << SUBPAGE_2POW)) + 1) + static const unsigned ssize_invs[] = { + SSIZE_INV(3), + SSIZE_INV(4), SSIZE_INV(5), SSIZE_INV(6), SSIZE_INV(7), + SSIZE_INV(8), SSIZE_INV(9), SSIZE_INV(10), SSIZE_INV(11), + SSIZE_INV(12), SSIZE_INV(13), SSIZE_INV(14), SSIZE_INV(15) +#if (PAGESIZE_2POW == 13) + , + SSIZE_INV(16), SSIZE_INV(17), SSIZE_INV(18), SSIZE_INV(19), + SSIZE_INV(20), SSIZE_INV(21), SSIZE_INV(22), SSIZE_INV(23), + SSIZE_INV(24), SSIZE_INV(25), SSIZE_INV(26), SSIZE_INV(27), + SSIZE_INV(28), SSIZE_INV(29), SSIZE_INV(29), SSIZE_INV(30) +#endif + }; + assert(SUBPAGE * (((sizeof(ssize_invs)) / sizeof(unsigned)) + 3) + >= (1U << PAGESIZE_2POW)); + + if (size < (((sizeof(ssize_invs) / sizeof(unsigned)) + 2) << + SUBPAGE_2POW)) { + regind = ssize_invs[(size >> SUBPAGE_2POW) - 3] * diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef SSIZE_INV + } +#undef SIZE_INV_SHIFT + assert(diff == regind * size); + assert(regind < bin->nregs); + + elm = regind >> (SIZEOF_INT_2POW + 3); + if (elm < run->regs_minelm) + run->regs_minelm = elm; + bit = regind - (elm << (SIZEOF_INT_2POW + 3)); + assert((run->regs_mask[elm] & (1U << bit)) == 0); + run->regs_mask[elm] |= (1U << bit); +} + +static void +arena_run_split(arena_t *arena, arena_run_t *run, size_t size, bool large, + bool zero) +{ + arena_chunk_t *chunk; + size_t old_ndirty, run_ind, total_pages, need_pages, rem_pages, i; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run); + old_ndirty = chunk->ndirty; + run_ind = (unsigned)(((uintptr_t)run - (uintptr_t)chunk) + >> pagesize_2pow); + total_pages = (chunk->map[run_ind].bits & ~pagesize_mask) >> + pagesize_2pow; + need_pages = (size >> pagesize_2pow); + assert(need_pages > 0); + assert(need_pages <= total_pages); + rem_pages = total_pages - need_pages; + + arena_avail_tree_remove(&arena->runs_avail, &chunk->map[run_ind]); + + /* Keep track of trailing unused pages for later use. */ + if (rem_pages > 0) { + chunk->map[run_ind+need_pages].bits = (rem_pages << + pagesize_2pow) | (chunk->map[run_ind+need_pages].bits & + pagesize_mask); + chunk->map[run_ind+total_pages-1].bits = (rem_pages << + pagesize_2pow) | (chunk->map[run_ind+total_pages-1].bits & + pagesize_mask); + arena_avail_tree_insert(&arena->runs_avail, + &chunk->map[run_ind+need_pages]); + } + + for (i = 0; i < need_pages; i++) { + /* Zero if necessary. */ + if (zero) { + if ((chunk->map[run_ind + i].bits & CHUNK_MAP_ZEROED) + == 0) { + memset((void *)((uintptr_t)chunk + ((run_ind + + i) << pagesize_2pow)), 0, pagesize); + /* CHUNK_MAP_ZEROED is cleared below. */ + } + } + + /* Update dirty page accounting. */ + if (chunk->map[run_ind + i].bits & CHUNK_MAP_DIRTY) { + chunk->ndirty--; + arena->ndirty--; + /* CHUNK_MAP_DIRTY is cleared below. */ + } + + /* Initialize the chunk map. */ + if (large) { + chunk->map[run_ind + i].bits = CHUNK_MAP_LARGE + | CHUNK_MAP_ALLOCATED; + } else { + chunk->map[run_ind + i].bits = (size_t)run + | CHUNK_MAP_ALLOCATED; + } + } + + /* + * Set the run size only in the first element for large runs. This is + * primarily a debugging aid, since the lack of size info for trailing + * pages only matters if the application tries to operate on an + * interior pointer. + */ + if (large) + chunk->map[run_ind].bits |= size; + + if (chunk->ndirty == 0 && old_ndirty > 0) + arena_chunk_tree_dirty_remove(&arena->chunks_dirty, chunk); +} + +static arena_chunk_t * +arena_chunk_alloc(arena_t *arena) +{ + arena_chunk_t *chunk; + size_t i; + + if (arena->spare != NULL) { + chunk = arena->spare; + arena->spare = NULL; + } else { + chunk = (arena_chunk_t *)chunk_alloc(chunksize, true); + if (chunk == NULL) + return (NULL); +#ifdef MALLOC_STATS + arena->stats.mapped += chunksize; +#endif + + chunk->arena = arena; + + /* + * Claim that no pages are in use, since the header is merely + * overhead. + */ + chunk->ndirty = 0; + + /* + * Initialize the map to contain one maximal free untouched run. + */ + for (i = 0; i < arena_chunk_header_npages; i++) + chunk->map[i].bits = 0; + chunk->map[i].bits = arena_maxclass | CHUNK_MAP_ZEROED; + for (i++; i < chunk_npages-1; i++) { + chunk->map[i].bits = CHUNK_MAP_ZEROED; + } + chunk->map[chunk_npages-1].bits = arena_maxclass | + CHUNK_MAP_ZEROED; + } + + /* Insert the run into the runs_avail tree. */ + arena_avail_tree_insert(&arena->runs_avail, + &chunk->map[arena_chunk_header_npages]); + + return (chunk); +} + +static void +arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk) +{ + + if (arena->spare != NULL) { + if (arena->spare->ndirty > 0) { + arena_chunk_tree_dirty_remove( + &chunk->arena->chunks_dirty, arena->spare); + arena->ndirty -= arena->spare->ndirty; + } + chunk_dealloc((void *)arena->spare, chunksize); +#ifdef MALLOC_STATS + arena->stats.mapped -= chunksize; +#endif + } + + /* + * Remove run from runs_avail, regardless of whether this chunk + * will be cached, so that the arena does not use it. Dirty page + * flushing only uses the chunks_dirty tree, so leaving this chunk in + * the chunks_* trees is sufficient for that purpose. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[arena_chunk_header_npages]); + + arena->spare = chunk; +} + +static arena_run_t * +arena_run_alloc(arena_t *arena, size_t size, bool large, bool zero) +{ + arena_chunk_t *chunk; + arena_run_t *run; + arena_chunk_map_t *mapelm, key; + + assert(size <= arena_maxclass); + assert((size & pagesize_mask) == 0); + + /* Search the arena's chunks for the lowest best fit. */ + key.bits = size | CHUNK_MAP_KEY; + mapelm = arena_avail_tree_nsearch(&arena->runs_avail, &key); + if (mapelm != NULL) { + arena_chunk_t *run_chunk = CHUNK_ADDR2BASE(mapelm); + size_t pageind = ((uintptr_t)mapelm - (uintptr_t)run_chunk->map) + / sizeof(arena_chunk_map_t); + + run = (arena_run_t *)((uintptr_t)run_chunk + (pageind + << pagesize_2pow)); + arena_run_split(arena, run, size, large, zero); + return (run); + } + + /* + * No usable runs. Create a new chunk from which to allocate the run. + */ + chunk = arena_chunk_alloc(arena); + if (chunk == NULL) + return (NULL); + run = (arena_run_t *)((uintptr_t)chunk + (arena_chunk_header_npages << + pagesize_2pow)); + /* Update page map. */ + arena_run_split(arena, run, size, large, zero); + return (run); +} + +static void +arena_purge(arena_t *arena) +{ + arena_chunk_t *chunk; + size_t i, npages; +#ifdef MALLOC_DEBUG + size_t ndirty = 0; + + rb_foreach_begin(arena_chunk_t, link_dirty, &arena->chunks_dirty, + chunk) { + ndirty += chunk->ndirty; + } rb_foreach_end(arena_chunk_t, link_dirty, &arena->chunks_dirty, chunk) + assert(ndirty == arena->ndirty); +#endif + assert(arena->ndirty > opt_dirty_max); + +#ifdef MALLOC_STATS + arena->stats.npurge++; +#endif + + /* + * Iterate downward through chunks until enough dirty memory has been + * purged. Terminate as soon as possible in order to minimize the + * number of system calls, even if a chunk has only been partially + * purged. + */ + while (arena->ndirty > (opt_dirty_max >> 1)) { + chunk = arena_chunk_tree_dirty_last(&arena->chunks_dirty); + assert(chunk != NULL); + + for (i = chunk_npages - 1; chunk->ndirty > 0; i--) { + assert(i >= arena_chunk_header_npages); + + if (chunk->map[i].bits & CHUNK_MAP_DIRTY) { + chunk->map[i].bits ^= CHUNK_MAP_DIRTY; + /* Find adjacent dirty run(s). */ + for (npages = 1; i > arena_chunk_header_npages + && (chunk->map[i - 1].bits & + CHUNK_MAP_DIRTY); npages++) { + i--; + chunk->map[i].bits ^= CHUNK_MAP_DIRTY; + } + chunk->ndirty -= npages; + arena->ndirty -= npages; + + madvise((void *)((uintptr_t)chunk + (i << + pagesize_2pow)), (npages << pagesize_2pow), + MADV_DONTNEED); +#ifdef MALLOC_STATS + arena->stats.nmadvise++; + arena->stats.purged += npages; +#endif + if (arena->ndirty <= (opt_dirty_max >> 1)) + break; + } + } + + if (chunk->ndirty == 0) { + arena_chunk_tree_dirty_remove(&arena->chunks_dirty, + chunk); + } + } +} + +static void +arena_run_dalloc(arena_t *arena, arena_run_t *run, bool dirty) +{ + arena_chunk_t *chunk; + size_t size, run_ind, run_pages; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run); + run_ind = (size_t)(((uintptr_t)run - (uintptr_t)chunk) + >> pagesize_2pow); + assert(run_ind >= arena_chunk_header_npages); + assert(run_ind < chunk_npages); + if ((chunk->map[run_ind].bits & CHUNK_MAP_LARGE) != 0) + size = chunk->map[run_ind].bits & ~pagesize_mask; + else + size = run->bin->run_size; + run_pages = (size >> pagesize_2pow); + + /* Mark pages as unallocated in the chunk map. */ + if (dirty) { + size_t i; + + for (i = 0; i < run_pages; i++) { + assert((chunk->map[run_ind + i].bits & CHUNK_MAP_DIRTY) + == 0); + chunk->map[run_ind + i].bits = CHUNK_MAP_DIRTY; + } + + if (chunk->ndirty == 0) { + arena_chunk_tree_dirty_insert(&arena->chunks_dirty, + chunk); + } + chunk->ndirty += run_pages; + arena->ndirty += run_pages; + } else { + size_t i; + + for (i = 0; i < run_pages; i++) { + chunk->map[run_ind + i].bits &= ~(CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED); + } + } + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + + /* Try to coalesce forward. */ + if (run_ind + run_pages < chunk_npages && + (chunk->map[run_ind+run_pages].bits & CHUNK_MAP_ALLOCATED) == 0) { + size_t nrun_size = chunk->map[run_ind+run_pages].bits & + ~pagesize_mask; + + /* + * Remove successor from runs_avail; the coalesced run is + * inserted later. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[run_ind+run_pages]); + + size += nrun_size; + run_pages = size >> pagesize_2pow; + + assert((chunk->map[run_ind+run_pages-1].bits & ~pagesize_mask) + == nrun_size); + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + } + + /* Try to coalesce backward. */ + if (run_ind > arena_chunk_header_npages && (chunk->map[run_ind-1].bits & + CHUNK_MAP_ALLOCATED) == 0) { + size_t prun_size = chunk->map[run_ind-1].bits & ~pagesize_mask; + + run_ind -= prun_size >> pagesize_2pow; + + /* + * Remove predecessor from runs_avail; the coalesced run is + * inserted later. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[run_ind]); + + size += prun_size; + run_pages = size >> pagesize_2pow; + + assert((chunk->map[run_ind].bits & ~pagesize_mask) == + prun_size); + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + } + + /* Insert into runs_avail, now that coalescing is complete. */ + arena_avail_tree_insert(&arena->runs_avail, &chunk->map[run_ind]); + + /* Deallocate chunk if it is now completely unused. */ + if ((chunk->map[arena_chunk_header_npages].bits & (~pagesize_mask | + CHUNK_MAP_ALLOCATED)) == arena_maxclass) + arena_chunk_dealloc(arena, chunk); + + /* Enforce opt_dirty_max. */ + if (arena->ndirty > opt_dirty_max) + arena_purge(arena); +} + +static void +arena_run_trim_head(arena_t *arena, arena_chunk_t *chunk, arena_run_t *run, + size_t oldsize, size_t newsize) +{ + size_t pageind = ((uintptr_t)run - (uintptr_t)chunk) >> pagesize_2pow; + size_t head_npages = (oldsize - newsize) >> pagesize_2pow; + + assert(oldsize > newsize); + + /* + * Update the chunk map so that arena_run_dalloc() can treat the + * leading run as separately allocated. + */ + chunk->map[pageind].bits = (oldsize - newsize) | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+head_npages].bits = newsize | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + + arena_run_dalloc(arena, run, false); +} + +static void +arena_run_trim_tail(arena_t *arena, arena_chunk_t *chunk, arena_run_t *run, + size_t oldsize, size_t newsize, bool dirty) +{ + size_t pageind = ((uintptr_t)run - (uintptr_t)chunk) >> pagesize_2pow; + size_t npages = newsize >> pagesize_2pow; + + assert(oldsize > newsize); + + /* + * Update the chunk map so that arena_run_dalloc() can treat the + * trailing run as separately allocated. + */ + chunk->map[pageind].bits = newsize | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+npages].bits = (oldsize - newsize) | CHUNK_MAP_LARGE + | CHUNK_MAP_ALLOCATED; + + arena_run_dalloc(arena, (arena_run_t *)((uintptr_t)run + newsize), + dirty); +} + +static arena_run_t * +arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin) +{ + arena_chunk_map_t *mapelm; + arena_run_t *run; + unsigned i, remainder; + + /* Look for a usable run. */ + mapelm = arena_run_tree_first(&bin->runs); + if (mapelm != NULL) { + /* run is guaranteed to have available space. */ + arena_run_tree_remove(&bin->runs, mapelm); + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); +#ifdef MALLOC_STATS + bin->stats.reruns++; +#endif + return (run); + } + /* No existing runs have any space available. */ + + /* Allocate a new run. */ + run = arena_run_alloc(arena, bin->run_size, false, false); + if (run == NULL) + return (NULL); + + /* Initialize run internals. */ + run->bin = bin; + + for (i = 0; i < bin->regs_mask_nelms - 1; i++) + run->regs_mask[i] = UINT_MAX; + remainder = bin->nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1); + if (remainder == 0) + run->regs_mask[i] = UINT_MAX; + else { + /* The last element has spare bits that need to be unset. */ + run->regs_mask[i] = (UINT_MAX >> ((1U << (SIZEOF_INT_2POW + 3)) + - remainder)); + } + + run->regs_minelm = 0; + + run->nfree = bin->nregs; +#ifdef MALLOC_DEBUG + run->magic = ARENA_RUN_MAGIC; +#endif + +#ifdef MALLOC_STATS + bin->stats.nruns++; + bin->stats.curruns++; + if (bin->stats.curruns > bin->stats.highruns) + bin->stats.highruns = bin->stats.curruns; +#endif + return (run); +} + +/* bin->runcur must have space available before this function is called. */ +static inline void * +arena_bin_malloc_easy(arena_t *arena, arena_bin_t *bin, arena_run_t *run) +{ + void *ret; + + assert(run->magic == ARENA_RUN_MAGIC); + assert(run->nfree > 0); + + ret = arena_run_reg_alloc(run, bin); + assert(ret != NULL); + run->nfree--; + + return (ret); +} + +/* Re-fill bin->runcur, then call arena_bin_malloc_easy(). */ +static void * +arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin) +{ + + bin->runcur = arena_bin_nonfull_run_get(arena, bin); + if (bin->runcur == NULL) + return (NULL); + assert(bin->runcur->magic == ARENA_RUN_MAGIC); + assert(bin->runcur->nfree > 0); + + return (arena_bin_malloc_easy(arena, bin, bin->runcur)); +} + +/* + * Calculate bin->run_size such that it meets the following constraints: + * + * *) bin->run_size >= min_run_size + * *) bin->run_size <= arena_maxclass + * *) bin->run_size <= RUN_MAX_SMALL + * *) run header overhead <= RUN_MAX_OVRHD (or header overhead relaxed). + * + * bin->nregs, bin->regs_mask_nelms, and bin->reg0_offset are + * also calculated here, since these settings are all interdependent. + */ +static size_t +arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size) +{ + size_t try_run_size, good_run_size; + unsigned good_nregs, good_mask_nelms, good_reg0_offset; + unsigned try_nregs, try_mask_nelms, try_reg0_offset; + + assert(min_run_size >= pagesize); + assert(min_run_size <= arena_maxclass); + assert(min_run_size <= RUN_MAX_SMALL); + + /* + * Calculate known-valid settings before entering the run_size + * expansion loop, so that the first part of the loop always copies + * valid settings. + * + * The do..while loop iteratively reduces the number of regions until + * the run header and the regions no longer overlap. A closed formula + * would be quite messy, since there is an interdependency between the + * header's mask length and the number of regions. + */ + try_run_size = min_run_size; + try_nregs = ((try_run_size - sizeof(arena_run_t)) / bin->reg_size) + + 1; /* Counter-act try_nregs-- in loop. */ + do { + try_nregs--; + try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) + + ((try_nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1)) ? 1 : 0); + try_reg0_offset = try_run_size - (try_nregs * bin->reg_size); + } while (sizeof(arena_run_t) + (sizeof(unsigned) * (try_mask_nelms - 1)) + > try_reg0_offset); + + /* run_size expansion loop. */ + do { + /* + * Copy valid settings before trying more aggressive settings. + */ + good_run_size = try_run_size; + good_nregs = try_nregs; + good_mask_nelms = try_mask_nelms; + good_reg0_offset = try_reg0_offset; + + /* Try more aggressive settings. */ + try_run_size += pagesize; + try_nregs = ((try_run_size - sizeof(arena_run_t)) / + bin->reg_size) + 1; /* Counter-act try_nregs-- in loop. */ + do { + try_nregs--; + try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) + + ((try_nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1)) ? + 1 : 0); + try_reg0_offset = try_run_size - (try_nregs * + bin->reg_size); + } while (sizeof(arena_run_t) + (sizeof(unsigned) * + (try_mask_nelms - 1)) > try_reg0_offset); + } while (try_run_size <= arena_maxclass && try_run_size <= RUN_MAX_SMALL + && RUN_MAX_OVRHD * (bin->reg_size << 3) > RUN_MAX_OVRHD_RELAX + && (try_reg0_offset << RUN_BFP) > RUN_MAX_OVRHD * try_run_size); + + assert(sizeof(arena_run_t) + (sizeof(unsigned) * (good_mask_nelms - 1)) + <= good_reg0_offset); + assert((good_mask_nelms << (SIZEOF_INT_2POW + 3)) >= good_nregs); + + /* Copy final settings. */ + bin->run_size = good_run_size; + bin->nregs = good_nregs; + bin->regs_mask_nelms = good_mask_nelms; + bin->reg0_offset = good_reg0_offset; + + return (good_run_size); +} + +#ifdef MALLOC_BALANCE +static inline void +arena_lock_balance(arena_t *arena) +{ + unsigned contention; + + contention = malloc_spin_lock(&arena->lock); + if (narenas > 1) { + /* + * Calculate the exponentially averaged contention for this + * arena. Due to integer math always rounding down, this value + * decays somewhat faster than normal. + */ + arena->contention = (((uint64_t)arena->contention + * (uint64_t)((1U << BALANCE_ALPHA_INV_2POW)-1)) + + (uint64_t)contention) >> BALANCE_ALPHA_INV_2POW; + if (arena->contention >= opt_balance_threshold) + arena_lock_balance_hard(arena); + } +} + +static void +arena_lock_balance_hard(arena_t *arena) +{ + uint32_t ind; + + arena->contention = 0; +#ifdef MALLOC_STATS + arena->stats.nbalance++; +#endif + ind = PRN(balance, narenas_2pow); + if (arenas[ind] != NULL) + arenas_map = arenas[ind]; + else { + malloc_spin_lock(&arenas_lock); + if (arenas[ind] != NULL) + arenas_map = arenas[ind]; + else + arenas_map = arenas_extend(ind); + malloc_spin_unlock(&arenas_lock); + } +} +#endif + +#ifdef MALLOC_MAG +static inline void * +mag_alloc(mag_t *mag) +{ + + if (mag->nrounds == 0) + return (NULL); + mag->nrounds--; + + return (mag->rounds[mag->nrounds]); +} + +static void +mag_load(mag_t *mag) +{ + arena_t *arena; + arena_bin_t *bin; + arena_run_t *run; + void *round; + size_t i; + + arena = choose_arena(); + bin = &arena->bins[mag->binind]; +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + for (i = mag->nrounds; i < max_rounds; i++) { + if ((run = bin->runcur) != NULL && run->nfree > 0) + round = arena_bin_malloc_easy(arena, bin, run); + else + round = arena_bin_malloc_hard(arena, bin); + if (round == NULL) + break; + mag->rounds[i] = round; + } +#ifdef MALLOC_STATS + bin->stats.nmags++; + arena->stats.nmalloc_small += (i - mag->nrounds); + arena->stats.allocated_small += (i - mag->nrounds) * bin->reg_size; +#endif + malloc_spin_unlock(&arena->lock); + mag->nrounds = i; +} + +static inline void * +mag_rack_alloc(mag_rack_t *rack, size_t size, bool zero) +{ + void *ret; + bin_mags_t *bin_mags; + mag_t *mag; + size_t binind; + + binind = size2bin[size]; + assert(binind < nbins); + bin_mags = &rack->bin_mags[binind]; + + mag = bin_mags->curmag; + if (mag == NULL) { + /* Create an initial magazine for this size class. */ + assert(bin_mags->sparemag == NULL); + mag = mag_create(choose_arena(), binind); + if (mag == NULL) + return (NULL); + bin_mags->curmag = mag; + mag_load(mag); + } + + ret = mag_alloc(mag); + if (ret == NULL) { + if (bin_mags->sparemag != NULL) { + if (bin_mags->sparemag->nrounds > 0) { + /* Swap magazines. */ + bin_mags->curmag = bin_mags->sparemag; + bin_mags->sparemag = mag; + mag = bin_mags->curmag; + } else { + /* Reload the current magazine. */ + mag_load(mag); + } + } else { + /* Create a second magazine. */ + mag = mag_create(choose_arena(), binind); + if (mag == NULL) + return (NULL); + mag_load(mag); + bin_mags->sparemag = bin_mags->curmag; + bin_mags->curmag = mag; + } + ret = mag_alloc(mag); + if (ret == NULL) + return (NULL); + } + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } else + memset(ret, 0, size); + + return (ret); +} +#endif + +static inline void * +arena_malloc_small(arena_t *arena, size_t size, bool zero) +{ + void *ret; + arena_bin_t *bin; + arena_run_t *run; + size_t binind; + + binind = size2bin[size]; + assert(binind < nbins); + bin = &arena->bins[binind]; + size = bin->reg_size; + +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + if ((run = bin->runcur) != NULL && run->nfree > 0) + ret = arena_bin_malloc_easy(arena, bin, run); + else + ret = arena_bin_malloc_hard(arena, bin); + + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } + +#ifdef MALLOC_STATS + bin->stats.nrequests++; + arena->stats.nmalloc_small++; + arena->stats.allocated_small += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } else + memset(ret, 0, size); + + return (ret); +} + +static void * +arena_malloc_large(arena_t *arena, size_t size, bool zero) +{ + void *ret; + + /* Large allocation. */ + size = PAGE_CEILING(size); +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + ret = (void *)arena_run_alloc(arena, size, true, zero); + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } +#ifdef MALLOC_STATS + arena->stats.nmalloc_large++; + arena->stats.allocated_large += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } + + return (ret); +} + +static inline void * +arena_malloc(arena_t *arena, size_t size, bool zero) +{ + + assert(arena != NULL); + assert(arena->magic == ARENA_MAGIC); + assert(size != 0); + assert(QUANTUM_CEILING(size) <= arena_maxclass); + + if (size <= bin_maxclass) { +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + mag_rack_t *rack = mag_rack; + if (rack == NULL) { + rack = mag_rack_create(arena); + if (rack == NULL) + return (NULL); + mag_rack = rack; + } + return (mag_rack_alloc(rack, size, zero)); + } else +#endif + return (arena_malloc_small(arena, size, zero)); + } else + return (arena_malloc_large(arena, size, zero)); +} + +static inline void * +imalloc(size_t size) +{ + + assert(size != 0); + + if (size <= arena_maxclass) + return (arena_malloc(choose_arena(), size, false)); + else + return (huge_malloc(size, false)); +} + +static inline void * +icalloc(size_t size) +{ + + if (size <= arena_maxclass) + return (arena_malloc(choose_arena(), size, true)); + else + return (huge_malloc(size, true)); +} + +/* Only handles large allocations that require more than page alignment. */ +static void * +arena_palloc(arena_t *arena, size_t alignment, size_t size, size_t alloc_size) +{ + void *ret; + size_t offset; + arena_chunk_t *chunk; + + assert((size & pagesize_mask) == 0); + assert((alignment & pagesize_mask) == 0); + +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + ret = (void *)arena_run_alloc(arena, alloc_size, true, false); + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ret); + + offset = (uintptr_t)ret & (alignment - 1); + assert((offset & pagesize_mask) == 0); + assert(offset < alloc_size); + if (offset == 0) + arena_run_trim_tail(arena, chunk, ret, alloc_size, size, false); + else { + size_t leadsize, trailsize; + + leadsize = alignment - offset; + if (leadsize > 0) { + arena_run_trim_head(arena, chunk, ret, alloc_size, + alloc_size - leadsize); + ret = (void *)((uintptr_t)ret + leadsize); + } + + trailsize = alloc_size - leadsize - size; + if (trailsize != 0) { + /* Trim trailing space. */ + assert(trailsize < alloc_size); + arena_run_trim_tail(arena, chunk, ret, size + trailsize, + size, false); + } + } + +#ifdef MALLOC_STATS + arena->stats.nmalloc_large++; + arena->stats.allocated_large += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + return (ret); +} + +static inline void * +ipalloc(size_t alignment, size_t size) +{ + void *ret; + size_t ceil_size; + + /* + * Round size up to the nearest multiple of alignment. + * + * This done, we can take advantage of the fact that for each small + * size class, every object is aligned at the smallest power of two + * that is non-zero in the base two representation of the size. For + * example: + * + * Size | Base 2 | Minimum alignment + * -----+----------+------------------ + * 96 | 1100000 | 32 + * 144 | 10100000 | 32 + * 192 | 11000000 | 64 + * + * Depending on runtime settings, it is possible that arena_malloc() + * will further round up to a power of two, but that never causes + * correctness issues. + */ + ceil_size = (size + (alignment - 1)) & (-alignment); + /* + * (ceil_size < size) protects against the combination of maximal + * alignment and size greater than maximal alignment. + */ + if (ceil_size < size) { + /* size_t overflow. */ + return (NULL); + } + + if (ceil_size <= pagesize || (alignment <= pagesize + && ceil_size <= arena_maxclass)) + ret = arena_malloc(choose_arena(), ceil_size, false); + else { + size_t run_size; + + /* + * We can't achieve subpage alignment, so round up alignment + * permanently; it makes later calculations simpler. + */ + alignment = PAGE_CEILING(alignment); + ceil_size = PAGE_CEILING(size); + /* + * (ceil_size < size) protects against very large sizes within + * pagesize of SIZE_T_MAX. + * + * (ceil_size + alignment < ceil_size) protects against the + * combination of maximal alignment and ceil_size large enough + * to cause overflow. This is similar to the first overflow + * check above, but it needs to be repeated due to the new + * ceil_size value, which may now be *equal* to maximal + * alignment, whereas before we only detected overflow if the + * original size was *greater* than maximal alignment. + */ + if (ceil_size < size || ceil_size + alignment < ceil_size) { + /* size_t overflow. */ + return (NULL); + } + + /* + * Calculate the size of the over-size run that arena_palloc() + * would need to allocate in order to guarantee the alignment. + */ + if (ceil_size >= alignment) + run_size = ceil_size + alignment - pagesize; + else { + /* + * It is possible that (alignment << 1) will cause + * overflow, but it doesn't matter because we also + * subtract pagesize, which in the case of overflow + * leaves us with a very large run_size. That causes + * the first conditional below to fail, which means + * that the bogus run_size value never gets used for + * anything important. + */ + run_size = (alignment << 1) - pagesize; + } + + if (run_size <= arena_maxclass) { + ret = arena_palloc(choose_arena(), alignment, ceil_size, + run_size); + } else if (alignment <= chunksize) + ret = huge_malloc(ceil_size, false); + else + ret = huge_palloc(alignment, ceil_size); + } + + assert(((uintptr_t)ret & (alignment - 1)) == 0); + return (ret); +} + +/* Return the size of the allocation pointed to by ptr. */ +static size_t +arena_salloc(const void *ptr) +{ + size_t ret; + arena_chunk_t *chunk; + size_t pageind, mapbits; + + assert(ptr != NULL); + assert(CHUNK_ADDR2BASE(ptr) != ptr); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapbits = chunk->map[pageind].bits; + assert((mapbits & CHUNK_MAP_ALLOCATED) != 0); + if ((mapbits & CHUNK_MAP_LARGE) == 0) { + arena_run_t *run = (arena_run_t *)(mapbits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + ret = run->bin->reg_size; + } else { + ret = mapbits & ~pagesize_mask; + assert(ret != 0); + } + + return (ret); +} + +static inline size_t +isalloc(const void *ptr) +{ + size_t ret; + arena_chunk_t *chunk; + + assert(ptr != NULL); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + if (chunk != ptr) { + /* Region. */ + assert(chunk->arena->magic == ARENA_MAGIC); + + ret = arena_salloc(ptr); + } else { + extent_node_t *node, key; + + /* Chunk (huge allocation). */ + + malloc_mutex_lock(&huge_mtx); + + /* Extract from tree of huge allocations. */ + key.addr = __DECONST(void *, ptr); + node = extent_tree_ad_search(&huge, &key); + assert(node != NULL); + + ret = node->size; + + malloc_mutex_unlock(&huge_mtx); + } + + return (ret); +} + +static inline void +arena_dalloc_small(arena_t *arena, arena_chunk_t *chunk, void *ptr, + arena_chunk_map_t *mapelm) +{ + arena_run_t *run; + arena_bin_t *bin; + size_t size; + + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + bin = run->bin; + size = bin->reg_size; + + if (opt_junk) + memset(ptr, 0x5a, size); + + arena_run_reg_dalloc(run, bin, ptr, size); + run->nfree++; + + if (run->nfree == bin->nregs) { + /* Deallocate run. */ + if (run == bin->runcur) + bin->runcur = NULL; + else if (bin->nregs != 1) { + size_t run_pageind = (((uintptr_t)run - + (uintptr_t)chunk)) >> pagesize_2pow; + arena_chunk_map_t *run_mapelm = + &chunk->map[run_pageind]; + /* + * This block's conditional is necessary because if the + * run only contains one region, then it never gets + * inserted into the non-full runs tree. + */ + arena_run_tree_remove(&bin->runs, run_mapelm); + } +#ifdef MALLOC_DEBUG + run->magic = 0; +#endif + arena_run_dalloc(arena, run, true); +#ifdef MALLOC_STATS + bin->stats.curruns--; +#endif + } else if (run->nfree == 1 && run != bin->runcur) { + /* + * Make sure that bin->runcur always refers to the lowest + * non-full run, if one exists. + */ + if (bin->runcur == NULL) + bin->runcur = run; + else if ((uintptr_t)run < (uintptr_t)bin->runcur) { + /* Switch runcur. */ + if (bin->runcur->nfree > 0) { + arena_chunk_t *runcur_chunk = + CHUNK_ADDR2BASE(bin->runcur); + size_t runcur_pageind = + (((uintptr_t)bin->runcur - + (uintptr_t)runcur_chunk)) >> pagesize_2pow; + arena_chunk_map_t *runcur_mapelm = + &runcur_chunk->map[runcur_pageind]; + + /* Insert runcur. */ + arena_run_tree_insert(&bin->runs, + runcur_mapelm); + } + bin->runcur = run; + } else { + size_t run_pageind = (((uintptr_t)run - + (uintptr_t)chunk)) >> pagesize_2pow; + arena_chunk_map_t *run_mapelm = + &chunk->map[run_pageind]; + + assert(arena_run_tree_search(&bin->runs, run_mapelm) == + NULL); + arena_run_tree_insert(&bin->runs, run_mapelm); + } + } +#ifdef MALLOC_STATS + arena->stats.allocated_small -= size; + arena->stats.ndalloc_small++; +#endif +} + +#ifdef MALLOC_MAG +static void +mag_unload(mag_t *mag) +{ + arena_chunk_t *chunk; + arena_t *arena; + void *round; + size_t i, ndeferred, nrounds; + + for (ndeferred = mag->nrounds; ndeferred > 0;) { + nrounds = ndeferred; + /* Lock the arena associated with the first round. */ + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(mag->rounds[0]); + arena = chunk->arena; +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + /* Deallocate every round that belongs to the locked arena. */ + for (i = ndeferred = 0; i < nrounds; i++) { + round = mag->rounds[i]; + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(round); + if (chunk->arena == arena) { + size_t pageind = (((uintptr_t)round - + (uintptr_t)chunk) >> pagesize_2pow); + arena_chunk_map_t *mapelm = + &chunk->map[pageind]; + arena_dalloc_small(arena, chunk, round, mapelm); + } else { + /* + * This round was allocated via a different + * arena than the one that is currently locked. + * Stash the round, so that it can be handled + * in a future pass. + */ + mag->rounds[ndeferred] = round; + ndeferred++; + } + } + malloc_spin_unlock(&arena->lock); + } + + mag->nrounds = 0; +} + +static inline void +mag_rack_dalloc(mag_rack_t *rack, void *ptr) +{ + arena_t *arena; + arena_chunk_t *chunk; + arena_run_t *run; + arena_bin_t *bin; + bin_mags_t *bin_mags; + mag_t *mag; + size_t pageind, binind; + arena_chunk_map_t *mapelm; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + arena = chunk->arena; + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + bin = run->bin; + binind = ((uintptr_t)bin - (uintptr_t)&arena->bins) / + sizeof(arena_bin_t); + assert(binind < nbins); + + if (opt_junk) + memset(ptr, 0x5a, arena->bins[binind].reg_size); + + bin_mags = &rack->bin_mags[binind]; + mag = bin_mags->curmag; + if (mag == NULL) { + /* Create an initial magazine for this size class. */ + assert(bin_mags->sparemag == NULL); + mag = mag_create(choose_arena(), binind); + if (mag == NULL) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, mapelm); + malloc_spin_unlock(&arena->lock); + return; + } + bin_mags->curmag = mag; + } + + if (mag->nrounds == max_rounds) { + if (bin_mags->sparemag != NULL) { + if (bin_mags->sparemag->nrounds < max_rounds) { + /* Swap magazines. */ + bin_mags->curmag = bin_mags->sparemag; + bin_mags->sparemag = mag; + mag = bin_mags->curmag; + } else { + /* Unload the current magazine. */ + mag_unload(mag); + } + } else { + /* Create a second magazine. */ + mag = mag_create(choose_arena(), binind); + if (mag == NULL) { + mag = rack->bin_mags[binind].curmag; + mag_unload(mag); + } else { + bin_mags->sparemag = bin_mags->curmag; + bin_mags->curmag = mag; + } + } + assert(mag->nrounds < max_rounds); + } + mag->rounds[mag->nrounds] = ptr; + mag->nrounds++; +} +#endif + +static void +arena_dalloc_large(arena_t *arena, arena_chunk_t *chunk, void *ptr) +{ + /* Large allocation. */ + malloc_spin_lock(&arena->lock); + +#ifndef MALLOC_STATS + if (opt_junk) +#endif + { + size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> + pagesize_2pow; + size_t size = chunk->map[pageind].bits & ~pagesize_mask; + +#ifdef MALLOC_STATS + if (opt_junk) +#endif + memset(ptr, 0x5a, size); +#ifdef MALLOC_STATS + arena->stats.allocated_large -= size; +#endif + } +#ifdef MALLOC_STATS + arena->stats.ndalloc_large++; +#endif + + arena_run_dalloc(arena, (arena_run_t *)ptr, true); + malloc_spin_unlock(&arena->lock); +} + +static inline void +arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr) +{ + size_t pageind; + arena_chunk_map_t *mapelm; + + assert(arena != NULL); + assert(arena->magic == ARENA_MAGIC); + assert(chunk->arena == arena); + assert(ptr != NULL); + assert(CHUNK_ADDR2BASE(ptr) != ptr); + + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + assert((mapelm->bits & CHUNK_MAP_ALLOCATED) != 0); + if ((mapelm->bits & CHUNK_MAP_LARGE) == 0) { + /* Small allocation. */ +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + mag_rack_t *rack = mag_rack; + if (rack == NULL) { + rack = mag_rack_create(arena); + if (rack == NULL) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, + mapelm); + malloc_spin_unlock(&arena->lock); + } + mag_rack = rack; + } + mag_rack_dalloc(rack, ptr); + } else { +#endif + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, mapelm); + malloc_spin_unlock(&arena->lock); +#ifdef MALLOC_MAG + } +#endif + } else + arena_dalloc_large(arena, chunk, ptr); +} + +static inline void +idalloc(void *ptr) +{ + arena_chunk_t *chunk; + + assert(ptr != NULL); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + if (chunk != ptr) + arena_dalloc(chunk->arena, chunk, ptr); + else + huge_dalloc(ptr); +} + +static void +arena_ralloc_large_shrink(arena_t *arena, arena_chunk_t *chunk, void *ptr, + size_t size, size_t oldsize) +{ + + assert(size < oldsize); + + /* + * Shrink the run, and make trailing pages available for other + * allocations. + */ +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + arena_run_trim_tail(arena, chunk, (arena_run_t *)ptr, oldsize, size, + true); +#ifdef MALLOC_STATS + arena->stats.allocated_large -= oldsize - size; +#endif + malloc_spin_unlock(&arena->lock); +} + +static bool +arena_ralloc_large_grow(arena_t *arena, arena_chunk_t *chunk, void *ptr, + size_t size, size_t oldsize) +{ + size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow; + size_t npages = oldsize >> pagesize_2pow; + + assert(oldsize == (chunk->map[pageind].bits & ~pagesize_mask)); + + /* Try to extend the run. */ + assert(size > oldsize); +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + if (pageind + npages < chunk_npages && (chunk->map[pageind+npages].bits + & CHUNK_MAP_ALLOCATED) == 0 && (chunk->map[pageind+npages].bits & + ~pagesize_mask) >= size - oldsize) { + /* + * The next run is available and sufficiently large. Split the + * following run, then merge the first part with the existing + * allocation. + */ + arena_run_split(arena, (arena_run_t *)((uintptr_t)chunk + + ((pageind+npages) << pagesize_2pow)), size - oldsize, true, + false); + + chunk->map[pageind].bits = size | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+npages].bits = CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + +#ifdef MALLOC_STATS + arena->stats.allocated_large += size - oldsize; +#endif + malloc_spin_unlock(&arena->lock); + return (false); + } + malloc_spin_unlock(&arena->lock); + + return (true); +} + +/* + * Try to resize a large allocation, in order to avoid copying. This will + * always fail if growing an object, and the following run is already in use. + */ +static bool +arena_ralloc_large(void *ptr, size_t size, size_t oldsize) +{ + size_t psize; + + psize = PAGE_CEILING(size); + if (psize == oldsize) { + /* Same size class. */ + if (opt_junk && size < oldsize) { + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize - + size); + } + return (false); + } else { + arena_chunk_t *chunk; + arena_t *arena; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + arena = chunk->arena; + assert(arena->magic == ARENA_MAGIC); + + if (psize < oldsize) { + /* Fill before shrinking in order avoid a race. */ + if (opt_junk) { + memset((void *)((uintptr_t)ptr + size), 0x5a, + oldsize - size); + } + arena_ralloc_large_shrink(arena, chunk, ptr, psize, + oldsize); + return (false); + } else { + bool ret = arena_ralloc_large_grow(arena, chunk, ptr, + psize, oldsize); + if (ret == false && opt_zero) { + memset((void *)((uintptr_t)ptr + oldsize), 0, + size - oldsize); + } + return (ret); + } + } +} + +static void * +arena_ralloc(void *ptr, size_t size, size_t oldsize) +{ + void *ret; + size_t copysize; + + /* Try to avoid moving the allocation. */ + if (size <= bin_maxclass) { + if (oldsize <= bin_maxclass && size2bin[size] == + size2bin[oldsize]) + goto IN_PLACE; + } else { + if (oldsize > bin_maxclass && oldsize <= arena_maxclass) { + assert(size > bin_maxclass); + if (arena_ralloc_large(ptr, size, oldsize) == false) + return (ptr); + } + } + + /* + * If we get here, then size and oldsize are different enough that we + * need to move the object. In that case, fall back to allocating new + * space and copying. + */ + ret = arena_malloc(choose_arena(), size, false); + if (ret == NULL) + return (NULL); + + /* Junk/zero-filling were already done by arena_malloc(). */ + copysize = (size < oldsize) ? size : oldsize; + memcpy(ret, ptr, copysize); + idalloc(ptr); + return (ret); +IN_PLACE: + if (opt_junk && size < oldsize) + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize - size); + else if (opt_zero && size > oldsize) + memset((void *)((uintptr_t)ptr + oldsize), 0, size - oldsize); + return (ptr); +} + +static inline void * +iralloc(void *ptr, size_t size) +{ + size_t oldsize; + + assert(ptr != NULL); + assert(size != 0); + + oldsize = isalloc(ptr); + + if (size <= arena_maxclass) + return (arena_ralloc(ptr, size, oldsize)); + else + return (huge_ralloc(ptr, size, oldsize)); +} + +static bool +arena_new(arena_t *arena) +{ + unsigned i; + arena_bin_t *bin; + size_t prev_run_size; + + if (malloc_spin_init(&arena->lock)) + return (true); + +#ifdef MALLOC_STATS + memset(&arena->stats, 0, sizeof(arena_stats_t)); +#endif + + /* Initialize chunks. */ + arena_chunk_tree_dirty_new(&arena->chunks_dirty); + arena->spare = NULL; + + arena->ndirty = 0; + + arena_avail_tree_new(&arena->runs_avail); + +#ifdef MALLOC_BALANCE + arena->contention = 0; +#endif + + /* Initialize bins. */ + prev_run_size = pagesize; + + i = 0; +#ifdef MALLOC_TINY + /* (2^n)-spaced tiny bins. */ + for (; i < ntbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = (1U << (TINY_MIN_2POW + i)); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } +#endif + + /* Quantum-spaced bins. */ + for (; i < ntbins + nqbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = (i - ntbins + 1) << QUANTUM_2POW; + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + + /* Cacheline-spaced bins. */ + for (; i < ntbins + nqbins + ncbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = cspace_min + ((i - (ntbins + nqbins)) << + CACHELINE_2POW); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + + /* Subpage-spaced bins. */ + for (; i < nbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = sspace_min + ((i - (ntbins + nqbins + ncbins)) + << SUBPAGE_2POW); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + +#ifdef MALLOC_DEBUG + arena->magic = ARENA_MAGIC; +#endif + + return (false); +} + +/* Create a new arena and insert it into the arenas array at index ind. */ +static arena_t * +arenas_extend(unsigned ind) +{ + arena_t *ret; + + /* Allocate enough space for trailing bins. */ + ret = (arena_t *)base_alloc(sizeof(arena_t) + + (sizeof(arena_bin_t) * (nbins - 1))); + if (ret != NULL && arena_new(ret) == false) { + arenas[ind] = ret; + return (ret); + } + /* Only reached if there is an OOM error. */ + + /* + * OOM here is quite inconvenient to propagate, since dealing with it + * would require a check for failure in the fast path. Instead, punt + * by using arenas[0]. In practice, this is an extremely unlikely + * failure. + */ + _malloc_message(_getprogname(), + ": (malloc) Error initializing arena\n", "", ""); + if (opt_abort) + abort(); + + return (arenas[0]); +} + +#ifdef MALLOC_MAG +static mag_t * +mag_create(arena_t *arena, size_t binind) +{ + mag_t *ret; + + if (sizeof(mag_t) + (sizeof(void *) * (max_rounds - 1)) <= + bin_maxclass) { + ret = arena_malloc_small(arena, sizeof(mag_t) + (sizeof(void *) + * (max_rounds - 1)), false); + } else { + ret = imalloc(sizeof(mag_t) + (sizeof(void *) * (max_rounds - + 1))); + } + if (ret == NULL) + return (NULL); + ret->binind = binind; + ret->nrounds = 0; + + return (ret); +} + +static void +mag_destroy(mag_t *mag) +{ + arena_t *arena; + arena_chunk_t *chunk; + size_t pageind; + arena_chunk_map_t *mapelm; + + chunk = CHUNK_ADDR2BASE(mag); + arena = chunk->arena; + pageind = (((uintptr_t)mag - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + + assert(mag->nrounds == 0); + if (sizeof(mag_t) + (sizeof(void *) * (max_rounds - 1)) <= + bin_maxclass) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, mag, mapelm); + malloc_spin_unlock(&arena->lock); + } else + idalloc(mag); +} + +static mag_rack_t * +mag_rack_create(arena_t *arena) +{ + + assert(sizeof(mag_rack_t) + (sizeof(bin_mags_t *) * (nbins - 1)) <= + bin_maxclass); + return (arena_malloc_small(arena, sizeof(mag_rack_t) + + (sizeof(bin_mags_t) * (nbins - 1)), true)); +} + +static void +mag_rack_destroy(mag_rack_t *rack) +{ + arena_t *arena; + arena_chunk_t *chunk; + bin_mags_t *bin_mags; + size_t i, pageind; + arena_chunk_map_t *mapelm; + + for (i = 0; i < nbins; i++) { + bin_mags = &rack->bin_mags[i]; + if (bin_mags->curmag != NULL) { + assert(bin_mags->curmag->binind == i); + mag_unload(bin_mags->curmag); + mag_destroy(bin_mags->curmag); + } + if (bin_mags->sparemag != NULL) { + assert(bin_mags->sparemag->binind == i); + mag_unload(bin_mags->sparemag); + mag_destroy(bin_mags->sparemag); + } + } + + chunk = CHUNK_ADDR2BASE(rack); + arena = chunk->arena; + pageind = (((uintptr_t)rack - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, rack, mapelm); + malloc_spin_unlock(&arena->lock); +} +#endif + +/* + * End arena. + */ +/******************************************************************************/ +/* + * Begin general internal functions. + */ + +static void * +huge_malloc(size_t size, bool zero) +{ + void *ret; + size_t csize; + extent_node_t *node; + + /* Allocate one or more contiguous chunks for this request. */ + + csize = CHUNK_CEILING(size); + if (csize == 0) { + /* size is large enough to cause size_t wrap-around. */ + return (NULL); + } + + /* Allocate an extent node with which to track the chunk. */ + node = base_node_alloc(); + if (node == NULL) + return (NULL); + + ret = chunk_alloc(csize, zero); + if (ret == NULL) { + base_node_dealloc(node); + return (NULL); + } + + /* Insert node into huge. */ + node->addr = ret; + node->size = csize; + + malloc_mutex_lock(&huge_mtx); + extent_tree_ad_insert(&huge, node); +#ifdef MALLOC_STATS + huge_nmalloc++; + huge_allocated += csize; +#endif + malloc_mutex_unlock(&huge_mtx); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, csize); + else if (opt_zero) + memset(ret, 0, csize); + } + + return (ret); +} + +/* Only handles large allocations that require more than chunk alignment. */ +static void * +huge_palloc(size_t alignment, size_t size) +{ + void *ret; + size_t alloc_size, chunk_size, offset; + extent_node_t *node; + + /* + * This allocation requires alignment that is even larger than chunk + * alignment. This means that huge_malloc() isn't good enough. + * + * Allocate almost twice as many chunks as are demanded by the size or + * alignment, in order to assure the alignment can be achieved, then + * unmap leading and trailing chunks. + */ + assert(alignment >= chunksize); + + chunk_size = CHUNK_CEILING(size); + + if (size >= alignment) + alloc_size = chunk_size + alignment - chunksize; + else + alloc_size = (alignment << 1) - chunksize; + + /* Allocate an extent node with which to track the chunk. */ + node = base_node_alloc(); + if (node == NULL) + return (NULL); + + ret = chunk_alloc(alloc_size, false); + if (ret == NULL) { + base_node_dealloc(node); + return (NULL); + } + + offset = (uintptr_t)ret & (alignment - 1); + assert((offset & chunksize_mask) == 0); + assert(offset < alloc_size); + if (offset == 0) { + /* Trim trailing space. */ + chunk_dealloc((void *)((uintptr_t)ret + chunk_size), alloc_size + - chunk_size); + } else { + size_t trailsize; + + /* Trim leading space. */ + chunk_dealloc(ret, alignment - offset); + + ret = (void *)((uintptr_t)ret + (alignment - offset)); + + trailsize = alloc_size - (alignment - offset) - chunk_size; + if (trailsize != 0) { + /* Trim trailing space. */ + assert(trailsize < alloc_size); + chunk_dealloc((void *)((uintptr_t)ret + chunk_size), + trailsize); + } + } + + /* Insert node into huge. */ + node->addr = ret; + node->size = chunk_size; + + malloc_mutex_lock(&huge_mtx); + extent_tree_ad_insert(&huge, node); +#ifdef MALLOC_STATS + huge_nmalloc++; + huge_allocated += chunk_size; +#endif + malloc_mutex_unlock(&huge_mtx); + + if (opt_junk) + memset(ret, 0xa5, chunk_size); + else if (opt_zero) + memset(ret, 0, chunk_size); + + return (ret); +} + +static void * +huge_ralloc(void *ptr, size_t size, size_t oldsize) +{ + void *ret; + size_t copysize; + + /* Avoid moving the allocation if the size class would not change. */ + if (oldsize > arena_maxclass && + CHUNK_CEILING(size) == CHUNK_CEILING(oldsize)) { + if (opt_junk && size < oldsize) { + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize + - size); + } else if (opt_zero && size > oldsize) { + memset((void *)((uintptr_t)ptr + oldsize), 0, size + - oldsize); + } + return (ptr); + } + + /* + * If we get here, then size and oldsize are different enough that we + * need to use a different size class. In that case, fall back to + * allocating new space and copying. + */ + ret = huge_malloc(size, false); + if (ret == NULL) + return (NULL); + + copysize = (size < oldsize) ? size : oldsize; + memcpy(ret, ptr, copysize); + idalloc(ptr); + return (ret); +} + +static void +huge_dalloc(void *ptr) +{ + extent_node_t *node, key; + + malloc_mutex_lock(&huge_mtx); + + /* Extract from tree of huge allocations. */ + key.addr = ptr; + node = extent_tree_ad_search(&huge, &key); + assert(node != NULL); + assert(node->addr == ptr); + extent_tree_ad_remove(&huge, node); + +#ifdef MALLOC_STATS + huge_ndalloc++; + huge_allocated -= node->size; +#endif + + malloc_mutex_unlock(&huge_mtx); + + /* Unmap chunk. */ +#ifdef MALLOC_DSS + if (opt_dss && opt_junk) + memset(node->addr, 0x5a, node->size); +#endif + chunk_dealloc(node->addr, node->size); + + base_node_dealloc(node); +} + +static void +malloc_print_stats(void) +{ + + if (opt_print_stats) { + char s[UMAX2S_BUFSIZE]; + _malloc_message("___ Begin malloc statistics ___\n", "", "", + ""); + _malloc_message("Assertions ", +#ifdef NDEBUG + "disabled", +#else + "enabled", +#endif + "\n", ""); + _malloc_message("Boolean MALLOC_OPTIONS: ", + opt_abort ? "A" : "a", "", ""); +#ifdef MALLOC_DSS + _malloc_message(opt_dss ? "D" : "d", "", "", ""); +#endif +#ifdef MALLOC_MAG + _malloc_message(opt_mag ? "G" : "g", "", "", ""); +#endif + _malloc_message(opt_junk ? "J" : "j", "", "", ""); +#ifdef MALLOC_DSS + _malloc_message(opt_mmap ? "M" : "m", "", "", ""); +#endif + _malloc_message(opt_utrace ? "PU" : "Pu", + opt_sysv ? "V" : "v", + opt_xmalloc ? "X" : "x", + opt_zero ? "Z\n" : "z\n"); + + _malloc_message("CPUs: ", umax2s(ncpus, s), "\n", ""); + _malloc_message("Max arenas: ", umax2s(narenas, s), "\n", ""); +#ifdef MALLOC_BALANCE + _malloc_message("Arena balance threshold: ", + umax2s(opt_balance_threshold, s), "\n", ""); +#endif + _malloc_message("Pointer size: ", umax2s(sizeof(void *), s), + "\n", ""); + _malloc_message("Quantum size: ", umax2s(QUANTUM, s), "\n", ""); + _malloc_message("Cacheline size (assumed): ", umax2s(CACHELINE, + s), "\n", ""); +#ifdef MALLOC_TINY + _malloc_message("Tiny 2^n-spaced sizes: [", umax2s((1U << + TINY_MIN_2POW), s), "..", ""); + _malloc_message(umax2s((qspace_min >> 1), s), "]\n", "", ""); +#endif + _malloc_message("Quantum-spaced sizes: [", umax2s(qspace_min, + s), "..", ""); + _malloc_message(umax2s(qspace_max, s), "]\n", "", ""); + _malloc_message("Cacheline-spaced sizes: [", umax2s(cspace_min, + s), "..", ""); + _malloc_message(umax2s(cspace_max, s), "]\n", "", ""); + _malloc_message("Subpage-spaced sizes: [", umax2s(sspace_min, + s), "..", ""); + _malloc_message(umax2s(sspace_max, s), "]\n", "", ""); +#ifdef MALLOC_MAG + _malloc_message("Rounds per magazine: ", umax2s(max_rounds, s), + "\n", ""); +#endif + _malloc_message("Max dirty pages per arena: ", + umax2s(opt_dirty_max, s), "\n", ""); + + _malloc_message("Chunk size: ", umax2s(chunksize, s), "", ""); + _malloc_message(" (2^", umax2s(opt_chunk_2pow, s), ")\n", ""); + +#ifdef MALLOC_STATS + { + size_t allocated, mapped; +#ifdef MALLOC_BALANCE + uint64_t nbalance = 0; +#endif + unsigned i; + arena_t *arena; + + /* Calculate and print allocated/mapped stats. */ + + /* arenas. */ + for (i = 0, allocated = 0; i < narenas; i++) { + if (arenas[i] != NULL) { + malloc_spin_lock(&arenas[i]->lock); + allocated += + arenas[i]->stats.allocated_small; + allocated += + arenas[i]->stats.allocated_large; +#ifdef MALLOC_BALANCE + nbalance += arenas[i]->stats.nbalance; +#endif + malloc_spin_unlock(&arenas[i]->lock); + } + } + + /* huge/base. */ + malloc_mutex_lock(&huge_mtx); + allocated += huge_allocated; + mapped = stats_chunks.curchunks * chunksize; + malloc_mutex_unlock(&huge_mtx); + + malloc_mutex_lock(&base_mtx); + mapped += base_mapped; + malloc_mutex_unlock(&base_mtx); + + malloc_printf("Allocated: %zu, mapped: %zu\n", + allocated, mapped); + +#ifdef MALLOC_BALANCE + malloc_printf("Arena balance reassignments: %llu\n", + nbalance); +#endif + + /* Print chunk stats. */ + { + chunk_stats_t chunks_stats; + + malloc_mutex_lock(&huge_mtx); + chunks_stats = stats_chunks; + malloc_mutex_unlock(&huge_mtx); + + malloc_printf("chunks: nchunks " + "highchunks curchunks\n"); + malloc_printf(" %13llu%13lu%13lu\n", + chunks_stats.nchunks, + chunks_stats.highchunks, + chunks_stats.curchunks); + } + + /* Print chunk stats. */ + malloc_printf( + "huge: nmalloc ndalloc allocated\n"); + malloc_printf(" %12llu %12llu %12zu\n", + huge_nmalloc, huge_ndalloc, huge_allocated); + + /* Print stats for each arena. */ + for (i = 0; i < narenas; i++) { + arena = arenas[i]; + if (arena != NULL) { + malloc_printf( + "\narenas[%u]:\n", i); + malloc_spin_lock(&arena->lock); + stats_print(arena); + malloc_spin_unlock(&arena->lock); + } + } + } +#endif /* #ifdef MALLOC_STATS */ + _malloc_message("--- End malloc statistics ---\n", "", "", ""); + } +} + +#ifdef MALLOC_DEBUG +static void +size2bin_validate(void) +{ + size_t i, size, binind; + + assert(size2bin[0] == 0xffU); + i = 1; +# ifdef MALLOC_TINY + /* Tiny. */ + for (; i < (1U << TINY_MIN_2POW); i++) { + size = pow2_ceil(1U << TINY_MIN_2POW); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + assert(size2bin[i] == binind); + } + for (; i < qspace_min; i++) { + size = pow2_ceil(i); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + assert(size2bin[i] == binind); + } +# endif + /* Quantum-spaced. */ + for (; i <= qspace_max; i++) { + size = QUANTUM_CEILING(i); + binind = ntbins + (size >> QUANTUM_2POW) - 1; + assert(size2bin[i] == binind); + } + /* Cacheline-spaced. */ + for (; i <= cspace_max; i++) { + size = CACHELINE_CEILING(i); + binind = ntbins + nqbins + ((size - cspace_min) >> + CACHELINE_2POW); + assert(size2bin[i] == binind); + } + /* Sub-page. */ + for (; i <= sspace_max; i++) { + size = SUBPAGE_CEILING(i); + binind = ntbins + nqbins + ncbins + ((size - sspace_min) + >> SUBPAGE_2POW); + assert(size2bin[i] == binind); + } +} +#endif + +static bool +size2bin_init(void) +{ + + if (opt_qspace_max_2pow != QSPACE_MAX_2POW_DEFAULT + || opt_cspace_max_2pow != CSPACE_MAX_2POW_DEFAULT) + return (size2bin_init_hard()); + + size2bin = const_size2bin; +#ifdef MALLOC_DEBUG + assert(sizeof(const_size2bin) == bin_maxclass + 1); + size2bin_validate(); +#endif + return (false); +} + +static bool +size2bin_init_hard(void) +{ + size_t i, size, binind; + uint8_t *custom_size2bin; + + assert(opt_qspace_max_2pow != QSPACE_MAX_2POW_DEFAULT + || opt_cspace_max_2pow != CSPACE_MAX_2POW_DEFAULT); + + custom_size2bin = (uint8_t *)base_alloc(bin_maxclass + 1); + if (custom_size2bin == NULL) + return (true); + + custom_size2bin[0] = 0xffU; + i = 1; +#ifdef MALLOC_TINY + /* Tiny. */ + for (; i < (1U << TINY_MIN_2POW); i++) { + size = pow2_ceil(1U << TINY_MIN_2POW); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + custom_size2bin[i] = binind; + } + for (; i < qspace_min; i++) { + size = pow2_ceil(i); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + custom_size2bin[i] = binind; + } +#endif + /* Quantum-spaced. */ + for (; i <= qspace_max; i++) { + size = QUANTUM_CEILING(i); + binind = ntbins + (size >> QUANTUM_2POW) - 1; + custom_size2bin[i] = binind; + } + /* Cacheline-spaced. */ + for (; i <= cspace_max; i++) { + size = CACHELINE_CEILING(i); + binind = ntbins + nqbins + ((size - cspace_min) >> + CACHELINE_2POW); + custom_size2bin[i] = binind; + } + /* Sub-page. */ + for (; i <= sspace_max; i++) { + size = SUBPAGE_CEILING(i); + binind = ntbins + nqbins + ncbins + ((size - sspace_min) >> + SUBPAGE_2POW); + custom_size2bin[i] = binind; + } + + size2bin = custom_size2bin; +#ifdef MALLOC_DEBUG + size2bin_validate(); +#endif + return (false); +} + +static unsigned +malloc_ncpus(void) +{ + unsigned ret; + int fd, nread, column; + char buf[1]; + static const char matchstr[] = "processor\t:"; + + /* + * sysconf(3) would be the preferred method for determining the number + * of CPUs, but it uses malloc internally, which causes untennable + * recursion during malloc initialization. + */ + fd = open("/proc/cpuinfo", O_RDONLY); + if (fd == -1) + return (1); /* Error. */ + /* + * Count the number of occurrences of matchstr at the beginnings of + * lines. This treats hyperthreaded CPUs as multiple processors. + */ + column = 0; + ret = 0; + while (true) { + nread = read(fd, &buf, sizeof(buf)); + if (nread <= 0) + break; /* EOF or error. */ + + if (buf[0] == '\n') + column = 0; + else if (column != -1) { + if (buf[0] == matchstr[column]) { + column++; + if (column == sizeof(matchstr) - 1) { + column = -1; + ret++; + } + } else + column = -1; + } + } + if (ret == 0) + ret = 1; /* Something went wrong in the parser. */ + close(fd); + + return (ret); +} +/* + * FreeBSD's pthreads implementation calls malloc(3), so the malloc + * implementation has to take pains to avoid infinite recursion during + * initialization. + */ +static inline bool +malloc_init(void) +{ + + if (malloc_initialized == false) + return (malloc_init_hard()); + + return (false); +} + +static bool +malloc_init_hard(void) +{ + unsigned i; + int linklen; + char buf[PATH_MAX + 1]; + const char *opts; + + malloc_mutex_lock(&init_lock); + if (malloc_initialized) { + /* + * Another thread initialized the allocator before this one + * acquired init_lock. + */ + malloc_mutex_unlock(&init_lock); + return (false); + } + + /* Get number of CPUs. */ + ncpus = malloc_ncpus(); + + /* Get page size. */ + { + long result; + + result = sysconf(_SC_PAGESIZE); + assert(result != -1); + pagesize = (unsigned)result; + + /* + * We assume that pagesize is a power of 2 when calculating + * pagesize_mask and pagesize_2pow. + */ + assert(((result - 1) & result) == 0); + pagesize_mask = result - 1; + pagesize_2pow = ffs((int)result) - 1; + } + + for (i = 0; i < 3; i++) { + unsigned j; + + /* Get runtime configuration. */ + switch (i) { + case 0: + if ((linklen = readlink("/etc/malloc.conf", buf, + sizeof(buf) - 1)) != -1) { + /* + * Use the contents of the "/etc/malloc.conf" + * symbolic link's name. + */ + buf[linklen] = '\0'; + opts = buf; + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + case 1: + if (issetugid() == 0 && (opts = + getenv("MALLOC_OPTIONS")) != NULL) { + /* + * Do nothing; opts is already initialized to + * the value of the MALLOC_OPTIONS environment + * variable. + */ + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + case 2: + if (_malloc_options != NULL) { + /* + * Use options that were compiled into the + * program. + */ + opts = _malloc_options; + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + default: + /* NOTREACHED */ + assert(false); + } + + for (j = 0; opts[j] != '\0'; j++) { + unsigned k, nreps; + bool nseen; + + /* Parse repetition count, if any. */ + for (nreps = 0, nseen = false;; j++, nseen = true) { + switch (opts[j]) { + case '0': case '1': case '2': case '3': + case '4': case '5': case '6': case '7': + case '8': case '9': + nreps *= 10; + nreps += opts[j] - '0'; + break; + default: + goto MALLOC_OUT; + } + } +MALLOC_OUT: + if (nseen == false) + nreps = 1; + + for (k = 0; k < nreps; k++) { + switch (opts[j]) { + case 'a': + opt_abort = false; + break; + case 'A': + opt_abort = true; + break; + case 'b': +#ifdef MALLOC_BALANCE + opt_balance_threshold >>= 1; +#endif + break; + case 'B': +#ifdef MALLOC_BALANCE + if (opt_balance_threshold == 0) + opt_balance_threshold = 1; + else if ((opt_balance_threshold << 1) + > opt_balance_threshold) + opt_balance_threshold <<= 1; +#endif + break; + case 'c': + if (opt_cspace_max_2pow - 1 > + opt_qspace_max_2pow && + opt_cspace_max_2pow > + CACHELINE_2POW) + opt_cspace_max_2pow--; + break; + case 'C': + if (opt_cspace_max_2pow < pagesize_2pow + - 1) + opt_cspace_max_2pow++; + break; + case 'd': +#ifdef MALLOC_DSS + opt_dss = false; +#endif + break; + case 'D': +#ifdef MALLOC_DSS + opt_dss = true; +#endif + break; + case 'f': + opt_dirty_max >>= 1; + break; + case 'F': + if (opt_dirty_max == 0) + opt_dirty_max = 1; + else if ((opt_dirty_max << 1) != 0) + opt_dirty_max <<= 1; + break; +#ifdef MALLOC_MAG + case 'g': + opt_mag = false; + break; + case 'G': + opt_mag = true; + break; +#endif + case 'j': + opt_junk = false; + break; + case 'J': + opt_junk = true; + break; + case 'k': + /* + * Chunks always require at least one + * header page, so chunks can never be + * smaller than two pages. + */ + if (opt_chunk_2pow > pagesize_2pow + 1) + opt_chunk_2pow--; + break; + case 'K': + if (opt_chunk_2pow + 1 < + (sizeof(size_t) << 3)) + opt_chunk_2pow++; + break; + case 'm': +#ifdef MALLOC_DSS + opt_mmap = false; +#endif + break; + case 'M': +#ifdef MALLOC_DSS + opt_mmap = true; +#endif + break; + case 'n': + opt_narenas_lshift--; + break; + case 'N': + opt_narenas_lshift++; + break; + case 'p': + opt_print_stats = false; + break; + case 'P': + opt_print_stats = true; + break; + case 'q': + if (opt_qspace_max_2pow > QUANTUM_2POW) + opt_qspace_max_2pow--; + break; + case 'Q': + if (opt_qspace_max_2pow + 1 < + opt_cspace_max_2pow) + opt_qspace_max_2pow++; + break; +#ifdef MALLOC_MAG + case 'R': + if (opt_mag_size_2pow + 1 < (8U << + SIZEOF_PTR_2POW)) + opt_mag_size_2pow++; + break; + case 'r': + /* + * Make sure there's always at least + * one round per magazine. + */ + if ((1U << (opt_mag_size_2pow-1)) >= + sizeof(mag_t)) + opt_mag_size_2pow--; + break; +#endif + case 'u': + opt_utrace = false; + break; + case 'U': + opt_utrace = true; + break; + case 'v': + opt_sysv = false; + break; + case 'V': + opt_sysv = true; + break; + case 'x': + opt_xmalloc = false; + break; + case 'X': + opt_xmalloc = true; + break; + case 'z': + opt_zero = false; + break; + case 'Z': + opt_zero = true; + break; + default: { + char cbuf[2]; + + cbuf[0] = opts[j]; + cbuf[1] = '\0'; + _malloc_message(_getprogname(), + ": (malloc) Unsupported character " + "in malloc options: '", cbuf, + "'\n"); + } + } + } + } + } + +#ifdef MALLOC_DSS + /* Make sure that there is some method for acquiring memory. */ + if (opt_dss == false && opt_mmap == false) + opt_mmap = true; +#endif + + /* Take care to call atexit() only once. */ + if (opt_print_stats) { + /* Print statistics at exit. */ + atexit(malloc_print_stats); + } + + /* Register fork handlers. */ + pthread_atfork(_malloc_prefork, _malloc_postfork, _malloc_postfork); + +#ifdef MALLOC_MAG + /* + * Calculate the actual number of rounds per magazine, taking into + * account header overhead. + */ + max_rounds = (1LLU << (opt_mag_size_2pow - SIZEOF_PTR_2POW)) - + (sizeof(mag_t) >> SIZEOF_PTR_2POW) + 1; +#endif + + /* Set variables according to the value of opt_[qc]space_max_2pow. */ + qspace_max = (1U << opt_qspace_max_2pow); + cspace_min = CACHELINE_CEILING(qspace_max); + if (cspace_min == qspace_max) + cspace_min += CACHELINE; + cspace_max = (1U << opt_cspace_max_2pow); + sspace_min = SUBPAGE_CEILING(cspace_max); + if (sspace_min == cspace_max) + sspace_min += SUBPAGE; + assert(sspace_min < pagesize); + sspace_max = pagesize - SUBPAGE; + +#ifdef MALLOC_TINY + assert(QUANTUM_2POW >= TINY_MIN_2POW); +#endif + assert(ntbins <= QUANTUM_2POW); + nqbins = qspace_max >> QUANTUM_2POW; + ncbins = ((cspace_max - cspace_min) >> CACHELINE_2POW) + 1; + nsbins = ((sspace_max - sspace_min) >> SUBPAGE_2POW) + 1; + nbins = ntbins + nqbins + ncbins + nsbins; + + if (size2bin_init()) { + malloc_mutex_unlock(&init_lock); + return (true); + } + + /* Set variables according to the value of opt_chunk_2pow. */ + chunksize = (1LU << opt_chunk_2pow); + chunksize_mask = chunksize - 1; + chunk_npages = (chunksize >> pagesize_2pow); + { + size_t header_size; + + /* + * Compute the header size such that it is large enough to + * contain the page map. + */ + header_size = sizeof(arena_chunk_t) + + (sizeof(arena_chunk_map_t) * (chunk_npages - 1)); + arena_chunk_header_npages = (header_size >> pagesize_2pow) + + ((header_size & pagesize_mask) != 0); + } + arena_maxclass = chunksize - (arena_chunk_header_npages << + pagesize_2pow); + + UTRACE(0, 0, 0); + +#ifdef MALLOC_STATS + memset(&stats_chunks, 0, sizeof(chunk_stats_t)); +#endif + + /* Various sanity checks that regard configuration. */ + assert(chunksize >= pagesize); + + /* Initialize chunks data. */ + if (malloc_mutex_init(&huge_mtx)) { + malloc_mutex_unlock(&init_lock); + return (true); + } + extent_tree_ad_new(&huge); +#ifdef MALLOC_DSS + if (malloc_mutex_init(&dss_mtx)) { + malloc_mutex_unlock(&init_lock); + return (true); + } + dss_base = sbrk(0); + dss_prev = dss_base; + dss_max = dss_base; + extent_tree_szad_new(&dss_chunks_szad); + extent_tree_ad_new(&dss_chunks_ad); +#endif +#ifdef MALLOC_STATS + huge_nmalloc = 0; + huge_ndalloc = 0; + huge_allocated = 0; +#endif + + /* Initialize base allocation data structures. */ +#ifdef MALLOC_STATS + base_mapped = 0; +#endif +#ifdef MALLOC_DSS + /* + * Allocate a base chunk here, since it doesn't actually have to be + * chunk-aligned. Doing this before allocating any other chunks allows + * the use of space that would otherwise be wasted. + */ + if (opt_dss) + base_pages_alloc(0); +#endif + base_nodes = NULL; + if (malloc_mutex_init(&base_mtx)) { + malloc_mutex_unlock(&init_lock); + return (true); + } + + if (ncpus > 1) { + /* + * For SMP systems, create twice as many arenas as there are + * CPUs by default. + */ + opt_narenas_lshift++; + } + + /* Determine how many arenas to use. */ + narenas = ncpus; + if (opt_narenas_lshift > 0) { + if ((narenas << opt_narenas_lshift) > narenas) + narenas <<= opt_narenas_lshift; + /* + * Make sure not to exceed the limits of what base_alloc() can + * handle. + */ + if (narenas * sizeof(arena_t *) > chunksize) + narenas = chunksize / sizeof(arena_t *); + } else if (opt_narenas_lshift < 0) { + if ((narenas >> -opt_narenas_lshift) < narenas) + narenas >>= -opt_narenas_lshift; + /* Make sure there is at least one arena. */ + if (narenas == 0) + narenas = 1; + } +#ifdef MALLOC_BALANCE + assert(narenas != 0); + for (narenas_2pow = 0; + (narenas >> (narenas_2pow + 1)) != 0; + narenas_2pow++); +#endif + +#ifdef NO_TLS + if (narenas > 1) { + static const unsigned primes[] = {1, 3, 5, 7, 11, 13, 17, 19, + 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, + 89, 97, 101, 103, 107, 109, 113, 127, 131, 137, 139, 149, + 151, 157, 163, 167, 173, 179, 181, 191, 193, 197, 199, 211, + 223, 227, 229, 233, 239, 241, 251, 257, 263}; + unsigned nprimes, parenas; + + /* + * Pick a prime number of hash arenas that is more than narenas + * so that direct hashing of pthread_self() pointers tends to + * spread allocations evenly among the arenas. + */ + assert((narenas & 1) == 0); /* narenas must be even. */ + nprimes = (sizeof(primes) >> SIZEOF_INT_2POW); + parenas = primes[nprimes - 1]; /* In case not enough primes. */ + for (i = 1; i < nprimes; i++) { + if (primes[i] > narenas) { + parenas = primes[i]; + break; + } + } + narenas = parenas; + } +#endif + +#ifndef NO_TLS +# ifndef MALLOC_BALANCE + next_arena = 0; +# endif +#endif + + /* Allocate and initialize arenas. */ + arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas); + if (arenas == NULL) { + malloc_mutex_unlock(&init_lock); + return (true); + } + /* + * Zero the array. In practice, this should always be pre-zeroed, + * since it was just mmap()ed, but let's be sure. + */ + memset(arenas, 0, sizeof(arena_t *) * narenas); + + /* + * Initialize one arena here. The rest are lazily created in + * choose_arena_hard(). + */ + arenas_extend(0); + if (arenas[0] == NULL) { + malloc_mutex_unlock(&init_lock); + return (true); + } +#ifndef NO_TLS + /* + * Assign the initial arena to the initial thread, in order to avoid + * spurious creation of an extra arena if the application switches to + * threaded mode. + */ + arenas_map = arenas[0]; +#endif + /* + * Seed here for the initial thread, since choose_arena_hard() is only + * called for other threads. The seed value doesn't really matter. + */ +#ifdef MALLOC_BALANCE + SPRN(balance, 42); +#endif + + malloc_spin_init(&arenas_lock); + + malloc_initialized = true; + malloc_mutex_unlock(&init_lock); + return (false); +} + +/* + * End general internal functions. + */ +/******************************************************************************/ +/* + * Begin malloc(3)-compatible functions. + */ + +void * +malloc(size_t size) +{ + void *ret; + + if (malloc_init()) { + ret = NULL; + goto RETURN; + } + + if (size == 0) { + if (opt_sysv == false) + size = 1; + else { + ret = NULL; + goto RETURN; + } + } + + ret = imalloc(size); + +RETURN: + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in malloc(): out of memory\n", "", + ""); + abort(); + } + errno = ENOMEM; + } + + UTRACE(0, size, ret); + return (ret); +} + +int +posix_memalign(void **memptr, size_t alignment, size_t size) +{ + int ret; + void *result; + + if (malloc_init()) + result = NULL; + else { + /* Make sure that alignment is a large enough power of 2. */ + if (((alignment - 1) & alignment) != 0 + || alignment < sizeof(void *)) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in posix_memalign(): " + "invalid alignment\n", "", ""); + abort(); + } + result = NULL; + ret = EINVAL; + goto RETURN; + } + + result = ipalloc(alignment, size); + } + + if (result == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in posix_memalign(): out of memory\n", + "", ""); + abort(); + } + ret = ENOMEM; + goto RETURN; + } + + *memptr = result; + ret = 0; + +RETURN: + UTRACE(0, size, result); + return (ret); +} + +void * +calloc(size_t num, size_t size) +{ + void *ret; + size_t num_size; + + if (malloc_init()) { + num_size = 0; + ret = NULL; + goto RETURN; + } + + num_size = num * size; + if (num_size == 0) { + if ((opt_sysv == false) && ((num == 0) || (size == 0))) + num_size = 1; + else { + ret = NULL; + goto RETURN; + } + /* + * Try to avoid division here. We know that it isn't possible to + * overflow during multiplication if neither operand uses any of the + * most significant half of the bits in a size_t. + */ + } else if (((num | size) & (SIZE_T_MAX << (sizeof(size_t) << 2))) + && (num_size / size != num)) { + /* size_t overflow. */ + ret = NULL; + goto RETURN; + } + + ret = icalloc(num_size); + +RETURN: + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in calloc(): out of memory\n", "", + ""); + abort(); + } + errno = ENOMEM; + } + + UTRACE(0, num_size, ret); + return (ret); +} + +void * +realloc(void *ptr, size_t size) +{ + void *ret; + + if (size == 0) { + if (opt_sysv == false) + size = 1; + else { + if (ptr != NULL) + idalloc(ptr); + ret = NULL; + goto RETURN; + } + } + + if (ptr != NULL) { + assert(malloc_initialized); + + ret = iralloc(ptr, size); + + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in realloc(): out of " + "memory\n", "", ""); + abort(); + } + errno = ENOMEM; + } + } else { + if (malloc_init()) + ret = NULL; + else + ret = imalloc(size); + + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in realloc(): out of " + "memory\n", "", ""); + abort(); + } + errno = ENOMEM; + } + } + +RETURN: + UTRACE(ptr, size, ret); + return (ret); +} + +void +free(void *ptr) +{ + + UTRACE(ptr, 0, 0); + if (ptr != NULL) { + assert(malloc_initialized); + + idalloc(ptr); + } +} + +/* + * End malloc(3)-compatible functions. + */ +/******************************************************************************/ +/* + * Begin non-standard functions. + */ + +size_t +malloc_usable_size(const void *ptr) +{ + + assert(ptr != NULL); + + return (isalloc(ptr)); +} + +/* + * End non-standard functions. + */ +/******************************************************************************/ +/* + * Begin library-private functions. + */ + +/******************************************************************************/ +/* + * Begin thread cache. + */ + +/* + * We provide an unpublished interface in order to receive notifications from + * the pthreads library whenever a thread exits. This allows us to clean up + * thread caches. + */ +void +_malloc_thread_cleanup(void) +{ + +#ifdef MALLOC_MAG + if (mag_rack != NULL) { + assert(mag_rack != (void *)-1); + mag_rack_destroy(mag_rack); +#ifdef MALLOC_DEBUG + mag_rack = (void *)-1; +#endif + } +#endif +} + +/* + * The following functions are used by threading libraries for protection of + * malloc during fork(). These functions are only called if the program is + * running in threaded mode, so there is no need to check whether the program + * is threaded here. + */ + +void +_malloc_prefork(void) +{ + unsigned i; + + /* Acquire all mutexes in a safe order. */ + + malloc_spin_lock(&arenas_lock); + for (i = 0; i < narenas; i++) { + if (arenas[i] != NULL) + malloc_spin_lock(&arenas[i]->lock); + } + malloc_spin_unlock(&arenas_lock); + + malloc_mutex_lock(&base_mtx); + + malloc_mutex_lock(&huge_mtx); + +#ifdef MALLOC_DSS + malloc_mutex_lock(&dss_mtx); +#endif +} + +void +_malloc_postfork(void) +{ + unsigned i; + + /* Release all mutexes, now that fork() has completed. */ + +#ifdef MALLOC_DSS + malloc_mutex_unlock(&dss_mtx); +#endif + + malloc_mutex_unlock(&huge_mtx); + + malloc_mutex_unlock(&base_mtx); + + malloc_spin_lock(&arenas_lock); + for (i = 0; i < narenas; i++) { + if (arenas[i] != NULL) + malloc_spin_unlock(&arenas[i]->lock); + } + malloc_spin_unlock(&arenas_lock); +} + +/* + * End library-private functions. + */ +/******************************************************************************/ diff --git a/varnish-cache/lib/libjemalloc/malloc.3 b/varnish-cache/lib/libjemalloc/malloc.3 new file mode 100644 index 00000000..67a52fbc --- /dev/null +++ b/varnish-cache/lib/libjemalloc/malloc.3 @@ -0,0 +1,584 @@ +.\" Copyright (c) 1980, 1991, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" This code is derived from software contributed to Berkeley by +.\" the American National Standards Committee X3, on Information +.\" Processing Systems. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)malloc.3 8.1 (Berkeley) 6/4/93 +.\" $FreeBSD: head/lib/libc/stdlib/malloc.3 182225 2008-08-27 02:00:53Z jasone $ +.\" +.Dd August 26, 2008 +.Dt MALLOC 3 +.Os +.Sh NAME +.Nm malloc , calloc , realloc , free , reallocf , malloc_usable_size +.Nd general purpose memory allocation functions +.Sh LIBRARY +.Lb libc +.Sh SYNOPSIS +.In stdlib.h +.Ft void * +.Fn malloc "size_t size" +.Ft void * +.Fn calloc "size_t number" "size_t size" +.Ft void * +.Fn realloc "void *ptr" "size_t size" +.Ft void * +.Fn reallocf "void *ptr" "size_t size" +.Ft void +.Fn free "void *ptr" +.Ft const char * +.Va _malloc_options ; +.Ft void +.Fo \*(lp*_malloc_message\*(rp +.Fa "const char *p1" "const char *p2" "const char *p3" "const char *p4" +.Fc +.In malloc_np.h +.Ft size_t +.Fn malloc_usable_size "const void *ptr" +.Sh DESCRIPTION +The +.Fn malloc +function allocates +.Fa size +bytes of uninitialized memory. +The allocated space is suitably aligned (after possible pointer coercion) +for storage of any type of object. +.Pp +The +.Fn calloc +function allocates space for +.Fa number +objects, +each +.Fa size +bytes in length. +The result is identical to calling +.Fn malloc +with an argument of +.Dq "number * size" , +with the exception that the allocated memory is explicitly initialized +to zero bytes. +.Pp +The +.Fn realloc +function changes the size of the previously allocated memory referenced by +.Fa ptr +to +.Fa size +bytes. +The contents of the memory are unchanged up to the lesser of the new and +old sizes. +If the new size is larger, +the contents of the newly allocated portion of the memory are undefined. +Upon success, the memory referenced by +.Fa ptr +is freed and a pointer to the newly allocated memory is returned. +Note that +.Fn realloc +and +.Fn reallocf +may move the memory allocation, resulting in a different return value than +.Fa ptr . +If +.Fa ptr +is +.Dv NULL , +the +.Fn realloc +function behaves identically to +.Fn malloc +for the specified size. +.Pp +The +.Fn reallocf +function is identical to the +.Fn realloc +function, except that it +will free the passed pointer when the requested memory cannot be allocated. +This is a +.Fx +specific API designed to ease the problems with traditional coding styles +for realloc causing memory leaks in libraries. +.Pp +The +.Fn free +function causes the allocated memory referenced by +.Fa ptr +to be made available for future allocations. +If +.Fa ptr +is +.Dv NULL , +no action occurs. +.Pp +The +.Fn malloc_usable_size +function returns the usable size of the allocation pointed to by +.Fa ptr . +The return value may be larger than the size that was requested during +allocation. +The +.Fn malloc_usable_size +function is not a mechanism for in-place +.Fn realloc ; +rather it is provided solely as a tool for introspection purposes. +Any discrepancy between the requested allocation size and the size reported by +.Fn malloc_usable_size +should not be depended on, since such behavior is entirely +implementation-dependent. +.Sh TUNING +Once, when the first call is made to one of these memory allocation +routines, various flags will be set or reset, which affects the +workings of this allocator implementation. +.Pp +The +.Dq name +of the file referenced by the symbolic link named +.Pa /etc/malloc.conf , +the value of the environment variable +.Ev MALLOC_OPTIONS , +and the string pointed to by the global variable +.Va _malloc_options +will be interpreted, in that order, from left to right as flags. +.Pp +Each flag is a single letter, optionally prefixed by a non-negative base 10 +integer repetition count. +For example, +.Dq 3N +is equivalent to +.Dq NNN . +Some flags control parameter magnitudes, where uppercase increases the +magnitude, and lowercase decreases the magnitude. +Other flags control boolean parameters, where uppercase indicates that a +behavior is set, or on, and lowercase means that a behavior is not set, or off. +.Bl -tag -width indent +.It A +All warnings (except for the warning about unknown +flags being set) become fatal. +The process will call +.Xr abort 3 +in these cases. +.It B +Double/halve the per-arena lock contention threshold at which a thread is +randomly re-assigned to an arena. +This dynamic load balancing tends to push threads away from highly contended +arenas, which avoids worst case contention scenarios in which threads +disproportionately utilize arenas. +However, due to the highly dynamic load that applications may place on the +allocator, it is impossible for the allocator to know in advance how sensitive +it should be to contention over arenas. +Therefore, some applications may benefit from increasing or decreasing this +threshold parameter. +This option is not available for some configurations (non-PIC). +.It C +Double/halve the size of the maximum size class that is a multiple of the +cacheline size (64). +Above this size, subpage spacing (256 bytes) is used for size classes. +The default value is 512 bytes. +.It D +Use +.Xr sbrk 2 +to acquire memory in the data storage segment (DSS). +This option is enabled by default. +See the +.Dq M +option for related information and interactions. +.It F +Double/halve the per-arena maximum number of dirty unused pages that are +allowed to accumulate before informing the kernel about at least half of those +pages via +.Xr madvise 2 . +This provides the kernel with sufficient information to recycle dirty pages if +physical memory becomes scarce and the pages remain unused. +The default is 512 pages per arena; +.Ev MALLOC_OPTIONS=10f +will prevent any dirty unused pages from accumulating. +.It G +When there are multiple threads, use thread-specific caching for objects that +are smaller than one page. +This option is enabled by default. +Thread-specific caching allows many allocations to be satisfied without +performing any thread synchronization, at the cost of increased memory use. +See the +.Dq R +option for related tuning information. +This option is not available for some configurations (non-PIC). +.It J +Each byte of new memory allocated by +.Fn malloc , +.Fn realloc +or +.Fn reallocf +will be initialized to 0xa5. +All memory returned by +.Fn free , +.Fn realloc +or +.Fn reallocf +will be initialized to 0x5a. +This is intended for debugging and will impact performance negatively. +.It K +Double/halve the virtual memory chunk size. +The default chunk size is 1 MB. +.It M +Use +.Xr mmap 2 +to acquire anonymously mapped memory. +This option is enabled by default. +If both the +.Dq D +and +.Dq M +options are enabled, the allocator prefers the DSS over anonymous mappings, +but allocation only fails if memory cannot be acquired via either method. +If neither option is enabled, then the +.Dq M +option is implicitly enabled in order to assure that there is a method for +acquiring memory. +.It N +Double/halve the number of arenas. +The default number of arenas is two times the number of CPUs, or one if there +is a single CPU. +.It P +Various statistics are printed at program exit via an +.Xr atexit 3 +function. +This has the potential to cause deadlock for a multi-threaded process that exits +while one or more threads are executing in the memory allocation functions. +Therefore, this option should only be used with care; it is primarily intended +as a performance tuning aid during application development. +.It Q +Double/halve the size of the maximum size class that is a multiple of the +quantum (8 or 16 bytes, depending on architecture). +Above this size, cacheline spacing is used for size classes. +The default value is 128 bytes. +.It R +Double/halve magazine size, which approximately doubles/halves the number of +rounds in each magazine. +Magazines are used by the thread-specific caching machinery to acquire and +release objects in bulk. +Increasing the magazine size decreases locking overhead, at the expense of +increased memory usage. +This option is not available for some configurations (non-PIC). +.It U +Generate +.Dq utrace +entries for +.Xr ktrace 1 , +for all operations. +Consult the source for details on this option. +.It V +Attempting to allocate zero bytes will return a +.Dv NULL +pointer instead of +a valid pointer. +(The default behavior is to make a minimal allocation and return a +pointer to it.) +This option is provided for System V compatibility. +This option is incompatible with the +.Dq X +option. +.It X +Rather than return failure for any allocation function, +display a diagnostic message on +.Dv stderr +and cause the program to drop +core (using +.Xr abort 3 ) . +This option should be set at compile time by including the following in +the source code: +.Bd -literal -offset indent +_malloc_options = "X"; +.Ed +.It Z +Each byte of new memory allocated by +.Fn malloc , +.Fn realloc +or +.Fn reallocf +will be initialized to 0. +Note that this initialization only happens once for each byte, so +.Fn realloc +and +.Fn reallocf +calls do not zero memory that was previously allocated. +This is intended for debugging and will impact performance negatively. +.El +.Pp +The +.Dq J +and +.Dq Z +options are intended for testing and debugging. +An application which changes its behavior when these options are used +is flawed. +.Sh IMPLEMENTATION NOTES +Traditionally, allocators have used +.Xr sbrk 2 +to obtain memory, which is suboptimal for several reasons, including race +conditions, increased fragmentation, and artificial limitations on maximum +usable memory. +This allocator uses both +.Xr sbrk 2 +and +.Xr mmap 2 +by default, but it can be configured at run time to use only one or the other. +If resource limits are not a primary concern, the preferred configuration is +.Ev MALLOC_OPTIONS=dM +or +.Ev MALLOC_OPTIONS=DM . +When so configured, the +.Ar datasize +resource limit has little practical effect for typical applications; use +.Ev MALLOC_OPTIONS=Dm +if that is a concern. +Regardless of allocator configuration, the +.Ar vmemoryuse +resource limit can be used to bound the total virtual memory used by a +process, as described in +.Xr limits 1 . +.Pp +This allocator uses multiple arenas in order to reduce lock contention for +threaded programs on multi-processor systems. +This works well with regard to threading scalability, but incurs some costs. +There is a small fixed per-arena overhead, and additionally, arenas manage +memory completely independently of each other, which means a small fixed +increase in overall memory fragmentation. +These overheads are not generally an issue, given the number of arenas normally +used. +Note that using substantially more arenas than the default is not likely to +improve performance, mainly due to reduced cache performance. +However, it may make sense to reduce the number of arenas if an application +does not make much use of the allocation functions. +.Pp +In addition to multiple arenas, this allocator supports thread-specific +caching for small objects (smaller than one page), in order to make it +possible to completely avoid synchronization for most small allocation requests. +Such caching allows very fast allocation in the common case, but it increases +memory usage and fragmentation, since a bounded number of objects can remain +allocated in each thread cache. +.Pp +Memory is conceptually broken into equal-sized chunks, where the chunk size is +a power of two that is greater than the page size. +Chunks are always aligned to multiples of the chunk size. +This alignment makes it possible to find metadata for user objects very +quickly. +.Pp +User objects are broken into three categories according to size: small, large, +and huge. +Small objects are smaller than one page. +Large objects are smaller than the chunk size. +Huge objects are a multiple of the chunk size. +Small and large objects are managed by arenas; huge objects are managed +separately in a single data structure that is shared by all threads. +Huge objects are used by applications infrequently enough that this single +data structure is not a scalability issue. +.Pp +Each chunk that is managed by an arena tracks its contents as runs of +contiguous pages (unused, backing a set of small objects, or backing one large +object). +The combination of chunk alignment and chunk page maps makes it possible to +determine all metadata regarding small and large allocations in constant time. +.Pp +Small objects are managed in groups by page runs. +Each run maintains a bitmap that tracks which regions are in use. +Allocation requests that are no more than half the quantum (8 or 16, depending +on architecture) are rounded up to the nearest power of two. +Allocation requests that are more than half the quantum, but no more than the +minimum cacheline-multiple size class (see the +.Dq Q +option) are rounded up to the nearest multiple of the quantum. +Allocation requests that are more than the minumum cacheline-multiple size +class, but no more than the minimum subpage-multiple size class (see the +.Dq C +option) are rounded up to the nearest multiple of the cacheline size (64). +Allocation requests that are more than the minimum subpage-multiple size class +are rounded up to the nearest multiple of the subpage size (256). +Allocation requests that are more than one page, but small enough to fit in +an arena-managed chunk (see the +.Dq K +option), are rounded up to the nearest run size. +Allocation requests that are too large to fit in an arena-managed chunk are +rounded up to the nearest multiple of the chunk size. +.Pp +Allocations are packed tightly together, which can be an issue for +multi-threaded applications. +If you need to assure that allocations do not suffer from cacheline sharing, +round your allocation requests up to the nearest multiple of the cacheline +size. +.Sh DEBUGGING MALLOC PROBLEMS +The first thing to do is to set the +.Dq A +option. +This option forces a coredump (if possible) at the first sign of trouble, +rather than the normal policy of trying to continue if at all possible. +.Pp +It is probably also a good idea to recompile the program with suitable +options and symbols for debugger support. +.Pp +If the program starts to give unusual results, coredump or generally behave +differently without emitting any of the messages mentioned in the next +section, it is likely because it depends on the storage being filled with +zero bytes. +Try running it with the +.Dq Z +option set; +if that improves the situation, this diagnosis has been confirmed. +If the program still misbehaves, +the likely problem is accessing memory outside the allocated area. +.Pp +Alternatively, if the symptoms are not easy to reproduce, setting the +.Dq J +option may help provoke the problem. +.Pp +In truly difficult cases, the +.Dq U +option, if supported by the kernel, can provide a detailed trace of +all calls made to these functions. +.Pp +Unfortunately this implementation does not provide much detail about +the problems it detects; the performance impact for storing such information +would be prohibitive. +There are a number of allocator implementations available on the Internet +which focus on detecting and pinpointing problems by trading performance for +extra sanity checks and detailed diagnostics. +.Sh DIAGNOSTIC MESSAGES +If any of the memory allocation/deallocation functions detect an error or +warning condition, a message will be printed to file descriptor +.Dv STDERR_FILENO . +Errors will result in the process dumping core. +If the +.Dq A +option is set, all warnings are treated as errors. +.Pp +The +.Va _malloc_message +variable allows the programmer to override the function which emits +the text strings forming the errors and warnings if for some reason +the +.Dv stderr +file descriptor is not suitable for this. +Please note that doing anything which tries to allocate memory in +this function is likely to result in a crash or deadlock. +.Pp +All messages are prefixed by +.Dq Ao Ar progname Ac Ns Li : (malloc) . +.Sh RETURN VALUES +The +.Fn malloc +and +.Fn calloc +functions return a pointer to the allocated memory if successful; otherwise +a +.Dv NULL +pointer is returned and +.Va errno +is set to +.Er ENOMEM . +.Pp +The +.Fn realloc +and +.Fn reallocf +functions return a pointer, possibly identical to +.Fa ptr , +to the allocated memory +if successful; otherwise a +.Dv NULL +pointer is returned, and +.Va errno +is set to +.Er ENOMEM +if the error was the result of an allocation failure. +The +.Fn realloc +function always leaves the original buffer intact +when an error occurs, whereas +.Fn reallocf +deallocates it in this case. +.Pp +The +.Fn free +function returns no value. +.Pp +The +.Fn malloc_usable_size +function returns the usable size of the allocation pointed to by +.Fa ptr . +.Sh ENVIRONMENT +The following environment variables affect the execution of the allocation +functions: +.Bl -tag -width ".Ev MALLOC_OPTIONS" +.It Ev MALLOC_OPTIONS +If the environment variable +.Ev MALLOC_OPTIONS +is set, the characters it contains will be interpreted as flags to the +allocation functions. +.El +.Sh EXAMPLES +To dump core whenever a problem occurs: +.Pp +.Bd -literal -offset indent +ln -s 'A' /etc/malloc.conf +.Ed +.Pp +To specify in the source that a program does no return value checking +on calls to these functions: +.Bd -literal -offset indent +_malloc_options = "X"; +.Ed +.Sh SEE ALSO +.Xr limits 1 , +.Xr madvise 2 , +.Xr mmap 2 , +.Xr sbrk 2 , +.Xr alloca 3 , +.Xr atexit 3 , +.Xr getpagesize 3 , +.Xr memory 3 , +.Xr posix_memalign 3 +.Sh STANDARDS +The +.Fn malloc , +.Fn calloc , +.Fn realloc +and +.Fn free +functions conform to +.St -isoC . +.Sh HISTORY +The +.Fn reallocf +function first appeared in +.Fx 3.0 . +.Pp +The +.Fn malloc_usable_size +function first appeared in +.Fx 7.0 . diff --git a/varnish-cache/lib/libjemalloc/malloc.c b/varnish-cache/lib/libjemalloc/malloc.c new file mode 100644 index 00000000..b030e742 --- /dev/null +++ b/varnish-cache/lib/libjemalloc/malloc.c @@ -0,0 +1,5589 @@ +/*- + * Copyright (C) 2006-2008 Jason Evans . + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice(s), this list of conditions and the following disclaimer as + * the first lines of this file unmodified other than the possible + * addition of one or more copyright notices. + * 2. Redistributions in binary form must reproduce the above copyright + * notice(s), this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE + * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR + * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, + * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE + * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, + * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + ******************************************************************************* + * + * This allocator implementation is designed to provide scalable performance + * for multi-threaded programs on multi-processor systems. The following + * features are included for this purpose: + * + * + Multiple arenas are used if there are multiple CPUs, which reduces lock + * contention and cache sloshing. + * + * + Thread-specific caching is used if there are multiple threads, which + * reduces the amount of locking. + * + * + Cache line sharing between arenas is avoided for internal data + * structures. + * + * + Memory is managed in chunks and runs (chunks can be split into runs), + * rather than as individual pages. This provides a constant-time + * mechanism for associating allocations with particular arenas. + * + * Allocation requests are rounded up to the nearest size class, and no record + * of the original request size is maintained. Allocations are broken into + * categories according to size class. Assuming runtime defaults, 4 kB pages + * and a 16 byte quantum on a 32-bit system, the size classes in each category + * are as follows: + * + * |=======================================| + * | Category | Subcategory | Size | + * |=======================================| + * | Small | Tiny | 2 | + * | | | 4 | + * | | | 8 | + * | |------------------+---------| + * | | Quantum-spaced | 16 | + * | | | 32 | + * | | | 48 | + * | | | ... | + * | | | 96 | + * | | | 112 | + * | | | 128 | + * | |------------------+---------| + * | | Cacheline-spaced | 192 | + * | | | 256 | + * | | | 320 | + * | | | 384 | + * | | | 448 | + * | | | 512 | + * | |------------------+---------| + * | | Sub-page | 760 | + * | | | 1024 | + * | | | 1280 | + * | | | ... | + * | | | 3328 | + * | | | 3584 | + * | | | 3840 | + * |=======================================| + * | Large | 4 kB | + * | | 8 kB | + * | | 12 kB | + * | | ... | + * | | 1012 kB | + * | | 1016 kB | + * | | 1020 kB | + * |=======================================| + * | Huge | 1 MB | + * | | 2 MB | + * | | 3 MB | + * | | ... | + * |=======================================| + * + * A different mechanism is used for each category: + * + * Small : Each size class is segregated into its own set of runs. Each run + * maintains a bitmap of which regions are free/allocated. + * + * Large : Each allocation is backed by a dedicated run. Metadata are stored + * in the associated arena chunk header maps. + * + * Huge : Each allocation is backed by a dedicated contiguous set of chunks. + * Metadata are stored in a separate red-black tree. + * + ******************************************************************************* + */ + +/* + * MALLOC_PRODUCTION disables assertions and statistics gathering. It also + * defaults the A and J runtime options to off. These settings are appropriate + * for production systems. + */ +/* #define MALLOC_PRODUCTION */ + +#ifndef MALLOC_PRODUCTION + /* + * MALLOC_DEBUG enables assertions and other sanity checks, and disables + * inline functions. + */ +# define MALLOC_DEBUG + + /* MALLOC_STATS enables statistics calculation. */ +# define MALLOC_STATS +#endif + +/* + * MALLOC_TINY enables support for tiny objects, which are smaller than one + * quantum. + */ +#define MALLOC_TINY + +/* + * MALLOC_MAG enables a magazine-based thread-specific caching layer for small + * objects. This makes it possible to allocate/deallocate objects without any + * locking when the cache is in the steady state. + */ +#define MALLOC_MAG + +/* + * MALLOC_BALANCE enables monitoring of arena lock contention and dynamically + * re-balances arena load if exponentially averaged contention exceeds a + * certain threshold. + */ +#define MALLOC_BALANCE + +/* + * MALLOC_DSS enables use of sbrk(2) to allocate chunks from the data storage + * segment (DSS). In an ideal world, this functionality would be completely + * unnecessary, but we are burdened by history and the lack of resource limits + * for anonymous mapped memory. + */ +#define MALLOC_DSS + +#include +__FBSDID("$FreeBSD: head/lib/libc/stdlib/malloc.c 182225 2008-08-27 02:00:53Z jasone $"); + +#include "libc_private.h" +#ifdef MALLOC_DEBUG +# define _LOCK_DEBUG +#endif +#include "spinlock.h" +#include "namespace.h" +#include +#include +#include +#include +#include +#include +#include +#include /* Must come after several other sys/ includes. */ + +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "un-namespace.h" + +#ifdef MALLOC_DEBUG +# ifdef NDEBUG +# undef NDEBUG +# endif +#else +# ifndef NDEBUG +# define NDEBUG +# endif +#endif +#include + +#include "rb.h" + +#ifdef MALLOC_DEBUG + /* Disable inlining to make debugging easier. */ +# define inline +#endif + +/* Size of stack-allocated buffer passed to strerror_r(). */ +#define STRERROR_BUF 64 + +/* + * The const_size2bin table is sized according to PAGESIZE_2POW, but for + * correctness reasons, we never assume that + * (pagesize == (1U << * PAGESIZE_2POW)). + * + * Minimum alignment of allocations is 2^QUANTUM_2POW bytes. + */ +#ifdef __i386__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 2 +# define CPU_SPINWAIT __asm__ volatile("pause") +#endif +#ifdef __ia64__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +#endif +#ifdef __alpha__ +# define PAGESIZE_2POW 13 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define NO_TLS +#endif +#ifdef __sparc64__ +# define PAGESIZE_2POW 13 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define NO_TLS +#endif +#ifdef __amd64__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 3 +# define CPU_SPINWAIT __asm__ volatile("pause") +#endif +#ifdef __arm__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 3 +# define SIZEOF_PTR_2POW 2 +# define NO_TLS +#endif +#ifdef __mips__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 3 +# define SIZEOF_PTR_2POW 2 +# define NO_TLS +#endif +#ifdef __powerpc__ +# define PAGESIZE_2POW 12 +# define QUANTUM_2POW 4 +# define SIZEOF_PTR_2POW 2 +#endif + +#define QUANTUM ((size_t)(1U << QUANTUM_2POW)) +#define QUANTUM_MASK (QUANTUM - 1) + +#define SIZEOF_PTR (1U << SIZEOF_PTR_2POW) + +/* sizeof(int) == (1U << SIZEOF_INT_2POW). */ +#ifndef SIZEOF_INT_2POW +# define SIZEOF_INT_2POW 2 +#endif + +/* We can't use TLS in non-PIC programs, since TLS relies on loader magic. */ +#if (!defined(PIC) && !defined(NO_TLS)) +# define NO_TLS +#endif + +#ifdef NO_TLS + /* MALLOC_MAG requires TLS. */ +# ifdef MALLOC_MAG +# undef MALLOC_MAG +# endif + /* MALLOC_BALANCE requires TLS. */ +# ifdef MALLOC_BALANCE +# undef MALLOC_BALANCE +# endif +#endif + +/* + * Size and alignment of memory chunks that are allocated by the OS's virtual + * memory system. + */ +#define CHUNK_2POW_DEFAULT 20 + +/* Maximum number of dirty pages per arena. */ +#define DIRTY_MAX_DEFAULT (1U << 9) + +/* + * Maximum size of L1 cache line. This is used to avoid cache line aliasing. + * In addition, this controls the spacing of cacheline-spaced size classes. + */ +#define CACHELINE_2POW 6 +#define CACHELINE ((size_t)(1U << CACHELINE_2POW)) +#define CACHELINE_MASK (CACHELINE - 1) + +/* + * Subpages are an artificially designated partitioning of pages. Their only + * purpose is to support subpage-spaced size classes. + * + * There must be at least 4 subpages per page, due to the way size classes are + * handled. + */ +#define SUBPAGE_2POW 8 +#define SUBPAGE ((size_t)(1U << SUBPAGE_2POW)) +#define SUBPAGE_MASK (SUBPAGE - 1) + +#ifdef MALLOC_TINY + /* Smallest size class to support. */ +# define TINY_MIN_2POW 1 +#endif + +/* + * Maximum size class that is a multiple of the quantum, but not (necessarily) + * a power of 2. Above this size, allocations are rounded up to the nearest + * power of 2. + */ +#define QSPACE_MAX_2POW_DEFAULT 7 + +/* + * Maximum size class that is a multiple of the cacheline, but not (necessarily) + * a power of 2. Above this size, allocations are rounded up to the nearest + * power of 2. + */ +#define CSPACE_MAX_2POW_DEFAULT 9 + +/* + * RUN_MAX_OVRHD indicates maximum desired run header overhead. Runs are sized + * as small as possible such that this setting is still honored, without + * violating other constraints. The goal is to make runs as small as possible + * without exceeding a per run external fragmentation threshold. + * + * We use binary fixed point math for overhead computations, where the binary + * point is implicitly RUN_BFP bits to the left. + * + * Note that it is possible to set RUN_MAX_OVRHD low enough that it cannot be + * honored for some/all object sizes, since there is one bit of header overhead + * per object (plus a constant). This constraint is relaxed (ignored) for runs + * that are so small that the per-region overhead is greater than: + * + * (RUN_MAX_OVRHD / (reg_size << (3+RUN_BFP)) + */ +#define RUN_BFP 12 +/* \/ Implicit binary fixed point. */ +#define RUN_MAX_OVRHD 0x0000003dU +#define RUN_MAX_OVRHD_RELAX 0x00001800U + +/* Put a cap on small object run size. This overrides RUN_MAX_OVRHD. */ +#define RUN_MAX_SMALL (12 * pagesize) + +/* + * Hyper-threaded CPUs may need a special instruction inside spin loops in + * order to yield to another virtual CPU. If no such instruction is defined + * above, make CPU_SPINWAIT a no-op. + */ +#ifndef CPU_SPINWAIT +# define CPU_SPINWAIT +#endif + +/* + * Adaptive spinning must eventually switch to blocking, in order to avoid the + * potential for priority inversion deadlock. Backing off past a certain point + * can actually waste time. + */ +#define SPIN_LIMIT_2POW 11 + +/* + * Conversion from spinning to blocking is expensive; we use (1U << + * BLOCK_COST_2POW) to estimate how many more times costly blocking is than + * worst-case spinning. + */ +#define BLOCK_COST_2POW 4 + +#ifdef MALLOC_MAG + /* + * Default magazine size, in bytes. max_rounds is calculated to make + * optimal use of the space, leaving just enough room for the magazine + * header. + */ +# define MAG_SIZE_2POW_DEFAULT 9 +#endif + +#ifdef MALLOC_BALANCE + /* + * We use an exponential moving average to track recent lock contention, + * where the size of the history window is N, and alpha=2/(N+1). + * + * Due to integer math rounding, very small values here can cause + * substantial degradation in accuracy, thus making the moving average decay + * faster than it would with precise calculation. + */ +# define BALANCE_ALPHA_INV_2POW 9 + + /* + * Threshold value for the exponential moving contention average at which to + * re-assign a thread. + */ +# define BALANCE_THRESHOLD_DEFAULT (1U << (SPIN_LIMIT_2POW-4)) +#endif + +/******************************************************************************/ + +/* + * Mutexes based on spinlocks. We can't use normal pthread spinlocks in all + * places, because they require malloc()ed memory, which causes bootstrapping + * issues in some cases. + */ +typedef struct { + spinlock_t lock; +} malloc_mutex_t; + +/* Set to true once the allocator has been initialized. */ +static bool malloc_initialized = false; + +/* Used to avoid initialization races. */ +static malloc_mutex_t init_lock = {_SPINLOCK_INITIALIZER}; + +/******************************************************************************/ +/* + * Statistics data structures. + */ + +#ifdef MALLOC_STATS + +typedef struct malloc_bin_stats_s malloc_bin_stats_t; +struct malloc_bin_stats_s { + /* + * Number of allocation requests that corresponded to the size of this + * bin. + */ + uint64_t nrequests; + +#ifdef MALLOC_MAG + /* Number of magazine reloads from this bin. */ + uint64_t nmags; +#endif + + /* Total number of runs created for this bin's size class. */ + uint64_t nruns; + + /* + * Total number of runs reused by extracting them from the runs tree for + * this bin's size class. + */ + uint64_t reruns; + + /* High-water mark for this bin. */ + unsigned long highruns; + + /* Current number of runs in this bin. */ + unsigned long curruns; +}; + +typedef struct arena_stats_s arena_stats_t; +struct arena_stats_s { + /* Number of bytes currently mapped. */ + size_t mapped; + + /* + * Total number of purge sweeps, total number of madvise calls made, + * and total pages purged in order to keep dirty unused memory under + * control. + */ + uint64_t npurge; + uint64_t nmadvise; + uint64_t purged; + + /* Per-size-category statistics. */ + size_t allocated_small; + uint64_t nmalloc_small; + uint64_t ndalloc_small; + + size_t allocated_large; + uint64_t nmalloc_large; + uint64_t ndalloc_large; + +#ifdef MALLOC_BALANCE + /* Number of times this arena reassigned a thread due to contention. */ + uint64_t nbalance; +#endif +}; + +typedef struct chunk_stats_s chunk_stats_t; +struct chunk_stats_s { + /* Number of chunks that were allocated. */ + uint64_t nchunks; + + /* High-water mark for number of chunks allocated. */ + unsigned long highchunks; + + /* + * Current number of chunks allocated. This value isn't maintained for + * any other purpose, so keep track of it in order to be able to set + * highchunks. + */ + unsigned long curchunks; +}; + +#endif /* #ifdef MALLOC_STATS */ + +/******************************************************************************/ +/* + * Extent data structures. + */ + +/* Tree of extents. */ +typedef struct extent_node_s extent_node_t; +struct extent_node_s { +#ifdef MALLOC_DSS + /* Linkage for the size/address-ordered tree. */ + rb_node(extent_node_t) link_szad; +#endif + + /* Linkage for the address-ordered tree. */ + rb_node(extent_node_t) link_ad; + + /* Pointer to the extent that this tree node is responsible for. */ + void *addr; + + /* Total region size. */ + size_t size; +}; +typedef rb_tree(extent_node_t) extent_tree_t; + +/******************************************************************************/ +/* + * Arena data structures. + */ + +typedef struct arena_s arena_t; +typedef struct arena_bin_s arena_bin_t; + +/* Each element of the chunk map corresponds to one page within the chunk. */ +typedef struct arena_chunk_map_s arena_chunk_map_t; +struct arena_chunk_map_s { + /* + * Linkage for run trees. There are two disjoint uses: + * + * 1) arena_t's runs_avail tree. + * 2) arena_run_t conceptually uses this linkage for in-use non-full + * runs, rather than directly embedding linkage. + */ + rb_node(arena_chunk_map_t) link; + + /* + * Run address (or size) and various flags are stored together. The bit + * layout looks like (assuming 32-bit system): + * + * ???????? ???????? ????---- ---kdzla + * + * ? : Unallocated: Run address for first/last pages, unset for internal + * pages. + * Small: Run address. + * Large: Run size for first page, unset for trailing pages. + * - : Unused. + * k : key? + * d : dirty? + * z : zeroed? + * l : large? + * a : allocated? + * + * Following are example bit patterns for the three types of runs. + * + * r : run address + * s : run size + * x : don't care + * - : 0 + * [dzla] : bit set + * + * Unallocated: + * ssssssss ssssssss ssss---- -------- + * xxxxxxxx xxxxxxxx xxxx---- ----d--- + * ssssssss ssssssss ssss---- -----z-- + * + * Small: + * rrrrrrrr rrrrrrrr rrrr---- -------a + * rrrrrrrr rrrrrrrr rrrr---- -------a + * rrrrrrrr rrrrrrrr rrrr---- -------a + * + * Large: + * ssssssss ssssssss ssss---- ------la + * -------- -------- -------- ------la + * -------- -------- -------- ------la + */ + size_t bits; +#define CHUNK_MAP_KEY ((size_t)0x10U) +#define CHUNK_MAP_DIRTY ((size_t)0x08U) +#define CHUNK_MAP_ZEROED ((size_t)0x04U) +#define CHUNK_MAP_LARGE ((size_t)0x02U) +#define CHUNK_MAP_ALLOCATED ((size_t)0x01U) +}; +typedef rb_tree(arena_chunk_map_t) arena_avail_tree_t; +typedef rb_tree(arena_chunk_map_t) arena_run_tree_t; + +/* Arena chunk header. */ +typedef struct arena_chunk_s arena_chunk_t; +struct arena_chunk_s { + /* Arena that owns the chunk. */ + arena_t *arena; + + /* Linkage for the arena's chunks_dirty tree. */ + rb_node(arena_chunk_t) link_dirty; + + /* Number of dirty pages. */ + size_t ndirty; + + /* Map of pages within chunk that keeps track of free/large/small. */ + arena_chunk_map_t map[1]; /* Dynamically sized. */ +}; +typedef rb_tree(arena_chunk_t) arena_chunk_tree_t; + +typedef struct arena_run_s arena_run_t; +struct arena_run_s { +#ifdef MALLOC_DEBUG + uint32_t magic; +# define ARENA_RUN_MAGIC 0x384adf93 +#endif + + /* Bin this run is associated with. */ + arena_bin_t *bin; + + /* Index of first element that might have a free region. */ + unsigned regs_minelm; + + /* Number of free regions in run. */ + unsigned nfree; + + /* Bitmask of in-use regions (0: in use, 1: free). */ + unsigned regs_mask[1]; /* Dynamically sized. */ +}; + +struct arena_bin_s { + /* + * Current run being used to service allocations of this bin's size + * class. + */ + arena_run_t *runcur; + + /* + * Tree of non-full runs. This tree is used when looking for an + * existing run when runcur is no longer usable. We choose the + * non-full run that is lowest in memory; this policy tends to keep + * objects packed well, and it can also help reduce the number of + * almost-empty chunks. + */ + arena_run_tree_t runs; + + /* Size of regions in a run for this bin's size class. */ + size_t reg_size; + + /* Total size of a run for this bin's size class. */ + size_t run_size; + + /* Total number of regions in a run for this bin's size class. */ + uint32_t nregs; + + /* Number of elements in a run's regs_mask for this bin's size class. */ + uint32_t regs_mask_nelms; + + /* Offset of first region in a run for this bin's size class. */ + uint32_t reg0_offset; + +#ifdef MALLOC_STATS + /* Bin statistics. */ + malloc_bin_stats_t stats; +#endif +}; + +struct arena_s { +#ifdef MALLOC_DEBUG + uint32_t magic; +# define ARENA_MAGIC 0x947d3d24 +#endif + + /* All operations on this arena require that lock be locked. */ + pthread_mutex_t lock; + +#ifdef MALLOC_STATS + arena_stats_t stats; +#endif + + /* Tree of dirty-page-containing chunks this arena manages. */ + arena_chunk_tree_t chunks_dirty; + + /* + * In order to avoid rapid chunk allocation/deallocation when an arena + * oscillates right on the cusp of needing a new chunk, cache the most + * recently freed chunk. The spare is left in the arena's chunk trees + * until it is deleted. + * + * There is one spare chunk per arena, rather than one spare total, in + * order to avoid interactions between multiple threads that could make + * a single spare inadequate. + */ + arena_chunk_t *spare; + + /* + * Current count of pages within unused runs that are potentially + * dirty, and for which madvise(... MADV_FREE) has not been called. By + * tracking this, we can institute a limit on how much dirty unused + * memory is mapped for each arena. + */ + size_t ndirty; + + /* + * Size/address-ordered tree of this arena's available runs. This tree + * is used for first-best-fit run allocation. + */ + arena_avail_tree_t runs_avail; + +#ifdef MALLOC_BALANCE + /* + * The arena load balancing machinery needs to keep track of how much + * lock contention there is. This value is exponentially averaged. + */ + uint32_t contention; +#endif + + /* + * bins is used to store rings of free regions of the following sizes, + * assuming a 16-byte quantum, 4kB pagesize, and default MALLOC_OPTIONS. + * + * bins[i] | size | + * --------+------+ + * 0 | 2 | + * 1 | 4 | + * 2 | 8 | + * --------+------+ + * 3 | 16 | + * 4 | 32 | + * 5 | 48 | + * 6 | 64 | + * : : + * : : + * 33 | 496 | + * 34 | 512 | + * --------+------+ + * 35 | 1024 | + * 36 | 2048 | + * --------+------+ + */ + arena_bin_t bins[1]; /* Dynamically sized. */ +}; + +/******************************************************************************/ +/* + * Magazine data structures. + */ + +#ifdef MALLOC_MAG +typedef struct mag_s mag_t; +struct mag_s { + size_t binind; /* Index of associated bin. */ + size_t nrounds; + void *rounds[1]; /* Dynamically sized. */ +}; + +/* + * Magazines are lazily allocated, but once created, they remain until the + * associated mag_rack is destroyed. + */ +typedef struct bin_mags_s bin_mags_t; +struct bin_mags_s { + mag_t *curmag; + mag_t *sparemag; +}; + +typedef struct mag_rack_s mag_rack_t; +struct mag_rack_s { + bin_mags_t bin_mags[1]; /* Dynamically sized. */ +}; +#endif + +/******************************************************************************/ +/* + * Data. + */ + +/* Number of CPUs. */ +static unsigned ncpus; + +/* VM page size. */ +static size_t pagesize; +static size_t pagesize_mask; +static size_t pagesize_2pow; + +/* Various bin-related settings. */ +#ifdef MALLOC_TINY /* Number of (2^n)-spaced tiny bins. */ +# define ntbins ((unsigned)(QUANTUM_2POW - TINY_MIN_2POW)) +#else +# define ntbins 0 +#endif +static unsigned nqbins; /* Number of quantum-spaced bins. */ +static unsigned ncbins; /* Number of cacheline-spaced bins. */ +static unsigned nsbins; /* Number of subpage-spaced bins. */ +static unsigned nbins; +#ifdef MALLOC_TINY +# define tspace_max ((size_t)(QUANTUM >> 1)) +#endif +#define qspace_min QUANTUM +static size_t qspace_max; +static size_t cspace_min; +static size_t cspace_max; +static size_t sspace_min; +static size_t sspace_max; +#define bin_maxclass sspace_max + +static uint8_t const *size2bin; +/* + * const_size2bin is a static constant lookup table that in the common case can + * be used as-is for size2bin. For dynamically linked programs, this avoids + * a page of memory overhead per process. + */ +#define S2B_1(i) i, +#define S2B_2(i) S2B_1(i) S2B_1(i) +#define S2B_4(i) S2B_2(i) S2B_2(i) +#define S2B_8(i) S2B_4(i) S2B_4(i) +#define S2B_16(i) S2B_8(i) S2B_8(i) +#define S2B_32(i) S2B_16(i) S2B_16(i) +#define S2B_64(i) S2B_32(i) S2B_32(i) +#define S2B_128(i) S2B_64(i) S2B_64(i) +#define S2B_256(i) S2B_128(i) S2B_128(i) +static const uint8_t const_size2bin[(1U << PAGESIZE_2POW) - 255] = { + S2B_1(0xffU) /* 0 */ +#if (QUANTUM_2POW == 4) +/* 64-bit system ************************/ +# ifdef MALLOC_TINY + S2B_2(0) /* 2 */ + S2B_2(1) /* 4 */ + S2B_4(2) /* 8 */ + S2B_8(3) /* 16 */ +# define S2B_QMIN 3 +# else + S2B_16(0) /* 16 */ +# define S2B_QMIN 0 +# endif + S2B_16(S2B_QMIN + 1) /* 32 */ + S2B_16(S2B_QMIN + 2) /* 48 */ + S2B_16(S2B_QMIN + 3) /* 64 */ + S2B_16(S2B_QMIN + 4) /* 80 */ + S2B_16(S2B_QMIN + 5) /* 96 */ + S2B_16(S2B_QMIN + 6) /* 112 */ + S2B_16(S2B_QMIN + 7) /* 128 */ +# define S2B_CMIN (S2B_QMIN + 8) +#else +/* 32-bit system ************************/ +# ifdef MALLOC_TINY + S2B_2(0) /* 2 */ + S2B_2(1) /* 4 */ + S2B_4(2) /* 8 */ +# define S2B_QMIN 2 +# else + S2B_8(0) /* 8 */ +# define S2B_QMIN 0 +# endif + S2B_8(S2B_QMIN + 1) /* 16 */ + S2B_8(S2B_QMIN + 2) /* 24 */ + S2B_8(S2B_QMIN + 3) /* 32 */ + S2B_8(S2B_QMIN + 4) /* 40 */ + S2B_8(S2B_QMIN + 5) /* 48 */ + S2B_8(S2B_QMIN + 6) /* 56 */ + S2B_8(S2B_QMIN + 7) /* 64 */ + S2B_8(S2B_QMIN + 8) /* 72 */ + S2B_8(S2B_QMIN + 9) /* 80 */ + S2B_8(S2B_QMIN + 10) /* 88 */ + S2B_8(S2B_QMIN + 11) /* 96 */ + S2B_8(S2B_QMIN + 12) /* 104 */ + S2B_8(S2B_QMIN + 13) /* 112 */ + S2B_8(S2B_QMIN + 14) /* 120 */ + S2B_8(S2B_QMIN + 15) /* 128 */ +# define S2B_CMIN (S2B_QMIN + 16) +#endif +/****************************************/ + S2B_64(S2B_CMIN + 0) /* 192 */ + S2B_64(S2B_CMIN + 1) /* 256 */ + S2B_64(S2B_CMIN + 2) /* 320 */ + S2B_64(S2B_CMIN + 3) /* 384 */ + S2B_64(S2B_CMIN + 4) /* 448 */ + S2B_64(S2B_CMIN + 5) /* 512 */ +# define S2B_SMIN (S2B_CMIN + 6) + S2B_256(S2B_SMIN + 0) /* 768 */ + S2B_256(S2B_SMIN + 1) /* 1024 */ + S2B_256(S2B_SMIN + 2) /* 1280 */ + S2B_256(S2B_SMIN + 3) /* 1536 */ + S2B_256(S2B_SMIN + 4) /* 1792 */ + S2B_256(S2B_SMIN + 5) /* 2048 */ + S2B_256(S2B_SMIN + 6) /* 2304 */ + S2B_256(S2B_SMIN + 7) /* 2560 */ + S2B_256(S2B_SMIN + 8) /* 2816 */ + S2B_256(S2B_SMIN + 9) /* 3072 */ + S2B_256(S2B_SMIN + 10) /* 3328 */ + S2B_256(S2B_SMIN + 11) /* 3584 */ + S2B_256(S2B_SMIN + 12) /* 3840 */ +#if (PAGESIZE_2POW == 13) + S2B_256(S2B_SMIN + 13) /* 4096 */ + S2B_256(S2B_SMIN + 14) /* 4352 */ + S2B_256(S2B_SMIN + 15) /* 4608 */ + S2B_256(S2B_SMIN + 16) /* 4864 */ + S2B_256(S2B_SMIN + 17) /* 5120 */ + S2B_256(S2B_SMIN + 18) /* 5376 */ + S2B_256(S2B_SMIN + 19) /* 5632 */ + S2B_256(S2B_SMIN + 20) /* 5888 */ + S2B_256(S2B_SMIN + 21) /* 6144 */ + S2B_256(S2B_SMIN + 22) /* 6400 */ + S2B_256(S2B_SMIN + 23) /* 6656 */ + S2B_256(S2B_SMIN + 24) /* 6912 */ + S2B_256(S2B_SMIN + 25) /* 7168 */ + S2B_256(S2B_SMIN + 26) /* 7424 */ + S2B_256(S2B_SMIN + 27) /* 7680 */ + S2B_256(S2B_SMIN + 28) /* 7936 */ +#endif +}; +#undef S2B_1 +#undef S2B_2 +#undef S2B_4 +#undef S2B_8 +#undef S2B_16 +#undef S2B_32 +#undef S2B_64 +#undef S2B_128 +#undef S2B_256 +#undef S2B_QMIN +#undef S2B_CMIN +#undef S2B_SMIN + +#ifdef MALLOC_MAG +static size_t max_rounds; +#endif + +/* Various chunk-related settings. */ +static size_t chunksize; +static size_t chunksize_mask; /* (chunksize - 1). */ +static size_t chunk_npages; +static size_t arena_chunk_header_npages; +static size_t arena_maxclass; /* Max size class for arenas. */ + +/********/ +/* + * Chunks. + */ + +/* Protects chunk-related data structures. */ +static malloc_mutex_t huge_mtx; + +/* Tree of chunks that are stand-alone huge allocations. */ +static extent_tree_t huge; + +#ifdef MALLOC_DSS +/* + * Protects sbrk() calls. This avoids malloc races among threads, though it + * does not protect against races with threads that call sbrk() directly. + */ +static malloc_mutex_t dss_mtx; +/* Base address of the DSS. */ +static void *dss_base; +/* Current end of the DSS, or ((void *)-1) if the DSS is exhausted. */ +static void *dss_prev; +/* Current upper limit on DSS addresses. */ +static void *dss_max; + +/* + * Trees of chunks that were previously allocated (trees differ only in node + * ordering). These are used when allocating chunks, in an attempt to re-use + * address space. Depending on function, different tree orderings are needed, + * which is why there are two trees with the same contents. + */ +static extent_tree_t dss_chunks_szad; +static extent_tree_t dss_chunks_ad; +#endif + +#ifdef MALLOC_STATS +/* Huge allocation statistics. */ +static uint64_t huge_nmalloc; +static uint64_t huge_ndalloc; +static size_t huge_allocated; +#endif + +/****************************/ +/* + * base (internal allocation). + */ + +/* + * Current pages that are being used for internal memory allocations. These + * pages are carved up in cacheline-size quanta, so that there is no chance of + * false cache line sharing. + */ +static void *base_pages; +static void *base_next_addr; +static void *base_past_addr; /* Addr immediately past base_pages. */ +static extent_node_t *base_nodes; +static malloc_mutex_t base_mtx; +#ifdef MALLOC_STATS +static size_t base_mapped; +#endif + +/********/ +/* + * Arenas. + */ + +/* + * Arenas that are used to service external requests. Not all elements of the + * arenas array are necessarily used; arenas are created lazily as needed. + */ +static arena_t **arenas; +static unsigned narenas; +#ifndef NO_TLS +# ifdef MALLOC_BALANCE +static unsigned narenas_2pow; +# else +static unsigned next_arena; +# endif +#endif +static pthread_mutex_t arenas_lock; /* Protects arenas initialization. */ + +#ifndef NO_TLS +/* + * Map of pthread_self() --> arenas[???], used for selecting an arena to use + * for allocations. + */ +static __thread arena_t *arenas_map; +#endif + +#ifdef MALLOC_MAG +/* + * Map of thread-specific magazine racks, used for thread-specific object + * caching. + */ +static __thread mag_rack_t *mag_rack; +#endif + +#ifdef MALLOC_STATS +/* Chunk statistics. */ +static chunk_stats_t stats_chunks; +#endif + +/*******************************/ +/* + * Runtime configuration options. + */ +const char *_malloc_options; + +#ifndef MALLOC_PRODUCTION +static bool opt_abort = true; +static bool opt_junk = true; +#else +static bool opt_abort = false; +static bool opt_junk = false; +#endif +#ifdef MALLOC_DSS +static bool opt_dss = true; +static bool opt_mmap = true; +#endif +#ifdef MALLOC_MAG +static bool opt_mag = true; +static size_t opt_mag_size_2pow = MAG_SIZE_2POW_DEFAULT; +#endif +static size_t opt_dirty_max = DIRTY_MAX_DEFAULT; +#ifdef MALLOC_BALANCE +static uint64_t opt_balance_threshold = BALANCE_THRESHOLD_DEFAULT; +#endif +static bool opt_print_stats = false; +static size_t opt_qspace_max_2pow = QSPACE_MAX_2POW_DEFAULT; +static size_t opt_cspace_max_2pow = CSPACE_MAX_2POW_DEFAULT; +static size_t opt_chunk_2pow = CHUNK_2POW_DEFAULT; +static bool opt_utrace = false; +static bool opt_sysv = false; +static bool opt_xmalloc = false; +static bool opt_zero = false; +static int opt_narenas_lshift = 0; + +typedef struct { + void *p; + size_t s; + void *r; +} malloc_utrace_t; + +#define UTRACE(a, b, c) \ + if (opt_utrace) { \ + malloc_utrace_t ut; \ + ut.p = (a); \ + ut.s = (b); \ + ut.r = (c); \ + utrace(&ut, sizeof(ut)); \ + } + +/******************************************************************************/ +/* + * Begin function prototypes for non-inline static functions. + */ + +static void malloc_mutex_init(malloc_mutex_t *mutex); +static bool malloc_spin_init(pthread_mutex_t *lock); +static void wrtmessage(const char *p1, const char *p2, const char *p3, + const char *p4); +#ifdef MALLOC_STATS +static void malloc_printf(const char *format, ...); +#endif +static char *umax2s(uintmax_t x, char *s); +#ifdef MALLOC_DSS +static bool base_pages_alloc_dss(size_t minsize); +#endif +static bool base_pages_alloc_mmap(size_t minsize); +static bool base_pages_alloc(size_t minsize); +static void *base_alloc(size_t size); +static void *base_calloc(size_t number, size_t size); +static extent_node_t *base_node_alloc(void); +static void base_node_dealloc(extent_node_t *node); +#ifdef MALLOC_STATS +static void stats_print(arena_t *arena); +#endif +static void *pages_map(void *addr, size_t size); +static void pages_unmap(void *addr, size_t size); +#ifdef MALLOC_DSS +static void *chunk_alloc_dss(size_t size); +static void *chunk_recycle_dss(size_t size, bool zero); +#endif +static void *chunk_alloc_mmap(size_t size); +static void *chunk_alloc(size_t size, bool zero); +#ifdef MALLOC_DSS +static extent_node_t *chunk_dealloc_dss_record(void *chunk, size_t size); +static bool chunk_dealloc_dss(void *chunk, size_t size); +#endif +static void chunk_dealloc_mmap(void *chunk, size_t size); +static void chunk_dealloc(void *chunk, size_t size); +#ifndef NO_TLS +static arena_t *choose_arena_hard(void); +#endif +static void arena_run_split(arena_t *arena, arena_run_t *run, size_t size, + bool large, bool zero); +static arena_chunk_t *arena_chunk_alloc(arena_t *arena); +static void arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk); +static arena_run_t *arena_run_alloc(arena_t *arena, size_t size, bool large, + bool zero); +static void arena_purge(arena_t *arena); +static void arena_run_dalloc(arena_t *arena, arena_run_t *run, bool dirty); +static void arena_run_trim_head(arena_t *arena, arena_chunk_t *chunk, + arena_run_t *run, size_t oldsize, size_t newsize); +static void arena_run_trim_tail(arena_t *arena, arena_chunk_t *chunk, + arena_run_t *run, size_t oldsize, size_t newsize, bool dirty); +static arena_run_t *arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin); +static void *arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin); +static size_t arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size); +#ifdef MALLOC_BALANCE +static void arena_lock_balance_hard(arena_t *arena); +#endif +#ifdef MALLOC_MAG +static void mag_load(mag_t *mag); +#endif +static void *arena_malloc_large(arena_t *arena, size_t size, bool zero); +static void *arena_palloc(arena_t *arena, size_t alignment, size_t size, + size_t alloc_size); +static size_t arena_salloc(const void *ptr); +#ifdef MALLOC_MAG +static void mag_unload(mag_t *mag); +#endif +static void arena_dalloc_large(arena_t *arena, arena_chunk_t *chunk, + void *ptr); +static void arena_ralloc_large_shrink(arena_t *arena, arena_chunk_t *chunk, + void *ptr, size_t size, size_t oldsize); +static bool arena_ralloc_large_grow(arena_t *arena, arena_chunk_t *chunk, + void *ptr, size_t size, size_t oldsize); +static bool arena_ralloc_large(void *ptr, size_t size, size_t oldsize); +static void *arena_ralloc(void *ptr, size_t size, size_t oldsize); +static bool arena_new(arena_t *arena); +static arena_t *arenas_extend(unsigned ind); +#ifdef MALLOC_MAG +static mag_t *mag_create(arena_t *arena, size_t binind); +static void mag_destroy(mag_t *mag); +static mag_rack_t *mag_rack_create(arena_t *arena); +static void mag_rack_destroy(mag_rack_t *rack); +#endif +static void *huge_malloc(size_t size, bool zero); +static void *huge_palloc(size_t alignment, size_t size); +static void *huge_ralloc(void *ptr, size_t size, size_t oldsize); +static void huge_dalloc(void *ptr); +static void malloc_print_stats(void); +#ifdef MALLOC_DEBUG +static void size2bin_validate(void); +#endif +static bool size2bin_init(void); +static bool size2bin_init_hard(void); +static bool malloc_init_hard(void); + +/* + * End function prototypes. + */ +/******************************************************************************/ +/* + * Begin mutex. We can't use normal pthread mutexes in all places, because + * they require malloc()ed memory, which causes bootstrapping issues in some + * cases. + */ + +static void +malloc_mutex_init(malloc_mutex_t *mutex) +{ + static const spinlock_t lock = _SPINLOCK_INITIALIZER; + + mutex->lock = lock; +} + +static inline void +malloc_mutex_lock(malloc_mutex_t *mutex) +{ + + if (__isthreaded) + _SPINLOCK(&mutex->lock); +} + +static inline void +malloc_mutex_unlock(malloc_mutex_t *mutex) +{ + + if (__isthreaded) + _SPINUNLOCK(&mutex->lock); +} + +/* + * End mutex. + */ +/******************************************************************************/ +/* + * Begin spin lock. Spin locks here are actually adaptive mutexes that block + * after a period of spinning, because unbounded spinning would allow for + * priority inversion. + */ + +/* + * We use an unpublished interface to initialize pthread mutexes with an + * allocation callback, in order to avoid infinite recursion. + */ +int _pthread_mutex_init_calloc_cb(pthread_mutex_t *mutex, + void *(calloc_cb)(size_t, size_t)); + +__weak_reference(_pthread_mutex_init_calloc_cb_stub, + _pthread_mutex_init_calloc_cb); + +int +_pthread_mutex_init_calloc_cb_stub(pthread_mutex_t *mutex, + void *(calloc_cb)(size_t, size_t)) +{ + + return (0); +} + +static bool +malloc_spin_init(pthread_mutex_t *lock) +{ + + if (_pthread_mutex_init_calloc_cb(lock, base_calloc) != 0) + return (true); + + return (false); +} + +static inline unsigned +malloc_spin_lock(pthread_mutex_t *lock) +{ + unsigned ret = 0; + + if (__isthreaded) { + if (_pthread_mutex_trylock(lock) != 0) { + unsigned i; + volatile unsigned j; + + /* Exponentially back off. */ + for (i = 1; i <= SPIN_LIMIT_2POW; i++) { + for (j = 0; j < (1U << i); j++) { + ret++; + CPU_SPINWAIT; + } + + if (_pthread_mutex_trylock(lock) == 0) + return (ret); + } + + /* + * Spinning failed. Block until the lock becomes + * available, in order to avoid indefinite priority + * inversion. + */ + _pthread_mutex_lock(lock); + assert((ret << BLOCK_COST_2POW) != 0); + return (ret << BLOCK_COST_2POW); + } + } + + return (ret); +} + +static inline void +malloc_spin_unlock(pthread_mutex_t *lock) +{ + + if (__isthreaded) + _pthread_mutex_unlock(lock); +} + +/* + * End spin lock. + */ +/******************************************************************************/ +/* + * Begin Utility functions/macros. + */ + +/* Return the chunk address for allocation address a. */ +#define CHUNK_ADDR2BASE(a) \ + ((void *)((uintptr_t)(a) & ~chunksize_mask)) + +/* Return the chunk offset of address a. */ +#define CHUNK_ADDR2OFFSET(a) \ + ((size_t)((uintptr_t)(a) & chunksize_mask)) + +/* Return the smallest chunk multiple that is >= s. */ +#define CHUNK_CEILING(s) \ + (((s) + chunksize_mask) & ~chunksize_mask) + +/* Return the smallest quantum multiple that is >= a. */ +#define QUANTUM_CEILING(a) \ + (((a) + QUANTUM_MASK) & ~QUANTUM_MASK) + +/* Return the smallest cacheline multiple that is >= s. */ +#define CACHELINE_CEILING(s) \ + (((s) + CACHELINE_MASK) & ~CACHELINE_MASK) + +/* Return the smallest subpage multiple that is >= s. */ +#define SUBPAGE_CEILING(s) \ + (((s) + SUBPAGE_MASK) & ~SUBPAGE_MASK) + +/* Return the smallest pagesize multiple that is >= s. */ +#define PAGE_CEILING(s) \ + (((s) + pagesize_mask) & ~pagesize_mask) + +#ifdef MALLOC_TINY +/* Compute the smallest power of 2 that is >= x. */ +static inline size_t +pow2_ceil(size_t x) +{ + + x--; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; +#if (SIZEOF_PTR == 8) + x |= x >> 32; +#endif + x++; + return (x); +} +#endif + +#ifdef MALLOC_BALANCE +/* + * Use a simple linear congruential pseudo-random number generator: + * + * prn(y) = (a*x + c) % m + * + * where the following constants ensure maximal period: + * + * a == Odd number (relatively prime to 2^n), and (a-1) is a multiple of 4. + * c == Odd number (relatively prime to 2^n). + * m == 2^32 + * + * See Knuth's TAOCP 3rd Ed., Vol. 2, pg. 17 for details on these constraints. + * + * This choice of m has the disadvantage that the quality of the bits is + * proportional to bit position. For example. the lowest bit has a cycle of 2, + * the next has a cycle of 4, etc. For this reason, we prefer to use the upper + * bits. + */ +# define PRN_DEFINE(suffix, var, a, c) \ +static inline void \ +sprn_##suffix(uint32_t seed) \ +{ \ + var = seed; \ +} \ + \ +static inline uint32_t \ +prn_##suffix(uint32_t lg_range) \ +{ \ + uint32_t ret, x; \ + \ + assert(lg_range > 0); \ + assert(lg_range <= 32); \ + \ + x = (var * (a)) + (c); \ + var = x; \ + ret = x >> (32 - lg_range); \ + \ + return (ret); \ +} +# define SPRN(suffix, seed) sprn_##suffix(seed) +# define PRN(suffix, lg_range) prn_##suffix(lg_range) +#endif + +#ifdef MALLOC_BALANCE +/* Define the PRNG used for arena assignment. */ +static __thread uint32_t balance_x; +PRN_DEFINE(balance, balance_x, 1297, 1301) +#endif + +static void +wrtmessage(const char *p1, const char *p2, const char *p3, const char *p4) +{ + + _write(STDERR_FILENO, p1, strlen(p1)); + _write(STDERR_FILENO, p2, strlen(p2)); + _write(STDERR_FILENO, p3, strlen(p3)); + _write(STDERR_FILENO, p4, strlen(p4)); +} + +void (*_malloc_message)(const char *p1, const char *p2, const char *p3, + const char *p4) = wrtmessage; + +#ifdef MALLOC_STATS +/* + * Print to stderr in such a way as to (hopefully) avoid memory allocation. + */ +static void +malloc_printf(const char *format, ...) +{ + char buf[4096]; + va_list ap; + + va_start(ap, format); + vsnprintf(buf, sizeof(buf), format, ap); + va_end(ap); + _malloc_message(buf, "", "", ""); +} +#endif + +/* + * We don't want to depend on vsnprintf() for production builds, since that can + * cause unnecessary bloat for static binaries. umax2s() provides minimal + * integer printing functionality, so that malloc_printf() use can be limited to + * MALLOC_STATS code. + */ +#define UMAX2S_BUFSIZE 21 +static char * +umax2s(uintmax_t x, char *s) +{ + unsigned i; + + /* Make sure UMAX2S_BUFSIZE is large enough. */ + assert(sizeof(uintmax_t) <= 8); + + i = UMAX2S_BUFSIZE - 1; + s[i] = '\0'; + do { + i--; + s[i] = "0123456789"[x % 10]; + x /= 10; + } while (x > 0); + + return (&s[i]); +} + +/******************************************************************************/ + +#ifdef MALLOC_DSS +static bool +base_pages_alloc_dss(size_t minsize) +{ + + /* + * Do special DSS allocation here, since base allocations don't need to + * be chunk-aligned. + */ + malloc_mutex_lock(&dss_mtx); + if (dss_prev != (void *)-1) { + intptr_t incr; + size_t csize = CHUNK_CEILING(minsize); + + do { + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Calculate how much padding is necessary to + * chunk-align the end of the DSS. Don't worry about + * dss_max not being chunk-aligned though. + */ + incr = (intptr_t)chunksize + - (intptr_t)CHUNK_ADDR2OFFSET(dss_max); + assert(incr >= 0); + if ((size_t)incr < minsize) + incr += csize; + + dss_prev = sbrk(incr); + if (dss_prev == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev + incr); + base_pages = dss_prev; + base_next_addr = base_pages; + base_past_addr = dss_max; +#ifdef MALLOC_STATS + base_mapped += incr; +#endif + malloc_mutex_unlock(&dss_mtx); + return (false); + } + } while (dss_prev != (void *)-1); + } + malloc_mutex_unlock(&dss_mtx); + + return (true); +} +#endif + +static bool +base_pages_alloc_mmap(size_t minsize) +{ + size_t csize; + + assert(minsize != 0); + csize = PAGE_CEILING(minsize); + base_pages = pages_map(NULL, csize); + if (base_pages == NULL) + return (true); + base_next_addr = base_pages; + base_past_addr = (void *)((uintptr_t)base_pages + csize); +#ifdef MALLOC_STATS + base_mapped += csize; +#endif + + return (false); +} + +static bool +base_pages_alloc(size_t minsize) +{ + +#ifdef MALLOC_DSS + if (opt_dss) { + if (base_pages_alloc_dss(minsize) == false) + return (false); + } + + if (opt_mmap && minsize != 0) +#endif + { + if (base_pages_alloc_mmap(minsize) == false) + return (false); + } + + return (true); +} + +static void * +base_alloc(size_t size) +{ + void *ret; + size_t csize; + + /* Round size up to nearest multiple of the cacheline size. */ + csize = CACHELINE_CEILING(size); + + malloc_mutex_lock(&base_mtx); + /* Make sure there's enough space for the allocation. */ + if ((uintptr_t)base_next_addr + csize > (uintptr_t)base_past_addr) { + if (base_pages_alloc(csize)) { + malloc_mutex_unlock(&base_mtx); + return (NULL); + } + } + /* Allocate. */ + ret = base_next_addr; + base_next_addr = (void *)((uintptr_t)base_next_addr + csize); + malloc_mutex_unlock(&base_mtx); + + return (ret); +} + +static void * +base_calloc(size_t number, size_t size) +{ + void *ret; + + ret = base_alloc(number * size); + memset(ret, 0, number * size); + + return (ret); +} + +static extent_node_t * +base_node_alloc(void) +{ + extent_node_t *ret; + + malloc_mutex_lock(&base_mtx); + if (base_nodes != NULL) { + ret = base_nodes; + base_nodes = *(extent_node_t **)ret; + malloc_mutex_unlock(&base_mtx); + } else { + malloc_mutex_unlock(&base_mtx); + ret = (extent_node_t *)base_alloc(sizeof(extent_node_t)); + } + + return (ret); +} + +static void +base_node_dealloc(extent_node_t *node) +{ + + malloc_mutex_lock(&base_mtx); + *(extent_node_t **)node = base_nodes; + base_nodes = node; + malloc_mutex_unlock(&base_mtx); +} + +/******************************************************************************/ + +#ifdef MALLOC_STATS +static void +stats_print(arena_t *arena) +{ + unsigned i, gap_start; + + malloc_printf("dirty: %zu page%s dirty, %llu sweep%s," + " %llu madvise%s, %llu page%s purged\n", + arena->ndirty, arena->ndirty == 1 ? "" : "s", + arena->stats.npurge, arena->stats.npurge == 1 ? "" : "s", + arena->stats.nmadvise, arena->stats.nmadvise == 1 ? "" : "s", + arena->stats.purged, arena->stats.purged == 1 ? "" : "s"); + + malloc_printf(" allocated nmalloc ndalloc\n"); + malloc_printf("small: %12zu %12llu %12llu\n", + arena->stats.allocated_small, arena->stats.nmalloc_small, + arena->stats.ndalloc_small); + malloc_printf("large: %12zu %12llu %12llu\n", + arena->stats.allocated_large, arena->stats.nmalloc_large, + arena->stats.ndalloc_large); + malloc_printf("total: %12zu %12llu %12llu\n", + arena->stats.allocated_small + arena->stats.allocated_large, + arena->stats.nmalloc_small + arena->stats.nmalloc_large, + arena->stats.ndalloc_small + arena->stats.ndalloc_large); + malloc_printf("mapped: %12zu\n", arena->stats.mapped); + +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + malloc_printf("bins: bin size regs pgs mags " + "newruns reruns maxruns curruns\n"); + } else { +#endif + malloc_printf("bins: bin size regs pgs requests " + "newruns reruns maxruns curruns\n"); +#ifdef MALLOC_MAG + } +#endif + for (i = 0, gap_start = UINT_MAX; i < nbins; i++) { + if (arena->bins[i].stats.nruns == 0) { + if (gap_start == UINT_MAX) + gap_start = i; + } else { + if (gap_start != UINT_MAX) { + if (i > gap_start + 1) { + /* Gap of more than one size class. */ + malloc_printf("[%u..%u]\n", + gap_start, i - 1); + } else { + /* Gap of one size class. */ + malloc_printf("[%u]\n", gap_start); + } + gap_start = UINT_MAX; + } + malloc_printf( + "%13u %1s %4u %4u %3u %9llu %9llu" + " %9llu %7lu %7lu\n", + i, + i < ntbins ? "T" : i < ntbins + nqbins ? "Q" : + i < ntbins + nqbins + ncbins ? "C" : "S", + arena->bins[i].reg_size, + arena->bins[i].nregs, + arena->bins[i].run_size >> pagesize_2pow, +#ifdef MALLOC_MAG + (__isthreaded && opt_mag) ? + arena->bins[i].stats.nmags : +#endif + arena->bins[i].stats.nrequests, + arena->bins[i].stats.nruns, + arena->bins[i].stats.reruns, + arena->bins[i].stats.highruns, + arena->bins[i].stats.curruns); + } + } + if (gap_start != UINT_MAX) { + if (i > gap_start + 1) { + /* Gap of more than one size class. */ + malloc_printf("[%u..%u]\n", gap_start, i - 1); + } else { + /* Gap of one size class. */ + malloc_printf("[%u]\n", gap_start); + } + } +} +#endif + +/* + * End Utility functions/macros. + */ +/******************************************************************************/ +/* + * Begin extent tree code. + */ + +#ifdef MALLOC_DSS +static inline int +extent_szad_comp(extent_node_t *a, extent_node_t *b) +{ + int ret; + size_t a_size = a->size; + size_t b_size = b->size; + + ret = (a_size > b_size) - (a_size < b_size); + if (ret == 0) { + uintptr_t a_addr = (uintptr_t)a->addr; + uintptr_t b_addr = (uintptr_t)b->addr; + + ret = (a_addr > b_addr) - (a_addr < b_addr); + } + + return (ret); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(__unused static, extent_tree_szad_, extent_tree_t, extent_node_t, + link_szad, extent_szad_comp) +#endif + +static inline int +extent_ad_comp(extent_node_t *a, extent_node_t *b) +{ + uintptr_t a_addr = (uintptr_t)a->addr; + uintptr_t b_addr = (uintptr_t)b->addr; + + return ((a_addr > b_addr) - (a_addr < b_addr)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(__unused static, extent_tree_ad_, extent_tree_t, extent_node_t, link_ad, + extent_ad_comp) + +/* + * End extent tree code. + */ +/******************************************************************************/ +/* + * Begin chunk management functions. + */ + +static void * +pages_map(void *addr, size_t size) +{ + void *ret; + + /* + * We don't use MAP_FIXED here, because it can cause the *replacement* + * of existing mappings, and we only want to create new mappings. + */ + ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, + -1, 0); + assert(ret != NULL); + + if (ret == MAP_FAILED) + ret = NULL; + else if (addr != NULL && ret != addr) { + /* + * We succeeded in mapping memory, but not in the right place. + */ + if (munmap(ret, size) == -1) { + char buf[STRERROR_BUF]; + + strerror_r(errno, buf, sizeof(buf)); + _malloc_message(_getprogname(), + ": (malloc) Error in munmap(): ", buf, "\n"); + if (opt_abort) + abort(); + } + ret = NULL; + } + + assert(ret == NULL || (addr == NULL && ret != addr) + || (addr != NULL && ret == addr)); + return (ret); +} + +static void +pages_unmap(void *addr, size_t size) +{ + + if (munmap(addr, size) == -1) { + char buf[STRERROR_BUF]; + + strerror_r(errno, buf, sizeof(buf)); + _malloc_message(_getprogname(), + ": (malloc) Error in munmap(): ", buf, "\n"); + if (opt_abort) + abort(); + } +} + +#ifdef MALLOC_DSS +static void * +chunk_alloc_dss(size_t size) +{ + + /* + * sbrk() uses a signed increment argument, so take care not to + * interpret a huge allocation request as a negative increment. + */ + if ((intptr_t)size < 0) + return (NULL); + + malloc_mutex_lock(&dss_mtx); + if (dss_prev != (void *)-1) { + intptr_t incr; + + /* + * The loop is necessary to recover from races with other + * threads that are using the DSS for something other than + * malloc. + */ + do { + void *ret; + + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Calculate how much padding is necessary to + * chunk-align the end of the DSS. + */ + incr = (intptr_t)size + - (intptr_t)CHUNK_ADDR2OFFSET(dss_max); + if (incr == (intptr_t)size) + ret = dss_max; + else { + ret = (void *)((intptr_t)dss_max + incr); + incr += size; + } + + dss_prev = sbrk(incr); + if (dss_prev == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev + incr); + malloc_mutex_unlock(&dss_mtx); + return (ret); + } + } while (dss_prev != (void *)-1); + } + malloc_mutex_unlock(&dss_mtx); + + return (NULL); +} + +static void * +chunk_recycle_dss(size_t size, bool zero) +{ + extent_node_t *node, key; + + key.addr = NULL; + key.size = size; + malloc_mutex_lock(&dss_mtx); + node = extent_tree_szad_nsearch(&dss_chunks_szad, &key); + if (node != NULL) { + void *ret = node->addr; + + /* Remove node from the tree. */ + extent_tree_szad_remove(&dss_chunks_szad, node); + if (node->size == size) { + extent_tree_ad_remove(&dss_chunks_ad, node); + base_node_dealloc(node); + } else { + /* + * Insert the remainder of node's address range as a + * smaller chunk. Its position within dss_chunks_ad + * does not change. + */ + assert(node->size > size); + node->addr = (void *)((uintptr_t)node->addr + size); + node->size -= size; + extent_tree_szad_insert(&dss_chunks_szad, node); + } + malloc_mutex_unlock(&dss_mtx); + + if (zero) + memset(ret, 0, size); + return (ret); + } + malloc_mutex_unlock(&dss_mtx); + + return (NULL); +} +#endif + +static void * +chunk_alloc_mmap(size_t size) +{ + void *ret; + size_t offset; + + /* + * Ideally, there would be a way to specify alignment to mmap() (like + * NetBSD has), but in the absence of such a feature, we have to work + * hard to efficiently create aligned mappings. The reliable, but + * expensive method is to create a mapping that is over-sized, then + * trim the excess. However, that always results in at least one call + * to pages_unmap(). + * + * A more optimistic approach is to try mapping precisely the right + * amount, then try to append another mapping if alignment is off. In + * practice, this works out well as long as the application is not + * interleaving mappings via direct mmap() calls. If we do run into a + * situation where there is an interleaved mapping and we are unable to + * extend an unaligned mapping, our best option is to momentarily + * revert to the reliable-but-expensive method. This will tend to + * leave a gap in the memory map that is too small to cause later + * problems for the optimistic method. + */ + + ret = pages_map(NULL, size); + if (ret == NULL) + return (NULL); + + offset = CHUNK_ADDR2OFFSET(ret); + if (offset != 0) { + /* Try to extend chunk boundary. */ + if (pages_map((void *)((uintptr_t)ret + size), + chunksize - offset) == NULL) { + /* + * Extension failed. Clean up, then revert to the + * reliable-but-expensive method. + */ + pages_unmap(ret, size); + + /* Beware size_t wrap-around. */ + if (size + chunksize <= size) + return NULL; + + ret = pages_map(NULL, size + chunksize); + if (ret == NULL) + return (NULL); + + /* Clean up unneeded leading/trailing space. */ + offset = CHUNK_ADDR2OFFSET(ret); + if (offset != 0) { + /* Leading space. */ + pages_unmap(ret, chunksize - offset); + + ret = (void *)((uintptr_t)ret + + (chunksize - offset)); + + /* Trailing space. */ + pages_unmap((void *)((uintptr_t)ret + size), + offset); + } else { + /* Trailing space only. */ + pages_unmap((void *)((uintptr_t)ret + size), + chunksize); + } + } else { + /* Clean up unneeded leading space. */ + pages_unmap(ret, chunksize - offset); + ret = (void *)((uintptr_t)ret + (chunksize - offset)); + } + } + + return (ret); +} + +static void * +chunk_alloc(size_t size, bool zero) +{ + void *ret; + + assert(size != 0); + assert((size & chunksize_mask) == 0); + +#ifdef MALLOC_DSS + if (opt_dss) { + ret = chunk_recycle_dss(size, zero); + if (ret != NULL) { + goto RETURN; + } + + ret = chunk_alloc_dss(size); + if (ret != NULL) + goto RETURN; + } + + if (opt_mmap) +#endif + { + ret = chunk_alloc_mmap(size); + if (ret != NULL) + goto RETURN; + } + + /* All strategies for allocation failed. */ + ret = NULL; +RETURN: +#ifdef MALLOC_STATS + if (ret != NULL) { + stats_chunks.nchunks += (size / chunksize); + stats_chunks.curchunks += (size / chunksize); + } + if (stats_chunks.curchunks > stats_chunks.highchunks) + stats_chunks.highchunks = stats_chunks.curchunks; +#endif + + assert(CHUNK_ADDR2BASE(ret) == ret); + return (ret); +} + +#ifdef MALLOC_DSS +static extent_node_t * +chunk_dealloc_dss_record(void *chunk, size_t size) +{ + extent_node_t *node, *prev, key; + + key.addr = (void *)((uintptr_t)chunk + size); + node = extent_tree_ad_nsearch(&dss_chunks_ad, &key); + /* Try to coalesce forward. */ + if (node != NULL && node->addr == key.addr) { + /* + * Coalesce chunk with the following address range. This does + * not change the position within dss_chunks_ad, so only + * remove/insert from/into dss_chunks_szad. + */ + extent_tree_szad_remove(&dss_chunks_szad, node); + node->addr = chunk; + node->size += size; + extent_tree_szad_insert(&dss_chunks_szad, node); + } else { + /* + * Coalescing forward failed, so insert a new node. Drop + * dss_mtx during node allocation, since it is possible that a + * new base chunk will be allocated. + */ + malloc_mutex_unlock(&dss_mtx); + node = base_node_alloc(); + malloc_mutex_lock(&dss_mtx); + if (node == NULL) + return (NULL); + node->addr = chunk; + node->size = size; + extent_tree_ad_insert(&dss_chunks_ad, node); + extent_tree_szad_insert(&dss_chunks_szad, node); + } + + /* Try to coalesce backward. */ + prev = extent_tree_ad_prev(&dss_chunks_ad, node); + if (prev != NULL && (void *)((uintptr_t)prev->addr + prev->size) == + chunk) { + /* + * Coalesce chunk with the previous address range. This does + * not change the position within dss_chunks_ad, so only + * remove/insert node from/into dss_chunks_szad. + */ + extent_tree_szad_remove(&dss_chunks_szad, prev); + extent_tree_ad_remove(&dss_chunks_ad, prev); + + extent_tree_szad_remove(&dss_chunks_szad, node); + node->addr = prev->addr; + node->size += prev->size; + extent_tree_szad_insert(&dss_chunks_szad, node); + + base_node_dealloc(prev); + } + + return (node); +} + +static bool +chunk_dealloc_dss(void *chunk, size_t size) +{ + + malloc_mutex_lock(&dss_mtx); + if ((uintptr_t)chunk >= (uintptr_t)dss_base + && (uintptr_t)chunk < (uintptr_t)dss_max) { + extent_node_t *node; + + /* Try to coalesce with other unused chunks. */ + node = chunk_dealloc_dss_record(chunk, size); + if (node != NULL) { + chunk = node->addr; + size = node->size; + } + + /* Get the current end of the DSS. */ + dss_max = sbrk(0); + + /* + * Try to shrink the DSS if this chunk is at the end of the + * DSS. The sbrk() call here is subject to a race condition + * with threads that use brk(2) or sbrk(2) directly, but the + * alternative would be to leak memory for the sake of poorly + * designed multi-threaded programs. + */ + if ((void *)((uintptr_t)chunk + size) == dss_max + && (dss_prev = sbrk(-(intptr_t)size)) == dss_max) { + /* Success. */ + dss_max = (void *)((intptr_t)dss_prev - (intptr_t)size); + + if (node != NULL) { + extent_tree_szad_remove(&dss_chunks_szad, node); + extent_tree_ad_remove(&dss_chunks_ad, node); + base_node_dealloc(node); + } + malloc_mutex_unlock(&dss_mtx); + } else { + malloc_mutex_unlock(&dss_mtx); + madvise(chunk, size, MADV_FREE); + } + + return (false); + } + malloc_mutex_unlock(&dss_mtx); + + return (true); +} +#endif + +static void +chunk_dealloc_mmap(void *chunk, size_t size) +{ + + pages_unmap(chunk, size); +} + +static void +chunk_dealloc(void *chunk, size_t size) +{ + + assert(chunk != NULL); + assert(CHUNK_ADDR2BASE(chunk) == chunk); + assert(size != 0); + assert((size & chunksize_mask) == 0); + +#ifdef MALLOC_STATS + stats_chunks.curchunks -= (size / chunksize); +#endif + +#ifdef MALLOC_DSS + if (opt_dss) { + if (chunk_dealloc_dss(chunk, size) == false) + return; + } + + if (opt_mmap) +#endif + chunk_dealloc_mmap(chunk, size); +} + +/* + * End chunk management functions. + */ +/******************************************************************************/ +/* + * Begin arena. + */ + +/* + * Choose an arena based on a per-thread value (fast-path code, calls slow-path + * code if necessary). + */ +static inline arena_t * +choose_arena(void) +{ + arena_t *ret; + + /* + * We can only use TLS if this is a PIC library, since for the static + * library version, libc's malloc is used by TLS allocation, which + * introduces a bootstrapping issue. + */ +#ifndef NO_TLS + if (__isthreaded == false) { + /* Avoid the overhead of TLS for single-threaded operation. */ + return (arenas[0]); + } + + ret = arenas_map; + if (ret == NULL) { + ret = choose_arena_hard(); + assert(ret != NULL); + } +#else + if (__isthreaded && narenas > 1) { + unsigned long ind; + + /* + * Hash _pthread_self() to one of the arenas. There is a prime + * number of arenas, so this has a reasonable chance of + * working. Even so, the hashing can be easily thwarted by + * inconvenient _pthread_self() values. Without specific + * knowledge of how _pthread_self() calculates values, we can't + * easily do much better than this. + */ + ind = (unsigned long) _pthread_self() % narenas; + + /* + * Optimistially assume that arenas[ind] has been initialized. + * At worst, we find out that some other thread has already + * done so, after acquiring the lock in preparation. Note that + * this lazy locking also has the effect of lazily forcing + * cache coherency; without the lock acquisition, there's no + * guarantee that modification of arenas[ind] by another thread + * would be seen on this CPU for an arbitrary amount of time. + * + * In general, this approach to modifying a synchronized value + * isn't a good idea, but in this case we only ever modify the + * value once, so things work out well. + */ + ret = arenas[ind]; + if (ret == NULL) { + /* + * Avoid races with another thread that may have already + * initialized arenas[ind]. + */ + malloc_spin_lock(&arenas_lock); + if (arenas[ind] == NULL) + ret = arenas_extend((unsigned)ind); + else + ret = arenas[ind]; + malloc_spin_unlock(&arenas_lock); + } + } else + ret = arenas[0]; +#endif + + assert(ret != NULL); + return (ret); +} + +#ifndef NO_TLS +/* + * Choose an arena based on a per-thread value (slow-path code only, called + * only by choose_arena()). + */ +static arena_t * +choose_arena_hard(void) +{ + arena_t *ret; + + assert(__isthreaded); + +#ifdef MALLOC_BALANCE + /* Seed the PRNG used for arena load balancing. */ + SPRN(balance, (uint32_t)(uintptr_t)(_pthread_self())); +#endif + + if (narenas > 1) { +#ifdef MALLOC_BALANCE + unsigned ind; + + ind = PRN(balance, narenas_2pow); + if ((ret = arenas[ind]) == NULL) { + malloc_spin_lock(&arenas_lock); + if ((ret = arenas[ind]) == NULL) + ret = arenas_extend(ind); + malloc_spin_unlock(&arenas_lock); + } +#else + malloc_spin_lock(&arenas_lock); + if ((ret = arenas[next_arena]) == NULL) + ret = arenas_extend(next_arena); + next_arena = (next_arena + 1) % narenas; + malloc_spin_unlock(&arenas_lock); +#endif + } else + ret = arenas[0]; + + arenas_map = ret; + + return (ret); +} +#endif + +static inline int +arena_chunk_comp(arena_chunk_t *a, arena_chunk_t *b) +{ + uintptr_t a_chunk = (uintptr_t)a; + uintptr_t b_chunk = (uintptr_t)b; + + assert(a != NULL); + assert(b != NULL); + + return ((a_chunk > b_chunk) - (a_chunk < b_chunk)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(__unused static, arena_chunk_tree_dirty_, arena_chunk_tree_t, + arena_chunk_t, link_dirty, arena_chunk_comp) + +static inline int +arena_run_comp(arena_chunk_map_t *a, arena_chunk_map_t *b) +{ + uintptr_t a_mapelm = (uintptr_t)a; + uintptr_t b_mapelm = (uintptr_t)b; + + assert(a != NULL); + assert(b != NULL); + + return ((a_mapelm > b_mapelm) - (a_mapelm < b_mapelm)); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(__unused static, arena_run_tree_, arena_run_tree_t, arena_chunk_map_t, + link, arena_run_comp) + +static inline int +arena_avail_comp(arena_chunk_map_t *a, arena_chunk_map_t *b) +{ + int ret; + size_t a_size = a->bits & ~pagesize_mask; + size_t b_size = b->bits & ~pagesize_mask; + + ret = (a_size > b_size) - (a_size < b_size); + if (ret == 0) { + uintptr_t a_mapelm, b_mapelm; + + if ((a->bits & CHUNK_MAP_KEY) == 0) + a_mapelm = (uintptr_t)a; + else { + /* + * Treat keys as though they are lower than anything + * else. + */ + a_mapelm = 0; + } + b_mapelm = (uintptr_t)b; + + ret = (a_mapelm > b_mapelm) - (a_mapelm < b_mapelm); + } + + return (ret); +} + +/* Wrap red-black tree macros in functions. */ +rb_wrap(__unused static, arena_avail_tree_, arena_avail_tree_t, + arena_chunk_map_t, link, arena_avail_comp) + +static inline void * +arena_run_reg_alloc(arena_run_t *run, arena_bin_t *bin) +{ + void *ret; + unsigned i, mask, bit, regind; + + assert(run->magic == ARENA_RUN_MAGIC); + assert(run->regs_minelm < bin->regs_mask_nelms); + + /* + * Move the first check outside the loop, so that run->regs_minelm can + * be updated unconditionally, without the possibility of updating it + * multiple times. + */ + i = run->regs_minelm; + mask = run->regs_mask[i]; + if (mask != 0) { + /* Usable allocation found. */ + bit = ffs((int)mask) - 1; + + regind = ((i << (SIZEOF_INT_2POW + 3)) + bit); + assert(regind < bin->nregs); + ret = (void *)(((uintptr_t)run) + bin->reg0_offset + + (bin->reg_size * regind)); + + /* Clear bit. */ + mask ^= (1U << bit); + run->regs_mask[i] = mask; + + return (ret); + } + + for (i++; i < bin->regs_mask_nelms; i++) { + mask = run->regs_mask[i]; + if (mask != 0) { + /* Usable allocation found. */ + bit = ffs((int)mask) - 1; + + regind = ((i << (SIZEOF_INT_2POW + 3)) + bit); + assert(regind < bin->nregs); + ret = (void *)(((uintptr_t)run) + bin->reg0_offset + + (bin->reg_size * regind)); + + /* Clear bit. */ + mask ^= (1U << bit); + run->regs_mask[i] = mask; + + /* + * Make a note that nothing before this element + * contains a free region. + */ + run->regs_minelm = i; /* Low payoff: + (mask == 0); */ + + return (ret); + } + } + /* Not reached. */ + assert(0); + return (NULL); +} + +static inline void +arena_run_reg_dalloc(arena_run_t *run, arena_bin_t *bin, void *ptr, size_t size) +{ + unsigned diff, regind, elm, bit; + + assert(run->magic == ARENA_RUN_MAGIC); + + /* + * Avoid doing division with a variable divisor if possible. Using + * actual division here can reduce allocator throughput by over 20%! + */ + diff = (unsigned)((uintptr_t)ptr - (uintptr_t)run - bin->reg0_offset); + if ((size & (size - 1)) == 0) { + /* + * log2_table allows fast division of a power of two in the + * [1..128] range. + * + * (x / divisor) becomes (x >> log2_table[divisor - 1]). + */ + static const unsigned char log2_table[] = { + 0, 1, 0, 2, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 4, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7 + }; + + if (size <= 128) + regind = (diff >> log2_table[size - 1]); + else if (size <= 32768) + regind = diff >> (8 + log2_table[(size >> 8) - 1]); + else + regind = diff / size; + } else if (size < qspace_max) { + /* + * To divide by a number D that is not a power of two we + * multiply by (2^21 / D) and then right shift by 21 positions. + * + * X / D + * + * becomes + * + * (X * qsize_invs[(D >> QUANTUM_2POW) - 3]) + * >> SIZE_INV_SHIFT + * + * We can omit the first three elements, because we never + * divide by 0, and QUANTUM and 2*QUANTUM are both powers of + * two, which are handled above. + */ +#define SIZE_INV_SHIFT 21 +#define QSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << QUANTUM_2POW)) + 1) + static const unsigned qsize_invs[] = { + QSIZE_INV(3), + QSIZE_INV(4), QSIZE_INV(5), QSIZE_INV(6), QSIZE_INV(7) +#if (QUANTUM_2POW < 4) + , + QSIZE_INV(8), QSIZE_INV(9), QSIZE_INV(10), QSIZE_INV(11), + QSIZE_INV(12),QSIZE_INV(13), QSIZE_INV(14), QSIZE_INV(15) +#endif + }; + assert(QUANTUM * (((sizeof(qsize_invs)) / sizeof(unsigned)) + 3) + >= (1U << QSPACE_MAX_2POW_DEFAULT)); + + if (size <= (((sizeof(qsize_invs) / sizeof(unsigned)) + 2) << + QUANTUM_2POW)) { + regind = qsize_invs[(size >> QUANTUM_2POW) - 3] * diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef QSIZE_INV + } else if (size < cspace_max) { +#define CSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << CACHELINE_2POW)) + 1) + static const unsigned csize_invs[] = { + CSIZE_INV(3), + CSIZE_INV(4), CSIZE_INV(5), CSIZE_INV(6), CSIZE_INV(7) + }; + assert(CACHELINE * (((sizeof(csize_invs)) / sizeof(unsigned)) + + 3) >= (1U << CSPACE_MAX_2POW_DEFAULT)); + + if (size <= (((sizeof(csize_invs) / sizeof(unsigned)) + 2) << + CACHELINE_2POW)) { + regind = csize_invs[(size >> CACHELINE_2POW) - 3] * + diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef CSIZE_INV + } else { +#define SSIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << SUBPAGE_2POW)) + 1) + static const unsigned ssize_invs[] = { + SSIZE_INV(3), + SSIZE_INV(4), SSIZE_INV(5), SSIZE_INV(6), SSIZE_INV(7), + SSIZE_INV(8), SSIZE_INV(9), SSIZE_INV(10), SSIZE_INV(11), + SSIZE_INV(12), SSIZE_INV(13), SSIZE_INV(14), SSIZE_INV(15) +#if (PAGESIZE_2POW == 13) + , + SSIZE_INV(16), SSIZE_INV(17), SSIZE_INV(18), SSIZE_INV(19), + SSIZE_INV(20), SSIZE_INV(21), SSIZE_INV(22), SSIZE_INV(23), + SSIZE_INV(24), SSIZE_INV(25), SSIZE_INV(26), SSIZE_INV(27), + SSIZE_INV(28), SSIZE_INV(29), SSIZE_INV(29), SSIZE_INV(30) +#endif + }; + assert(SUBPAGE * (((sizeof(ssize_invs)) / sizeof(unsigned)) + 3) + >= (1U << PAGESIZE_2POW)); + + if (size < (((sizeof(ssize_invs) / sizeof(unsigned)) + 2) << + SUBPAGE_2POW)) { + regind = ssize_invs[(size >> SUBPAGE_2POW) - 3] * diff; + regind >>= SIZE_INV_SHIFT; + } else + regind = diff / size; +#undef SSIZE_INV + } +#undef SIZE_INV_SHIFT + assert(diff == regind * size); + assert(regind < bin->nregs); + + elm = regind >> (SIZEOF_INT_2POW + 3); + if (elm < run->regs_minelm) + run->regs_minelm = elm; + bit = regind - (elm << (SIZEOF_INT_2POW + 3)); + assert((run->regs_mask[elm] & (1U << bit)) == 0); + run->regs_mask[elm] |= (1U << bit); +} + +static void +arena_run_split(arena_t *arena, arena_run_t *run, size_t size, bool large, + bool zero) +{ + arena_chunk_t *chunk; + size_t old_ndirty, run_ind, total_pages, need_pages, rem_pages, i; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run); + old_ndirty = chunk->ndirty; + run_ind = (unsigned)(((uintptr_t)run - (uintptr_t)chunk) + >> pagesize_2pow); + total_pages = (chunk->map[run_ind].bits & ~pagesize_mask) >> + pagesize_2pow; + need_pages = (size >> pagesize_2pow); + assert(need_pages > 0); + assert(need_pages <= total_pages); + rem_pages = total_pages - need_pages; + + arena_avail_tree_remove(&arena->runs_avail, &chunk->map[run_ind]); + + /* Keep track of trailing unused pages for later use. */ + if (rem_pages > 0) { + chunk->map[run_ind+need_pages].bits = (rem_pages << + pagesize_2pow) | (chunk->map[run_ind+need_pages].bits & + pagesize_mask); + chunk->map[run_ind+total_pages-1].bits = (rem_pages << + pagesize_2pow) | (chunk->map[run_ind+total_pages-1].bits & + pagesize_mask); + arena_avail_tree_insert(&arena->runs_avail, + &chunk->map[run_ind+need_pages]); + } + + for (i = 0; i < need_pages; i++) { + /* Zero if necessary. */ + if (zero) { + if ((chunk->map[run_ind + i].bits & CHUNK_MAP_ZEROED) + == 0) { + memset((void *)((uintptr_t)chunk + ((run_ind + + i) << pagesize_2pow)), 0, pagesize); + /* CHUNK_MAP_ZEROED is cleared below. */ + } + } + + /* Update dirty page accounting. */ + if (chunk->map[run_ind + i].bits & CHUNK_MAP_DIRTY) { + chunk->ndirty--; + arena->ndirty--; + /* CHUNK_MAP_DIRTY is cleared below. */ + } + + /* Initialize the chunk map. */ + if (large) { + chunk->map[run_ind + i].bits = CHUNK_MAP_LARGE + | CHUNK_MAP_ALLOCATED; + } else { + chunk->map[run_ind + i].bits = (size_t)run + | CHUNK_MAP_ALLOCATED; + } + } + + /* + * Set the run size only in the first element for large runs. This is + * primarily a debugging aid, since the lack of size info for trailing + * pages only matters if the application tries to operate on an + * interior pointer. + */ + if (large) + chunk->map[run_ind].bits |= size; + + if (chunk->ndirty == 0 && old_ndirty > 0) + arena_chunk_tree_dirty_remove(&arena->chunks_dirty, chunk); +} + +static arena_chunk_t * +arena_chunk_alloc(arena_t *arena) +{ + arena_chunk_t *chunk; + size_t i; + + if (arena->spare != NULL) { + chunk = arena->spare; + arena->spare = NULL; + } else { + chunk = (arena_chunk_t *)chunk_alloc(chunksize, true); + if (chunk == NULL) + return (NULL); +#ifdef MALLOC_STATS + arena->stats.mapped += chunksize; +#endif + + chunk->arena = arena; + + /* + * Claim that no pages are in use, since the header is merely + * overhead. + */ + chunk->ndirty = 0; + + /* + * Initialize the map to contain one maximal free untouched run. + */ + for (i = 0; i < arena_chunk_header_npages; i++) + chunk->map[i].bits = 0; + chunk->map[i].bits = arena_maxclass | CHUNK_MAP_ZEROED; + for (i++; i < chunk_npages-1; i++) { + chunk->map[i].bits = CHUNK_MAP_ZEROED; + } + chunk->map[chunk_npages-1].bits = arena_maxclass | + CHUNK_MAP_ZEROED; + } + + /* Insert the run into the runs_avail tree. */ + arena_avail_tree_insert(&arena->runs_avail, + &chunk->map[arena_chunk_header_npages]); + + return (chunk); +} + +static void +arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk) +{ + + if (arena->spare != NULL) { + if (arena->spare->ndirty > 0) { + arena_chunk_tree_dirty_remove( + &chunk->arena->chunks_dirty, arena->spare); + arena->ndirty -= arena->spare->ndirty; + } + chunk_dealloc((void *)arena->spare, chunksize); +#ifdef MALLOC_STATS + arena->stats.mapped -= chunksize; +#endif + } + + /* + * Remove run from runs_avail, regardless of whether this chunk + * will be cached, so that the arena does not use it. Dirty page + * flushing only uses the chunks_dirty tree, so leaving this chunk in + * the chunks_* trees is sufficient for that purpose. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[arena_chunk_header_npages]); + + arena->spare = chunk; +} + +static arena_run_t * +arena_run_alloc(arena_t *arena, size_t size, bool large, bool zero) +{ + arena_chunk_t *chunk; + arena_run_t *run; + arena_chunk_map_t *mapelm, key; + + assert(size <= arena_maxclass); + assert((size & pagesize_mask) == 0); + + /* Search the arena's chunks for the lowest best fit. */ + key.bits = size | CHUNK_MAP_KEY; + mapelm = arena_avail_tree_nsearch(&arena->runs_avail, &key); + if (mapelm != NULL) { + arena_chunk_t *run_chunk = CHUNK_ADDR2BASE(mapelm); + size_t pageind = ((uintptr_t)mapelm - (uintptr_t)run_chunk->map) + / sizeof(arena_chunk_map_t); + + run = (arena_run_t *)((uintptr_t)run_chunk + (pageind + << pagesize_2pow)); + arena_run_split(arena, run, size, large, zero); + return (run); + } + + /* + * No usable runs. Create a new chunk from which to allocate the run. + */ + chunk = arena_chunk_alloc(arena); + if (chunk == NULL) + return (NULL); + run = (arena_run_t *)((uintptr_t)chunk + (arena_chunk_header_npages << + pagesize_2pow)); + /* Update page map. */ + arena_run_split(arena, run, size, large, zero); + return (run); +} + +static void +arena_purge(arena_t *arena) +{ + arena_chunk_t *chunk; + size_t i, npages; +#ifdef MALLOC_DEBUG + size_t ndirty = 0; + + rb_foreach_begin(arena_chunk_t, link_dirty, &arena->chunks_dirty, + chunk) { + ndirty += chunk->ndirty; + } rb_foreach_end(arena_chunk_t, link_dirty, &arena->chunks_dirty, chunk) + assert(ndirty == arena->ndirty); +#endif + assert(arena->ndirty > opt_dirty_max); + +#ifdef MALLOC_STATS + arena->stats.npurge++; +#endif + + /* + * Iterate downward through chunks until enough dirty memory has been + * purged. Terminate as soon as possible in order to minimize the + * number of system calls, even if a chunk has only been partially + * purged. + */ + while (arena->ndirty > (opt_dirty_max >> 1)) { + chunk = arena_chunk_tree_dirty_last(&arena->chunks_dirty); + assert(chunk != NULL); + + for (i = chunk_npages - 1; chunk->ndirty > 0; i--) { + assert(i >= arena_chunk_header_npages); + + if (chunk->map[i].bits & CHUNK_MAP_DIRTY) { + chunk->map[i].bits ^= CHUNK_MAP_DIRTY; + /* Find adjacent dirty run(s). */ + for (npages = 1; i > arena_chunk_header_npages + && (chunk->map[i - 1].bits & + CHUNK_MAP_DIRTY); npages++) { + i--; + chunk->map[i].bits ^= CHUNK_MAP_DIRTY; + } + chunk->ndirty -= npages; + arena->ndirty -= npages; + + madvise((void *)((uintptr_t)chunk + (i << + pagesize_2pow)), (npages << pagesize_2pow), + MADV_FREE); +#ifdef MALLOC_STATS + arena->stats.nmadvise++; + arena->stats.purged += npages; +#endif + if (arena->ndirty <= (opt_dirty_max >> 1)) + break; + } + } + + if (chunk->ndirty == 0) { + arena_chunk_tree_dirty_remove(&arena->chunks_dirty, + chunk); + } + } +} + +static void +arena_run_dalloc(arena_t *arena, arena_run_t *run, bool dirty) +{ + arena_chunk_t *chunk; + size_t size, run_ind, run_pages; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run); + run_ind = (size_t)(((uintptr_t)run - (uintptr_t)chunk) + >> pagesize_2pow); + assert(run_ind >= arena_chunk_header_npages); + assert(run_ind < chunk_npages); + if ((chunk->map[run_ind].bits & CHUNK_MAP_LARGE) != 0) + size = chunk->map[run_ind].bits & ~pagesize_mask; + else + size = run->bin->run_size; + run_pages = (size >> pagesize_2pow); + + /* Mark pages as unallocated in the chunk map. */ + if (dirty) { + size_t i; + + for (i = 0; i < run_pages; i++) { + assert((chunk->map[run_ind + i].bits & CHUNK_MAP_DIRTY) + == 0); + chunk->map[run_ind + i].bits = CHUNK_MAP_DIRTY; + } + + if (chunk->ndirty == 0) { + arena_chunk_tree_dirty_insert(&arena->chunks_dirty, + chunk); + } + chunk->ndirty += run_pages; + arena->ndirty += run_pages; + } else { + size_t i; + + for (i = 0; i < run_pages; i++) { + chunk->map[run_ind + i].bits &= ~(CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED); + } + } + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + + /* Try to coalesce forward. */ + if (run_ind + run_pages < chunk_npages && + (chunk->map[run_ind+run_pages].bits & CHUNK_MAP_ALLOCATED) == 0) { + size_t nrun_size = chunk->map[run_ind+run_pages].bits & + ~pagesize_mask; + + /* + * Remove successor from runs_avail; the coalesced run is + * inserted later. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[run_ind+run_pages]); + + size += nrun_size; + run_pages = size >> pagesize_2pow; + + assert((chunk->map[run_ind+run_pages-1].bits & ~pagesize_mask) + == nrun_size); + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + } + + /* Try to coalesce backward. */ + if (run_ind > arena_chunk_header_npages && (chunk->map[run_ind-1].bits & + CHUNK_MAP_ALLOCATED) == 0) { + size_t prun_size = chunk->map[run_ind-1].bits & ~pagesize_mask; + + run_ind -= prun_size >> pagesize_2pow; + + /* + * Remove predecessor from runs_avail; the coalesced run is + * inserted later. + */ + arena_avail_tree_remove(&arena->runs_avail, + &chunk->map[run_ind]); + + size += prun_size; + run_pages = size >> pagesize_2pow; + + assert((chunk->map[run_ind].bits & ~pagesize_mask) == + prun_size); + chunk->map[run_ind].bits = size | (chunk->map[run_ind].bits & + pagesize_mask); + chunk->map[run_ind+run_pages-1].bits = size | + (chunk->map[run_ind+run_pages-1].bits & pagesize_mask); + } + + /* Insert into runs_avail, now that coalescing is complete. */ + arena_avail_tree_insert(&arena->runs_avail, &chunk->map[run_ind]); + + /* Deallocate chunk if it is now completely unused. */ + if ((chunk->map[arena_chunk_header_npages].bits & (~pagesize_mask | + CHUNK_MAP_ALLOCATED)) == arena_maxclass) + arena_chunk_dealloc(arena, chunk); + + /* Enforce opt_dirty_max. */ + if (arena->ndirty > opt_dirty_max) + arena_purge(arena); +} + +static void +arena_run_trim_head(arena_t *arena, arena_chunk_t *chunk, arena_run_t *run, + size_t oldsize, size_t newsize) +{ + size_t pageind = ((uintptr_t)run - (uintptr_t)chunk) >> pagesize_2pow; + size_t head_npages = (oldsize - newsize) >> pagesize_2pow; + + assert(oldsize > newsize); + + /* + * Update the chunk map so that arena_run_dalloc() can treat the + * leading run as separately allocated. + */ + chunk->map[pageind].bits = (oldsize - newsize) | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+head_npages].bits = newsize | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + + arena_run_dalloc(arena, run, false); +} + +static void +arena_run_trim_tail(arena_t *arena, arena_chunk_t *chunk, arena_run_t *run, + size_t oldsize, size_t newsize, bool dirty) +{ + size_t pageind = ((uintptr_t)run - (uintptr_t)chunk) >> pagesize_2pow; + size_t npages = newsize >> pagesize_2pow; + + assert(oldsize > newsize); + + /* + * Update the chunk map so that arena_run_dalloc() can treat the + * trailing run as separately allocated. + */ + chunk->map[pageind].bits = newsize | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+npages].bits = (oldsize - newsize) | CHUNK_MAP_LARGE + | CHUNK_MAP_ALLOCATED; + + arena_run_dalloc(arena, (arena_run_t *)((uintptr_t)run + newsize), + dirty); +} + +static arena_run_t * +arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin) +{ + arena_chunk_map_t *mapelm; + arena_run_t *run; + unsigned i, remainder; + + /* Look for a usable run. */ + mapelm = arena_run_tree_first(&bin->runs); + if (mapelm != NULL) { + /* run is guaranteed to have available space. */ + arena_run_tree_remove(&bin->runs, mapelm); + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); +#ifdef MALLOC_STATS + bin->stats.reruns++; +#endif + return (run); + } + /* No existing runs have any space available. */ + + /* Allocate a new run. */ + run = arena_run_alloc(arena, bin->run_size, false, false); + if (run == NULL) + return (NULL); + + /* Initialize run internals. */ + run->bin = bin; + + for (i = 0; i < bin->regs_mask_nelms - 1; i++) + run->regs_mask[i] = UINT_MAX; + remainder = bin->nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1); + if (remainder == 0) + run->regs_mask[i] = UINT_MAX; + else { + /* The last element has spare bits that need to be unset. */ + run->regs_mask[i] = (UINT_MAX >> ((1U << (SIZEOF_INT_2POW + 3)) + - remainder)); + } + + run->regs_minelm = 0; + + run->nfree = bin->nregs; +#ifdef MALLOC_DEBUG + run->magic = ARENA_RUN_MAGIC; +#endif + +#ifdef MALLOC_STATS + bin->stats.nruns++; + bin->stats.curruns++; + if (bin->stats.curruns > bin->stats.highruns) + bin->stats.highruns = bin->stats.curruns; +#endif + return (run); +} + +/* bin->runcur must have space available before this function is called. */ +static inline void * +arena_bin_malloc_easy(arena_t *arena, arena_bin_t *bin, arena_run_t *run) +{ + void *ret; + + assert(run->magic == ARENA_RUN_MAGIC); + assert(run->nfree > 0); + + ret = arena_run_reg_alloc(run, bin); + assert(ret != NULL); + run->nfree--; + + return (ret); +} + +/* Re-fill bin->runcur, then call arena_bin_malloc_easy(). */ +static void * +arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin) +{ + + bin->runcur = arena_bin_nonfull_run_get(arena, bin); + if (bin->runcur == NULL) + return (NULL); + assert(bin->runcur->magic == ARENA_RUN_MAGIC); + assert(bin->runcur->nfree > 0); + + return (arena_bin_malloc_easy(arena, bin, bin->runcur)); +} + +/* + * Calculate bin->run_size such that it meets the following constraints: + * + * *) bin->run_size >= min_run_size + * *) bin->run_size <= arena_maxclass + * *) bin->run_size <= RUN_MAX_SMALL + * *) run header overhead <= RUN_MAX_OVRHD (or header overhead relaxed). + * + * bin->nregs, bin->regs_mask_nelms, and bin->reg0_offset are + * also calculated here, since these settings are all interdependent. + */ +static size_t +arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size) +{ + size_t try_run_size, good_run_size; + unsigned good_nregs, good_mask_nelms, good_reg0_offset; + unsigned try_nregs, try_mask_nelms, try_reg0_offset; + + assert(min_run_size >= pagesize); + assert(min_run_size <= arena_maxclass); + assert(min_run_size <= RUN_MAX_SMALL); + + /* + * Calculate known-valid settings before entering the run_size + * expansion loop, so that the first part of the loop always copies + * valid settings. + * + * The do..while loop iteratively reduces the number of regions until + * the run header and the regions no longer overlap. A closed formula + * would be quite messy, since there is an interdependency between the + * header's mask length and the number of regions. + */ + try_run_size = min_run_size; + try_nregs = ((try_run_size - sizeof(arena_run_t)) / bin->reg_size) + + 1; /* Counter-act try_nregs-- in loop. */ + do { + try_nregs--; + try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) + + ((try_nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1)) ? 1 : 0); + try_reg0_offset = try_run_size - (try_nregs * bin->reg_size); + } while (sizeof(arena_run_t) + (sizeof(unsigned) * (try_mask_nelms - 1)) + > try_reg0_offset); + + /* run_size expansion loop. */ + do { + /* + * Copy valid settings before trying more aggressive settings. + */ + good_run_size = try_run_size; + good_nregs = try_nregs; + good_mask_nelms = try_mask_nelms; + good_reg0_offset = try_reg0_offset; + + /* Try more aggressive settings. */ + try_run_size += pagesize; + try_nregs = ((try_run_size - sizeof(arena_run_t)) / + bin->reg_size) + 1; /* Counter-act try_nregs-- in loop. */ + do { + try_nregs--; + try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) + + ((try_nregs & ((1U << (SIZEOF_INT_2POW + 3)) - 1)) ? + 1 : 0); + try_reg0_offset = try_run_size - (try_nregs * + bin->reg_size); + } while (sizeof(arena_run_t) + (sizeof(unsigned) * + (try_mask_nelms - 1)) > try_reg0_offset); + } while (try_run_size <= arena_maxclass && try_run_size <= RUN_MAX_SMALL + && RUN_MAX_OVRHD * (bin->reg_size << 3) > RUN_MAX_OVRHD_RELAX + && (try_reg0_offset << RUN_BFP) > RUN_MAX_OVRHD * try_run_size); + + assert(sizeof(arena_run_t) + (sizeof(unsigned) * (good_mask_nelms - 1)) + <= good_reg0_offset); + assert((good_mask_nelms << (SIZEOF_INT_2POW + 3)) >= good_nregs); + + /* Copy final settings. */ + bin->run_size = good_run_size; + bin->nregs = good_nregs; + bin->regs_mask_nelms = good_mask_nelms; + bin->reg0_offset = good_reg0_offset; + + return (good_run_size); +} + +#ifdef MALLOC_BALANCE +static inline void +arena_lock_balance(arena_t *arena) +{ + unsigned contention; + + contention = malloc_spin_lock(&arena->lock); + if (narenas > 1) { + /* + * Calculate the exponentially averaged contention for this + * arena. Due to integer math always rounding down, this value + * decays somewhat faster than normal. + */ + arena->contention = (((uint64_t)arena->contention + * (uint64_t)((1U << BALANCE_ALPHA_INV_2POW)-1)) + + (uint64_t)contention) >> BALANCE_ALPHA_INV_2POW; + if (arena->contention >= opt_balance_threshold) + arena_lock_balance_hard(arena); + } +} + +static void +arena_lock_balance_hard(arena_t *arena) +{ + uint32_t ind; + + arena->contention = 0; +#ifdef MALLOC_STATS + arena->stats.nbalance++; +#endif + ind = PRN(balance, narenas_2pow); + if (arenas[ind] != NULL) + arenas_map = arenas[ind]; + else { + malloc_spin_lock(&arenas_lock); + if (arenas[ind] != NULL) + arenas_map = arenas[ind]; + else + arenas_map = arenas_extend(ind); + malloc_spin_unlock(&arenas_lock); + } +} +#endif + +#ifdef MALLOC_MAG +static inline void * +mag_alloc(mag_t *mag) +{ + + if (mag->nrounds == 0) + return (NULL); + mag->nrounds--; + + return (mag->rounds[mag->nrounds]); +} + +static void +mag_load(mag_t *mag) +{ + arena_t *arena; + arena_bin_t *bin; + arena_run_t *run; + void *round; + size_t i; + + arena = choose_arena(); + bin = &arena->bins[mag->binind]; +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + for (i = mag->nrounds; i < max_rounds; i++) { + if ((run = bin->runcur) != NULL && run->nfree > 0) + round = arena_bin_malloc_easy(arena, bin, run); + else + round = arena_bin_malloc_hard(arena, bin); + if (round == NULL) + break; + mag->rounds[i] = round; + } +#ifdef MALLOC_STATS + bin->stats.nmags++; + arena->stats.nmalloc_small += (i - mag->nrounds); + arena->stats.allocated_small += (i - mag->nrounds) * bin->reg_size; +#endif + malloc_spin_unlock(&arena->lock); + mag->nrounds = i; +} + +static inline void * +mag_rack_alloc(mag_rack_t *rack, size_t size, bool zero) +{ + void *ret; + bin_mags_t *bin_mags; + mag_t *mag; + size_t binind; + + binind = size2bin[size]; + assert(binind < nbins); + bin_mags = &rack->bin_mags[binind]; + + mag = bin_mags->curmag; + if (mag == NULL) { + /* Create an initial magazine for this size class. */ + assert(bin_mags->sparemag == NULL); + mag = mag_create(choose_arena(), binind); + if (mag == NULL) + return (NULL); + bin_mags->curmag = mag; + mag_load(mag); + } + + ret = mag_alloc(mag); + if (ret == NULL) { + if (bin_mags->sparemag != NULL) { + if (bin_mags->sparemag->nrounds > 0) { + /* Swap magazines. */ + bin_mags->curmag = bin_mags->sparemag; + bin_mags->sparemag = mag; + mag = bin_mags->curmag; + } else { + /* Reload the current magazine. */ + mag_load(mag); + } + } else { + /* Create a second magazine. */ + mag = mag_create(choose_arena(), binind); + if (mag == NULL) + return (NULL); + mag_load(mag); + bin_mags->sparemag = bin_mags->curmag; + bin_mags->curmag = mag; + } + ret = mag_alloc(mag); + if (ret == NULL) + return (NULL); + } + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } else + memset(ret, 0, size); + + return (ret); +} +#endif + +static inline void * +arena_malloc_small(arena_t *arena, size_t size, bool zero) +{ + void *ret; + arena_bin_t *bin; + arena_run_t *run; + size_t binind; + + binind = size2bin[size]; + assert(binind < nbins); + bin = &arena->bins[binind]; + size = bin->reg_size; + +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + if ((run = bin->runcur) != NULL && run->nfree > 0) + ret = arena_bin_malloc_easy(arena, bin, run); + else + ret = arena_bin_malloc_hard(arena, bin); + + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } + +#ifdef MALLOC_STATS + bin->stats.nrequests++; + arena->stats.nmalloc_small++; + arena->stats.allocated_small += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } else + memset(ret, 0, size); + + return (ret); +} + +static void * +arena_malloc_large(arena_t *arena, size_t size, bool zero) +{ + void *ret; + + /* Large allocation. */ + size = PAGE_CEILING(size); +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + ret = (void *)arena_run_alloc(arena, size, true, zero); + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } +#ifdef MALLOC_STATS + arena->stats.nmalloc_large++; + arena->stats.allocated_large += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + } + + return (ret); +} + +static inline void * +arena_malloc(arena_t *arena, size_t size, bool zero) +{ + + assert(arena != NULL); + assert(arena->magic == ARENA_MAGIC); + assert(size != 0); + assert(QUANTUM_CEILING(size) <= arena_maxclass); + + if (size <= bin_maxclass) { +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + mag_rack_t *rack = mag_rack; + if (rack == NULL) { + rack = mag_rack_create(arena); + if (rack == NULL) + return (NULL); + mag_rack = rack; + } + return (mag_rack_alloc(rack, size, zero)); + } else +#endif + return (arena_malloc_small(arena, size, zero)); + } else + return (arena_malloc_large(arena, size, zero)); +} + +static inline void * +imalloc(size_t size) +{ + + assert(size != 0); + + if (size <= arena_maxclass) + return (arena_malloc(choose_arena(), size, false)); + else + return (huge_malloc(size, false)); +} + +static inline void * +icalloc(size_t size) +{ + + if (size <= arena_maxclass) + return (arena_malloc(choose_arena(), size, true)); + else + return (huge_malloc(size, true)); +} + +/* Only handles large allocations that require more than page alignment. */ +static void * +arena_palloc(arena_t *arena, size_t alignment, size_t size, size_t alloc_size) +{ + void *ret; + size_t offset; + arena_chunk_t *chunk; + + assert((size & pagesize_mask) == 0); + assert((alignment & pagesize_mask) == 0); + +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + ret = (void *)arena_run_alloc(arena, alloc_size, true, false); + if (ret == NULL) { + malloc_spin_unlock(&arena->lock); + return (NULL); + } + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ret); + + offset = (uintptr_t)ret & (alignment - 1); + assert((offset & pagesize_mask) == 0); + assert(offset < alloc_size); + if (offset == 0) + arena_run_trim_tail(arena, chunk, ret, alloc_size, size, false); + else { + size_t leadsize, trailsize; + + leadsize = alignment - offset; + if (leadsize > 0) { + arena_run_trim_head(arena, chunk, ret, alloc_size, + alloc_size - leadsize); + ret = (void *)((uintptr_t)ret + leadsize); + } + + trailsize = alloc_size - leadsize - size; + if (trailsize != 0) { + /* Trim trailing space. */ + assert(trailsize < alloc_size); + arena_run_trim_tail(arena, chunk, ret, size + trailsize, + size, false); + } + } + +#ifdef MALLOC_STATS + arena->stats.nmalloc_large++; + arena->stats.allocated_large += size; +#endif + malloc_spin_unlock(&arena->lock); + + if (opt_junk) + memset(ret, 0xa5, size); + else if (opt_zero) + memset(ret, 0, size); + return (ret); +} + +static inline void * +ipalloc(size_t alignment, size_t size) +{ + void *ret; + size_t ceil_size; + + /* + * Round size up to the nearest multiple of alignment. + * + * This done, we can take advantage of the fact that for each small + * size class, every object is aligned at the smallest power of two + * that is non-zero in the base two representation of the size. For + * example: + * + * Size | Base 2 | Minimum alignment + * -----+----------+------------------ + * 96 | 1100000 | 32 + * 144 | 10100000 | 32 + * 192 | 11000000 | 64 + * + * Depending on runtime settings, it is possible that arena_malloc() + * will further round up to a power of two, but that never causes + * correctness issues. + */ + ceil_size = (size + (alignment - 1)) & (-alignment); + /* + * (ceil_size < size) protects against the combination of maximal + * alignment and size greater than maximal alignment. + */ + if (ceil_size < size) { + /* size_t overflow. */ + return (NULL); + } + + if (ceil_size <= pagesize || (alignment <= pagesize + && ceil_size <= arena_maxclass)) + ret = arena_malloc(choose_arena(), ceil_size, false); + else { + size_t run_size; + + /* + * We can't achieve subpage alignment, so round up alignment + * permanently; it makes later calculations simpler. + */ + alignment = PAGE_CEILING(alignment); + ceil_size = PAGE_CEILING(size); + /* + * (ceil_size < size) protects against very large sizes within + * pagesize of SIZE_T_MAX. + * + * (ceil_size + alignment < ceil_size) protects against the + * combination of maximal alignment and ceil_size large enough + * to cause overflow. This is similar to the first overflow + * check above, but it needs to be repeated due to the new + * ceil_size value, which may now be *equal* to maximal + * alignment, whereas before we only detected overflow if the + * original size was *greater* than maximal alignment. + */ + if (ceil_size < size || ceil_size + alignment < ceil_size) { + /* size_t overflow. */ + return (NULL); + } + + /* + * Calculate the size of the over-size run that arena_palloc() + * would need to allocate in order to guarantee the alignment. + */ + if (ceil_size >= alignment) + run_size = ceil_size + alignment - pagesize; + else { + /* + * It is possible that (alignment << 1) will cause + * overflow, but it doesn't matter because we also + * subtract pagesize, which in the case of overflow + * leaves us with a very large run_size. That causes + * the first conditional below to fail, which means + * that the bogus run_size value never gets used for + * anything important. + */ + run_size = (alignment << 1) - pagesize; + } + + if (run_size <= arena_maxclass) { + ret = arena_palloc(choose_arena(), alignment, ceil_size, + run_size); + } else if (alignment <= chunksize) + ret = huge_malloc(ceil_size, false); + else + ret = huge_palloc(alignment, ceil_size); + } + + assert(((uintptr_t)ret & (alignment - 1)) == 0); + return (ret); +} + +/* Return the size of the allocation pointed to by ptr. */ +static size_t +arena_salloc(const void *ptr) +{ + size_t ret; + arena_chunk_t *chunk; + size_t pageind, mapbits; + + assert(ptr != NULL); + assert(CHUNK_ADDR2BASE(ptr) != ptr); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapbits = chunk->map[pageind].bits; + assert((mapbits & CHUNK_MAP_ALLOCATED) != 0); + if ((mapbits & CHUNK_MAP_LARGE) == 0) { + arena_run_t *run = (arena_run_t *)(mapbits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + ret = run->bin->reg_size; + } else { + ret = mapbits & ~pagesize_mask; + assert(ret != 0); + } + + return (ret); +} + +static inline size_t +isalloc(const void *ptr) +{ + size_t ret; + arena_chunk_t *chunk; + + assert(ptr != NULL); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + if (chunk != ptr) { + /* Region. */ + assert(chunk->arena->magic == ARENA_MAGIC); + + ret = arena_salloc(ptr); + } else { + extent_node_t *node, key; + + /* Chunk (huge allocation). */ + + malloc_mutex_lock(&huge_mtx); + + /* Extract from tree of huge allocations. */ + key.addr = __DECONST(void *, ptr); + node = extent_tree_ad_search(&huge, &key); + assert(node != NULL); + + ret = node->size; + + malloc_mutex_unlock(&huge_mtx); + } + + return (ret); +} + +static inline void +arena_dalloc_small(arena_t *arena, arena_chunk_t *chunk, void *ptr, + arena_chunk_map_t *mapelm) +{ + arena_run_t *run; + arena_bin_t *bin; + size_t size; + + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + bin = run->bin; + size = bin->reg_size; + + if (opt_junk) + memset(ptr, 0x5a, size); + + arena_run_reg_dalloc(run, bin, ptr, size); + run->nfree++; + + if (run->nfree == bin->nregs) { + /* Deallocate run. */ + if (run == bin->runcur) + bin->runcur = NULL; + else if (bin->nregs != 1) { + size_t run_pageind = (((uintptr_t)run - + (uintptr_t)chunk)) >> pagesize_2pow; + arena_chunk_map_t *run_mapelm = + &chunk->map[run_pageind]; + /* + * This block's conditional is necessary because if the + * run only contains one region, then it never gets + * inserted into the non-full runs tree. + */ + arena_run_tree_remove(&bin->runs, run_mapelm); + } +#ifdef MALLOC_DEBUG + run->magic = 0; +#endif + arena_run_dalloc(arena, run, true); +#ifdef MALLOC_STATS + bin->stats.curruns--; +#endif + } else if (run->nfree == 1 && run != bin->runcur) { + /* + * Make sure that bin->runcur always refers to the lowest + * non-full run, if one exists. + */ + if (bin->runcur == NULL) + bin->runcur = run; + else if ((uintptr_t)run < (uintptr_t)bin->runcur) { + /* Switch runcur. */ + if (bin->runcur->nfree > 0) { + arena_chunk_t *runcur_chunk = + CHUNK_ADDR2BASE(bin->runcur); + size_t runcur_pageind = + (((uintptr_t)bin->runcur - + (uintptr_t)runcur_chunk)) >> pagesize_2pow; + arena_chunk_map_t *runcur_mapelm = + &runcur_chunk->map[runcur_pageind]; + + /* Insert runcur. */ + arena_run_tree_insert(&bin->runs, + runcur_mapelm); + } + bin->runcur = run; + } else { + size_t run_pageind = (((uintptr_t)run - + (uintptr_t)chunk)) >> pagesize_2pow; + arena_chunk_map_t *run_mapelm = + &chunk->map[run_pageind]; + + assert(arena_run_tree_search(&bin->runs, run_mapelm) == + NULL); + arena_run_tree_insert(&bin->runs, run_mapelm); + } + } +#ifdef MALLOC_STATS + arena->stats.allocated_small -= size; + arena->stats.ndalloc_small++; +#endif +} + +#ifdef MALLOC_MAG +static void +mag_unload(mag_t *mag) +{ + arena_chunk_t *chunk; + arena_t *arena; + void *round; + size_t i, ndeferred, nrounds; + + for (ndeferred = mag->nrounds; ndeferred > 0;) { + nrounds = ndeferred; + /* Lock the arena associated with the first round. */ + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(mag->rounds[0]); + arena = chunk->arena; +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + /* Deallocate every round that belongs to the locked arena. */ + for (i = ndeferred = 0; i < nrounds; i++) { + round = mag->rounds[i]; + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(round); + if (chunk->arena == arena) { + size_t pageind = (((uintptr_t)round - + (uintptr_t)chunk) >> pagesize_2pow); + arena_chunk_map_t *mapelm = + &chunk->map[pageind]; + arena_dalloc_small(arena, chunk, round, mapelm); + } else { + /* + * This round was allocated via a different + * arena than the one that is currently locked. + * Stash the round, so that it can be handled + * in a future pass. + */ + mag->rounds[ndeferred] = round; + ndeferred++; + } + } + malloc_spin_unlock(&arena->lock); + } + + mag->nrounds = 0; +} + +static inline void +mag_rack_dalloc(mag_rack_t *rack, void *ptr) +{ + arena_t *arena; + arena_chunk_t *chunk; + arena_run_t *run; + arena_bin_t *bin; + bin_mags_t *bin_mags; + mag_t *mag; + size_t pageind, binind; + arena_chunk_map_t *mapelm; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + arena = chunk->arena; + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + run = (arena_run_t *)(mapelm->bits & ~pagesize_mask); + assert(run->magic == ARENA_RUN_MAGIC); + bin = run->bin; + binind = ((uintptr_t)bin - (uintptr_t)&arena->bins) / + sizeof(arena_bin_t); + assert(binind < nbins); + + if (opt_junk) + memset(ptr, 0x5a, arena->bins[binind].reg_size); + + bin_mags = &rack->bin_mags[binind]; + mag = bin_mags->curmag; + if (mag == NULL) { + /* Create an initial magazine for this size class. */ + assert(bin_mags->sparemag == NULL); + mag = mag_create(choose_arena(), binind); + if (mag == NULL) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, mapelm); + malloc_spin_unlock(&arena->lock); + return; + } + bin_mags->curmag = mag; + } + + if (mag->nrounds == max_rounds) { + if (bin_mags->sparemag != NULL) { + if (bin_mags->sparemag->nrounds < max_rounds) { + /* Swap magazines. */ + bin_mags->curmag = bin_mags->sparemag; + bin_mags->sparemag = mag; + mag = bin_mags->curmag; + } else { + /* Unload the current magazine. */ + mag_unload(mag); + } + } else { + /* Create a second magazine. */ + mag = mag_create(choose_arena(), binind); + if (mag == NULL) { + mag = rack->bin_mags[binind].curmag; + mag_unload(mag); + } else { + bin_mags->sparemag = bin_mags->curmag; + bin_mags->curmag = mag; + } + } + assert(mag->nrounds < max_rounds); + } + mag->rounds[mag->nrounds] = ptr; + mag->nrounds++; +} +#endif + +static void +arena_dalloc_large(arena_t *arena, arena_chunk_t *chunk, void *ptr) +{ + /* Large allocation. */ + malloc_spin_lock(&arena->lock); + +#ifndef MALLOC_STATS + if (opt_junk) +#endif + { + size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> + pagesize_2pow; + size_t size = chunk->map[pageind].bits & ~pagesize_mask; + +#ifdef MALLOC_STATS + if (opt_junk) +#endif + memset(ptr, 0x5a, size); +#ifdef MALLOC_STATS + arena->stats.allocated_large -= size; +#endif + } +#ifdef MALLOC_STATS + arena->stats.ndalloc_large++; +#endif + + arena_run_dalloc(arena, (arena_run_t *)ptr, true); + malloc_spin_unlock(&arena->lock); +} + +static inline void +arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr) +{ + size_t pageind; + arena_chunk_map_t *mapelm; + + assert(arena != NULL); + assert(arena->magic == ARENA_MAGIC); + assert(chunk->arena == arena); + assert(ptr != NULL); + assert(CHUNK_ADDR2BASE(ptr) != ptr); + + pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + assert((mapelm->bits & CHUNK_MAP_ALLOCATED) != 0); + if ((mapelm->bits & CHUNK_MAP_LARGE) == 0) { + /* Small allocation. */ +#ifdef MALLOC_MAG + if (__isthreaded && opt_mag) { + mag_rack_t *rack = mag_rack; + if (rack == NULL) { + rack = mag_rack_create(arena); + if (rack == NULL) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, + mapelm); + malloc_spin_unlock(&arena->lock); + } + mag_rack = rack; + } + mag_rack_dalloc(rack, ptr); + } else { +#endif + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, ptr, mapelm); + malloc_spin_unlock(&arena->lock); +#ifdef MALLOC_MAG + } +#endif + } else + arena_dalloc_large(arena, chunk, ptr); +} + +static inline void +idalloc(void *ptr) +{ + arena_chunk_t *chunk; + + assert(ptr != NULL); + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + if (chunk != ptr) + arena_dalloc(chunk->arena, chunk, ptr); + else + huge_dalloc(ptr); +} + +static void +arena_ralloc_large_shrink(arena_t *arena, arena_chunk_t *chunk, void *ptr, + size_t size, size_t oldsize) +{ + + assert(size < oldsize); + + /* + * Shrink the run, and make trailing pages available for other + * allocations. + */ +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + arena_run_trim_tail(arena, chunk, (arena_run_t *)ptr, oldsize, size, + true); +#ifdef MALLOC_STATS + arena->stats.allocated_large -= oldsize - size; +#endif + malloc_spin_unlock(&arena->lock); +} + +static bool +arena_ralloc_large_grow(arena_t *arena, arena_chunk_t *chunk, void *ptr, + size_t size, size_t oldsize) +{ + size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow; + size_t npages = oldsize >> pagesize_2pow; + + assert(oldsize == (chunk->map[pageind].bits & ~pagesize_mask)); + + /* Try to extend the run. */ + assert(size > oldsize); +#ifdef MALLOC_BALANCE + arena_lock_balance(arena); +#else + malloc_spin_lock(&arena->lock); +#endif + if (pageind + npages < chunk_npages && (chunk->map[pageind+npages].bits + & CHUNK_MAP_ALLOCATED) == 0 && (chunk->map[pageind+npages].bits & + ~pagesize_mask) >= size - oldsize) { + /* + * The next run is available and sufficiently large. Split the + * following run, then merge the first part with the existing + * allocation. + */ + arena_run_split(arena, (arena_run_t *)((uintptr_t)chunk + + ((pageind+npages) << pagesize_2pow)), size - oldsize, true, + false); + + chunk->map[pageind].bits = size | CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + chunk->map[pageind+npages].bits = CHUNK_MAP_LARGE | + CHUNK_MAP_ALLOCATED; + +#ifdef MALLOC_STATS + arena->stats.allocated_large += size - oldsize; +#endif + malloc_spin_unlock(&arena->lock); + return (false); + } + malloc_spin_unlock(&arena->lock); + + return (true); +} + +/* + * Try to resize a large allocation, in order to avoid copying. This will + * always fail if growing an object, and the following run is already in use. + */ +static bool +arena_ralloc_large(void *ptr, size_t size, size_t oldsize) +{ + size_t psize; + + psize = PAGE_CEILING(size); + if (psize == oldsize) { + /* Same size class. */ + if (opt_junk && size < oldsize) { + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize - + size); + } + return (false); + } else { + arena_chunk_t *chunk; + arena_t *arena; + + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + arena = chunk->arena; + assert(arena->magic == ARENA_MAGIC); + + if (psize < oldsize) { + /* Fill before shrinking in order avoid a race. */ + if (opt_junk) { + memset((void *)((uintptr_t)ptr + size), 0x5a, + oldsize - size); + } + arena_ralloc_large_shrink(arena, chunk, ptr, psize, + oldsize); + return (false); + } else { + bool ret = arena_ralloc_large_grow(arena, chunk, ptr, + psize, oldsize); + if (ret == false && opt_zero) { + memset((void *)((uintptr_t)ptr + oldsize), 0, + size - oldsize); + } + return (ret); + } + } +} + +static void * +arena_ralloc(void *ptr, size_t size, size_t oldsize) +{ + void *ret; + size_t copysize; + + /* Try to avoid moving the allocation. */ + if (size <= bin_maxclass) { + if (oldsize <= bin_maxclass && size2bin[size] == + size2bin[oldsize]) + goto IN_PLACE; + } else { + if (oldsize > bin_maxclass && oldsize <= arena_maxclass) { + assert(size > bin_maxclass); + if (arena_ralloc_large(ptr, size, oldsize) == false) + return (ptr); + } + } + + /* + * If we get here, then size and oldsize are different enough that we + * need to move the object. In that case, fall back to allocating new + * space and copying. + */ + ret = arena_malloc(choose_arena(), size, false); + if (ret == NULL) + return (NULL); + + /* Junk/zero-filling were already done by arena_malloc(). */ + copysize = (size < oldsize) ? size : oldsize; + memcpy(ret, ptr, copysize); + idalloc(ptr); + return (ret); +IN_PLACE: + if (opt_junk && size < oldsize) + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize - size); + else if (opt_zero && size > oldsize) + memset((void *)((uintptr_t)ptr + oldsize), 0, size - oldsize); + return (ptr); +} + +static inline void * +iralloc(void *ptr, size_t size) +{ + size_t oldsize; + + assert(ptr != NULL); + assert(size != 0); + + oldsize = isalloc(ptr); + + if (size <= arena_maxclass) + return (arena_ralloc(ptr, size, oldsize)); + else + return (huge_ralloc(ptr, size, oldsize)); +} + +static bool +arena_new(arena_t *arena) +{ + unsigned i; + arena_bin_t *bin; + size_t prev_run_size; + + if (malloc_spin_init(&arena->lock)) + return (true); + +#ifdef MALLOC_STATS + memset(&arena->stats, 0, sizeof(arena_stats_t)); +#endif + + /* Initialize chunks. */ + arena_chunk_tree_dirty_new(&arena->chunks_dirty); + arena->spare = NULL; + + arena->ndirty = 0; + + arena_avail_tree_new(&arena->runs_avail); + +#ifdef MALLOC_BALANCE + arena->contention = 0; +#endif + + /* Initialize bins. */ + prev_run_size = pagesize; + + i = 0; +#ifdef MALLOC_TINY + /* (2^n)-spaced tiny bins. */ + for (; i < ntbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = (1U << (TINY_MIN_2POW + i)); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } +#endif + + /* Quantum-spaced bins. */ + for (; i < ntbins + nqbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = (i - ntbins + 1) << QUANTUM_2POW; + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + + /* Cacheline-spaced bins. */ + for (; i < ntbins + nqbins + ncbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = cspace_min + ((i - (ntbins + nqbins)) << + CACHELINE_2POW); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + + /* Subpage-spaced bins. */ + for (; i < nbins; i++) { + bin = &arena->bins[i]; + bin->runcur = NULL; + arena_run_tree_new(&bin->runs); + + bin->reg_size = sspace_min + ((i - (ntbins + nqbins + ncbins)) + << SUBPAGE_2POW); + + prev_run_size = arena_bin_run_size_calc(bin, prev_run_size); + +#ifdef MALLOC_STATS + memset(&bin->stats, 0, sizeof(malloc_bin_stats_t)); +#endif + } + +#ifdef MALLOC_DEBUG + arena->magic = ARENA_MAGIC; +#endif + + return (false); +} + +/* Create a new arena and insert it into the arenas array at index ind. */ +static arena_t * +arenas_extend(unsigned ind) +{ + arena_t *ret; + + /* Allocate enough space for trailing bins. */ + ret = (arena_t *)base_alloc(sizeof(arena_t) + + (sizeof(arena_bin_t) * (nbins - 1))); + if (ret != NULL && arena_new(ret) == false) { + arenas[ind] = ret; + return (ret); + } + /* Only reached if there is an OOM error. */ + + /* + * OOM here is quite inconvenient to propagate, since dealing with it + * would require a check for failure in the fast path. Instead, punt + * by using arenas[0]. In practice, this is an extremely unlikely + * failure. + */ + _malloc_message(_getprogname(), + ": (malloc) Error initializing arena\n", "", ""); + if (opt_abort) + abort(); + + return (arenas[0]); +} + +#ifdef MALLOC_MAG +static mag_t * +mag_create(arena_t *arena, size_t binind) +{ + mag_t *ret; + + if (sizeof(mag_t) + (sizeof(void *) * (max_rounds - 1)) <= + bin_maxclass) { + ret = arena_malloc_small(arena, sizeof(mag_t) + (sizeof(void *) + * (max_rounds - 1)), false); + } else { + ret = imalloc(sizeof(mag_t) + (sizeof(void *) * (max_rounds - + 1))); + } + if (ret == NULL) + return (NULL); + ret->binind = binind; + ret->nrounds = 0; + + return (ret); +} + +static void +mag_destroy(mag_t *mag) +{ + arena_t *arena; + arena_chunk_t *chunk; + size_t pageind; + arena_chunk_map_t *mapelm; + + chunk = CHUNK_ADDR2BASE(mag); + arena = chunk->arena; + pageind = (((uintptr_t)mag - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + + assert(mag->nrounds == 0); + if (sizeof(mag_t) + (sizeof(void *) * (max_rounds - 1)) <= + bin_maxclass) { + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, mag, mapelm); + malloc_spin_unlock(&arena->lock); + } else + idalloc(mag); +} + +static mag_rack_t * +mag_rack_create(arena_t *arena) +{ + + assert(sizeof(mag_rack_t) + (sizeof(bin_mags_t *) * (nbins - 1)) <= + bin_maxclass); + return (arena_malloc_small(arena, sizeof(mag_rack_t) + + (sizeof(bin_mags_t) * (nbins - 1)), true)); +} + +static void +mag_rack_destroy(mag_rack_t *rack) +{ + arena_t *arena; + arena_chunk_t *chunk; + bin_mags_t *bin_mags; + size_t i, pageind; + arena_chunk_map_t *mapelm; + + for (i = 0; i < nbins; i++) { + bin_mags = &rack->bin_mags[i]; + if (bin_mags->curmag != NULL) { + assert(bin_mags->curmag->binind == i); + mag_unload(bin_mags->curmag); + mag_destroy(bin_mags->curmag); + } + if (bin_mags->sparemag != NULL) { + assert(bin_mags->sparemag->binind == i); + mag_unload(bin_mags->sparemag); + mag_destroy(bin_mags->sparemag); + } + } + + chunk = CHUNK_ADDR2BASE(rack); + arena = chunk->arena; + pageind = (((uintptr_t)rack - (uintptr_t)chunk) >> pagesize_2pow); + mapelm = &chunk->map[pageind]; + + malloc_spin_lock(&arena->lock); + arena_dalloc_small(arena, chunk, rack, mapelm); + malloc_spin_unlock(&arena->lock); +} +#endif + +/* + * End arena. + */ +/******************************************************************************/ +/* + * Begin general internal functions. + */ + +static void * +huge_malloc(size_t size, bool zero) +{ + void *ret; + size_t csize; + extent_node_t *node; + + /* Allocate one or more contiguous chunks for this request. */ + + csize = CHUNK_CEILING(size); + if (csize == 0) { + /* size is large enough to cause size_t wrap-around. */ + return (NULL); + } + + /* Allocate an extent node with which to track the chunk. */ + node = base_node_alloc(); + if (node == NULL) + return (NULL); + + ret = chunk_alloc(csize, zero); + if (ret == NULL) { + base_node_dealloc(node); + return (NULL); + } + + /* Insert node into huge. */ + node->addr = ret; + node->size = csize; + + malloc_mutex_lock(&huge_mtx); + extent_tree_ad_insert(&huge, node); +#ifdef MALLOC_STATS + huge_nmalloc++; + huge_allocated += csize; +#endif + malloc_mutex_unlock(&huge_mtx); + + if (zero == false) { + if (opt_junk) + memset(ret, 0xa5, csize); + else if (opt_zero) + memset(ret, 0, csize); + } + + return (ret); +} + +/* Only handles large allocations that require more than chunk alignment. */ +static void * +huge_palloc(size_t alignment, size_t size) +{ + void *ret; + size_t alloc_size, chunk_size, offset; + extent_node_t *node; + + /* + * This allocation requires alignment that is even larger than chunk + * alignment. This means that huge_malloc() isn't good enough. + * + * Allocate almost twice as many chunks as are demanded by the size or + * alignment, in order to assure the alignment can be achieved, then + * unmap leading and trailing chunks. + */ + assert(alignment >= chunksize); + + chunk_size = CHUNK_CEILING(size); + + if (size >= alignment) + alloc_size = chunk_size + alignment - chunksize; + else + alloc_size = (alignment << 1) - chunksize; + + /* Allocate an extent node with which to track the chunk. */ + node = base_node_alloc(); + if (node == NULL) + return (NULL); + + ret = chunk_alloc(alloc_size, false); + if (ret == NULL) { + base_node_dealloc(node); + return (NULL); + } + + offset = (uintptr_t)ret & (alignment - 1); + assert((offset & chunksize_mask) == 0); + assert(offset < alloc_size); + if (offset == 0) { + /* Trim trailing space. */ + chunk_dealloc((void *)((uintptr_t)ret + chunk_size), alloc_size + - chunk_size); + } else { + size_t trailsize; + + /* Trim leading space. */ + chunk_dealloc(ret, alignment - offset); + + ret = (void *)((uintptr_t)ret + (alignment - offset)); + + trailsize = alloc_size - (alignment - offset) - chunk_size; + if (trailsize != 0) { + /* Trim trailing space. */ + assert(trailsize < alloc_size); + chunk_dealloc((void *)((uintptr_t)ret + chunk_size), + trailsize); + } + } + + /* Insert node into huge. */ + node->addr = ret; + node->size = chunk_size; + + malloc_mutex_lock(&huge_mtx); + extent_tree_ad_insert(&huge, node); +#ifdef MALLOC_STATS + huge_nmalloc++; + huge_allocated += chunk_size; +#endif + malloc_mutex_unlock(&huge_mtx); + + if (opt_junk) + memset(ret, 0xa5, chunk_size); + else if (opt_zero) + memset(ret, 0, chunk_size); + + return (ret); +} + +static void * +huge_ralloc(void *ptr, size_t size, size_t oldsize) +{ + void *ret; + size_t copysize; + + /* Avoid moving the allocation if the size class would not change. */ + if (oldsize > arena_maxclass && + CHUNK_CEILING(size) == CHUNK_CEILING(oldsize)) { + if (opt_junk && size < oldsize) { + memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize + - size); + } else if (opt_zero && size > oldsize) { + memset((void *)((uintptr_t)ptr + oldsize), 0, size + - oldsize); + } + return (ptr); + } + + /* + * If we get here, then size and oldsize are different enough that we + * need to use a different size class. In that case, fall back to + * allocating new space and copying. + */ + ret = huge_malloc(size, false); + if (ret == NULL) + return (NULL); + + copysize = (size < oldsize) ? size : oldsize; + memcpy(ret, ptr, copysize); + idalloc(ptr); + return (ret); +} + +static void +huge_dalloc(void *ptr) +{ + extent_node_t *node, key; + + malloc_mutex_lock(&huge_mtx); + + /* Extract from tree of huge allocations. */ + key.addr = ptr; + node = extent_tree_ad_search(&huge, &key); + assert(node != NULL); + assert(node->addr == ptr); + extent_tree_ad_remove(&huge, node); + +#ifdef MALLOC_STATS + huge_ndalloc++; + huge_allocated -= node->size; +#endif + + malloc_mutex_unlock(&huge_mtx); + + /* Unmap chunk. */ +#ifdef MALLOC_DSS + if (opt_dss && opt_junk) + memset(node->addr, 0x5a, node->size); +#endif + chunk_dealloc(node->addr, node->size); + + base_node_dealloc(node); +} + +static void +malloc_print_stats(void) +{ + + if (opt_print_stats) { + char s[UMAX2S_BUFSIZE]; + _malloc_message("___ Begin malloc statistics ___\n", "", "", + ""); + _malloc_message("Assertions ", +#ifdef NDEBUG + "disabled", +#else + "enabled", +#endif + "\n", ""); + _malloc_message("Boolean MALLOC_OPTIONS: ", + opt_abort ? "A" : "a", "", ""); +#ifdef MALLOC_DSS + _malloc_message(opt_dss ? "D" : "d", "", "", ""); +#endif +#ifdef MALLOC_MAG + _malloc_message(opt_mag ? "G" : "g", "", "", ""); +#endif + _malloc_message(opt_junk ? "J" : "j", "", "", ""); +#ifdef MALLOC_DSS + _malloc_message(opt_mmap ? "M" : "m", "", "", ""); +#endif + _malloc_message(opt_utrace ? "PU" : "Pu", + opt_sysv ? "V" : "v", + opt_xmalloc ? "X" : "x", + opt_zero ? "Z\n" : "z\n"); + + _malloc_message("CPUs: ", umax2s(ncpus, s), "\n", ""); + _malloc_message("Max arenas: ", umax2s(narenas, s), "\n", ""); +#ifdef MALLOC_BALANCE + _malloc_message("Arena balance threshold: ", + umax2s(opt_balance_threshold, s), "\n", ""); +#endif + _malloc_message("Pointer size: ", umax2s(sizeof(void *), s), + "\n", ""); + _malloc_message("Quantum size: ", umax2s(QUANTUM, s), "\n", ""); + _malloc_message("Cacheline size (assumed): ", umax2s(CACHELINE, + s), "\n", ""); +#ifdef MALLOC_TINY + _malloc_message("Tiny 2^n-spaced sizes: [", umax2s((1U << + TINY_MIN_2POW), s), "..", ""); + _malloc_message(umax2s((qspace_min >> 1), s), "]\n", "", ""); +#endif + _malloc_message("Quantum-spaced sizes: [", umax2s(qspace_min, + s), "..", ""); + _malloc_message(umax2s(qspace_max, s), "]\n", "", ""); + _malloc_message("Cacheline-spaced sizes: [", umax2s(cspace_min, + s), "..", ""); + _malloc_message(umax2s(cspace_max, s), "]\n", "", ""); + _malloc_message("Subpage-spaced sizes: [", umax2s(sspace_min, + s), "..", ""); + _malloc_message(umax2s(sspace_max, s), "]\n", "", ""); +#ifdef MALLOC_MAG + _malloc_message("Rounds per magazine: ", umax2s(max_rounds, s), + "\n", ""); +#endif + _malloc_message("Max dirty pages per arena: ", + umax2s(opt_dirty_max, s), "\n", ""); + + _malloc_message("Chunk size: ", umax2s(chunksize, s), "", ""); + _malloc_message(" (2^", umax2s(opt_chunk_2pow, s), ")\n", ""); + +#ifdef MALLOC_STATS + { + size_t allocated, mapped; +#ifdef MALLOC_BALANCE + uint64_t nbalance = 0; +#endif + unsigned i; + arena_t *arena; + + /* Calculate and print allocated/mapped stats. */ + + /* arenas. */ + for (i = 0, allocated = 0; i < narenas; i++) { + if (arenas[i] != NULL) { + malloc_spin_lock(&arenas[i]->lock); + allocated += + arenas[i]->stats.allocated_small; + allocated += + arenas[i]->stats.allocated_large; +#ifdef MALLOC_BALANCE + nbalance += arenas[i]->stats.nbalance; +#endif + malloc_spin_unlock(&arenas[i]->lock); + } + } + + /* huge/base. */ + malloc_mutex_lock(&huge_mtx); + allocated += huge_allocated; + mapped = stats_chunks.curchunks * chunksize; + malloc_mutex_unlock(&huge_mtx); + + malloc_mutex_lock(&base_mtx); + mapped += base_mapped; + malloc_mutex_unlock(&base_mtx); + + malloc_printf("Allocated: %zu, mapped: %zu\n", + allocated, mapped); + +#ifdef MALLOC_BALANCE + malloc_printf("Arena balance reassignments: %llu\n", + nbalance); +#endif + + /* Print chunk stats. */ + { + chunk_stats_t chunks_stats; + + malloc_mutex_lock(&huge_mtx); + chunks_stats = stats_chunks; + malloc_mutex_unlock(&huge_mtx); + + malloc_printf("chunks: nchunks " + "highchunks curchunks\n"); + malloc_printf(" %13llu%13lu%13lu\n", + chunks_stats.nchunks, + chunks_stats.highchunks, + chunks_stats.curchunks); + } + + /* Print chunk stats. */ + malloc_printf( + "huge: nmalloc ndalloc allocated\n"); + malloc_printf(" %12llu %12llu %12zu\n", + huge_nmalloc, huge_ndalloc, huge_allocated); + + /* Print stats for each arena. */ + for (i = 0; i < narenas; i++) { + arena = arenas[i]; + if (arena != NULL) { + malloc_printf( + "\narenas[%u]:\n", i); + malloc_spin_lock(&arena->lock); + stats_print(arena); + malloc_spin_unlock(&arena->lock); + } + } + } +#endif /* #ifdef MALLOC_STATS */ + _malloc_message("--- End malloc statistics ---\n", "", "", ""); + } +} + +#ifdef MALLOC_DEBUG +static void +size2bin_validate(void) +{ + size_t i, size, binind; + + assert(size2bin[0] == 0xffU); + i = 1; +# ifdef MALLOC_TINY + /* Tiny. */ + for (; i < (1U << TINY_MIN_2POW); i++) { + size = pow2_ceil(1U << TINY_MIN_2POW); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + assert(size2bin[i] == binind); + } + for (; i < qspace_min; i++) { + size = pow2_ceil(i); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + assert(size2bin[i] == binind); + } +# endif + /* Quantum-spaced. */ + for (; i <= qspace_max; i++) { + size = QUANTUM_CEILING(i); + binind = ntbins + (size >> QUANTUM_2POW) - 1; + assert(size2bin[i] == binind); + } + /* Cacheline-spaced. */ + for (; i <= cspace_max; i++) { + size = CACHELINE_CEILING(i); + binind = ntbins + nqbins + ((size - cspace_min) >> + CACHELINE_2POW); + assert(size2bin[i] == binind); + } + /* Sub-page. */ + for (; i <= sspace_max; i++) { + size = SUBPAGE_CEILING(i); + binind = ntbins + nqbins + ncbins + ((size - sspace_min) + >> SUBPAGE_2POW); + assert(size2bin[i] == binind); + } +} +#endif + +static bool +size2bin_init(void) +{ + + if (opt_qspace_max_2pow != QSPACE_MAX_2POW_DEFAULT + || opt_cspace_max_2pow != CSPACE_MAX_2POW_DEFAULT) + return (size2bin_init_hard()); + + size2bin = const_size2bin; +#ifdef MALLOC_DEBUG + assert(sizeof(const_size2bin) == bin_maxclass + 1); + size2bin_validate(); +#endif + return (false); +} + +static bool +size2bin_init_hard(void) +{ + size_t i, size, binind; + uint8_t *custom_size2bin; + + assert(opt_qspace_max_2pow != QSPACE_MAX_2POW_DEFAULT + || opt_cspace_max_2pow != CSPACE_MAX_2POW_DEFAULT); + + custom_size2bin = (uint8_t *)base_alloc(bin_maxclass + 1); + if (custom_size2bin == NULL) + return (true); + + custom_size2bin[0] = 0xffU; + i = 1; +#ifdef MALLOC_TINY + /* Tiny. */ + for (; i < (1U << TINY_MIN_2POW); i++) { + size = pow2_ceil(1U << TINY_MIN_2POW); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + custom_size2bin[i] = binind; + } + for (; i < qspace_min; i++) { + size = pow2_ceil(i); + binind = ffs((int)(size >> (TINY_MIN_2POW + 1))); + custom_size2bin[i] = binind; + } +#endif + /* Quantum-spaced. */ + for (; i <= qspace_max; i++) { + size = QUANTUM_CEILING(i); + binind = ntbins + (size >> QUANTUM_2POW) - 1; + custom_size2bin[i] = binind; + } + /* Cacheline-spaced. */ + for (; i <= cspace_max; i++) { + size = CACHELINE_CEILING(i); + binind = ntbins + nqbins + ((size - cspace_min) >> + CACHELINE_2POW); + custom_size2bin[i] = binind; + } + /* Sub-page. */ + for (; i <= sspace_max; i++) { + size = SUBPAGE_CEILING(i); + binind = ntbins + nqbins + ncbins + ((size - sspace_min) >> + SUBPAGE_2POW); + custom_size2bin[i] = binind; + } + + size2bin = custom_size2bin; +#ifdef MALLOC_DEBUG + size2bin_validate(); +#endif + return (false); +} + +/* + * FreeBSD's pthreads implementation calls malloc(3), so the malloc + * implementation has to take pains to avoid infinite recursion during + * initialization. + */ +static inline bool +malloc_init(void) +{ + + if (malloc_initialized == false) + return (malloc_init_hard()); + + return (false); +} + +static bool +malloc_init_hard(void) +{ + unsigned i; + int linklen; + char buf[PATH_MAX + 1]; + const char *opts; + + malloc_mutex_lock(&init_lock); + if (malloc_initialized) { + /* + * Another thread initialized the allocator before this one + * acquired init_lock. + */ + malloc_mutex_unlock(&init_lock); + return (false); + } + + /* Get number of CPUs. */ + { + int mib[2]; + size_t len; + + mib[0] = CTL_HW; + mib[1] = HW_NCPU; + len = sizeof(ncpus); + if (sysctl(mib, 2, &ncpus, &len, (void *) 0, 0) == -1) { + /* Error. */ + ncpus = 1; + } + } + + /* Get page size. */ + { + long result; + + result = sysconf(_SC_PAGESIZE); + assert(result != -1); + pagesize = (unsigned)result; + + /* + * We assume that pagesize is a power of 2 when calculating + * pagesize_mask and pagesize_2pow. + */ + assert(((result - 1) & result) == 0); + pagesize_mask = result - 1; + pagesize_2pow = ffs((int)result) - 1; + } + + for (i = 0; i < 3; i++) { + unsigned j; + + /* Get runtime configuration. */ + switch (i) { + case 0: + if ((linklen = readlink("/etc/malloc.conf", buf, + sizeof(buf) - 1)) != -1) { + /* + * Use the contents of the "/etc/malloc.conf" + * symbolic link's name. + */ + buf[linklen] = '\0'; + opts = buf; + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + case 1: + if (issetugid() == 0 && (opts = + getenv("MALLOC_OPTIONS")) != NULL) { + /* + * Do nothing; opts is already initialized to + * the value of the MALLOC_OPTIONS environment + * variable. + */ + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + case 2: + if (_malloc_options != NULL) { + /* + * Use options that were compiled into the + * program. + */ + opts = _malloc_options; + } else { + /* No configuration specified. */ + buf[0] = '\0'; + opts = buf; + } + break; + default: + /* NOTREACHED */ + assert(false); + } + + for (j = 0; opts[j] != '\0'; j++) { + unsigned k, nreps; + bool nseen; + + /* Parse repetition count, if any. */ + for (nreps = 0, nseen = false;; j++, nseen = true) { + switch (opts[j]) { + case '0': case '1': case '2': case '3': + case '4': case '5': case '6': case '7': + case '8': case '9': + nreps *= 10; + nreps += opts[j] - '0'; + break; + default: + goto MALLOC_OUT; + } + } +MALLOC_OUT: + if (nseen == false) + nreps = 1; + + for (k = 0; k < nreps; k++) { + switch (opts[j]) { + case 'a': + opt_abort = false; + break; + case 'A': + opt_abort = true; + break; + case 'b': +#ifdef MALLOC_BALANCE + opt_balance_threshold >>= 1; +#endif + break; + case 'B': +#ifdef MALLOC_BALANCE + if (opt_balance_threshold == 0) + opt_balance_threshold = 1; + else if ((opt_balance_threshold << 1) + > opt_balance_threshold) + opt_balance_threshold <<= 1; +#endif + break; + case 'c': + if (opt_cspace_max_2pow - 1 > + opt_qspace_max_2pow && + opt_cspace_max_2pow > + CACHELINE_2POW) + opt_cspace_max_2pow--; + break; + case 'C': + if (opt_cspace_max_2pow < pagesize_2pow + - 1) + opt_cspace_max_2pow++; + break; + case 'd': +#ifdef MALLOC_DSS + opt_dss = false; +#endif + break; + case 'D': +#ifdef MALLOC_DSS + opt_dss = true; +#endif + break; + case 'f': + opt_dirty_max >>= 1; + break; + case 'F': + if (opt_dirty_max == 0) + opt_dirty_max = 1; + else if ((opt_dirty_max << 1) != 0) + opt_dirty_max <<= 1; + break; +#ifdef MALLOC_MAG + case 'g': + opt_mag = false; + break; + case 'G': + opt_mag = true; + break; +#endif + case 'j': + opt_junk = false; + break; + case 'J': + opt_junk = true; + break; + case 'k': + /* + * Chunks always require at least one + * header page, so chunks can never be + * smaller than two pages. + */ + if (opt_chunk_2pow > pagesize_2pow + 1) + opt_chunk_2pow--; + break; + case 'K': + if (opt_chunk_2pow + 1 < + (sizeof(size_t) << 3)) + opt_chunk_2pow++; + break; + case 'm': +#ifdef MALLOC_DSS + opt_mmap = false; +#endif + break; + case 'M': +#ifdef MALLOC_DSS + opt_mmap = true; +#endif + break; + case 'n': + opt_narenas_lshift--; + break; + case 'N': + opt_narenas_lshift++; + break; + case 'p': + opt_print_stats = false; + break; + case 'P': + opt_print_stats = true; + break; + case 'q': + if (opt_qspace_max_2pow > QUANTUM_2POW) + opt_qspace_max_2pow--; + break; + case 'Q': + if (opt_qspace_max_2pow + 1 < + opt_cspace_max_2pow) + opt_qspace_max_2pow++; + break; +#ifdef MALLOC_MAG + case 'R': + if (opt_mag_size_2pow + 1 < (8U << + SIZEOF_PTR_2POW)) + opt_mag_size_2pow++; + break; + case 'r': + /* + * Make sure there's always at least + * one round per magazine. + */ + if ((1U << (opt_mag_size_2pow-1)) >= + sizeof(mag_t)) + opt_mag_size_2pow--; + break; +#endif + case 'u': + opt_utrace = false; + break; + case 'U': + opt_utrace = true; + break; + case 'v': + opt_sysv = false; + break; + case 'V': + opt_sysv = true; + break; + case 'x': + opt_xmalloc = false; + break; + case 'X': + opt_xmalloc = true; + break; + case 'z': + opt_zero = false; + break; + case 'Z': + opt_zero = true; + break; + default: { + char cbuf[2]; + + cbuf[0] = opts[j]; + cbuf[1] = '\0'; + _malloc_message(_getprogname(), + ": (malloc) Unsupported character " + "in malloc options: '", cbuf, + "'\n"); + } + } + } + } + } + +#ifdef MALLOC_DSS + /* Make sure that there is some method for acquiring memory. */ + if (opt_dss == false && opt_mmap == false) + opt_mmap = true; +#endif + + /* Take care to call atexit() only once. */ + if (opt_print_stats) { + /* Print statistics at exit. */ + atexit(malloc_print_stats); + } + +#ifdef MALLOC_MAG + /* + * Calculate the actual number of rounds per magazine, taking into + * account header overhead. + */ + max_rounds = (1LLU << (opt_mag_size_2pow - SIZEOF_PTR_2POW)) - + (sizeof(mag_t) >> SIZEOF_PTR_2POW) + 1; +#endif + + /* Set variables according to the value of opt_[qc]space_max_2pow. */ + qspace_max = (1U << opt_qspace_max_2pow); + cspace_min = CACHELINE_CEILING(qspace_max); + if (cspace_min == qspace_max) + cspace_min += CACHELINE; + cspace_max = (1U << opt_cspace_max_2pow); + sspace_min = SUBPAGE_CEILING(cspace_max); + if (sspace_min == cspace_max) + sspace_min += SUBPAGE; + assert(sspace_min < pagesize); + sspace_max = pagesize - SUBPAGE; + +#ifdef MALLOC_TINY + assert(QUANTUM_2POW >= TINY_MIN_2POW); +#endif + assert(ntbins <= QUANTUM_2POW); + nqbins = qspace_max >> QUANTUM_2POW; + ncbins = ((cspace_max - cspace_min) >> CACHELINE_2POW) + 1; + nsbins = ((sspace_max - sspace_min) >> SUBPAGE_2POW) + 1; + nbins = ntbins + nqbins + ncbins + nsbins; + + if (size2bin_init()) { + malloc_mutex_unlock(&init_lock); + return (true); + } + + /* Set variables according to the value of opt_chunk_2pow. */ + chunksize = (1LU << opt_chunk_2pow); + chunksize_mask = chunksize - 1; + chunk_npages = (chunksize >> pagesize_2pow); + { + size_t header_size; + + /* + * Compute the header size such that it is large enough to + * contain the page map. + */ + header_size = sizeof(arena_chunk_t) + + (sizeof(arena_chunk_map_t) * (chunk_npages - 1)); + arena_chunk_header_npages = (header_size >> pagesize_2pow) + + ((header_size & pagesize_mask) != 0); + } + arena_maxclass = chunksize - (arena_chunk_header_npages << + pagesize_2pow); + + UTRACE(0, 0, 0); + +#ifdef MALLOC_STATS + memset(&stats_chunks, 0, sizeof(chunk_stats_t)); +#endif + + /* Various sanity checks that regard configuration. */ + assert(chunksize >= pagesize); + + /* Initialize chunks data. */ + malloc_mutex_init(&huge_mtx); + extent_tree_ad_new(&huge); +#ifdef MALLOC_DSS + malloc_mutex_init(&dss_mtx); + dss_base = sbrk(0); + dss_prev = dss_base; + dss_max = dss_base; + extent_tree_szad_new(&dss_chunks_szad); + extent_tree_ad_new(&dss_chunks_ad); +#endif +#ifdef MALLOC_STATS + huge_nmalloc = 0; + huge_ndalloc = 0; + huge_allocated = 0; +#endif + + /* Initialize base allocation data structures. */ +#ifdef MALLOC_STATS + base_mapped = 0; +#endif +#ifdef MALLOC_DSS + /* + * Allocate a base chunk here, since it doesn't actually have to be + * chunk-aligned. Doing this before allocating any other chunks allows + * the use of space that would otherwise be wasted. + */ + if (opt_dss) + base_pages_alloc(0); +#endif + base_nodes = NULL; + malloc_mutex_init(&base_mtx); + + if (ncpus > 1) { + /* + * For SMP systems, create twice as many arenas as there are + * CPUs by default. + */ + opt_narenas_lshift++; + } + + /* Determine how many arenas to use. */ + narenas = ncpus; + if (opt_narenas_lshift > 0) { + if ((narenas << opt_narenas_lshift) > narenas) + narenas <<= opt_narenas_lshift; + /* + * Make sure not to exceed the limits of what base_alloc() can + * handle. + */ + if (narenas * sizeof(arena_t *) > chunksize) + narenas = chunksize / sizeof(arena_t *); + } else if (opt_narenas_lshift < 0) { + if ((narenas >> -opt_narenas_lshift) < narenas) + narenas >>= -opt_narenas_lshift; + /* Make sure there is at least one arena. */ + if (narenas == 0) + narenas = 1; + } +#ifdef MALLOC_BALANCE + assert(narenas != 0); + for (narenas_2pow = 0; + (narenas >> (narenas_2pow + 1)) != 0; + narenas_2pow++); +#endif + +#ifdef NO_TLS + if (narenas > 1) { + static const unsigned primes[] = {1, 3, 5, 7, 11, 13, 17, 19, + 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, + 89, 97, 101, 103, 107, 109, 113, 127, 131, 137, 139, 149, + 151, 157, 163, 167, 173, 179, 181, 191, 193, 197, 199, 211, + 223, 227, 229, 233, 239, 241, 251, 257, 263}; + unsigned nprimes, parenas; + + /* + * Pick a prime number of hash arenas that is more than narenas + * so that direct hashing of pthread_self() pointers tends to + * spread allocations evenly among the arenas. + */ + assert((narenas & 1) == 0); /* narenas must be even. */ + nprimes = (sizeof(primes) >> SIZEOF_INT_2POW); + parenas = primes[nprimes - 1]; /* In case not enough primes. */ + for (i = 1; i < nprimes; i++) { + if (primes[i] > narenas) { + parenas = primes[i]; + break; + } + } + narenas = parenas; + } +#endif + +#ifndef NO_TLS +# ifndef MALLOC_BALANCE + next_arena = 0; +# endif +#endif + + /* Allocate and initialize arenas. */ + arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas); + if (arenas == NULL) { + malloc_mutex_unlock(&init_lock); + return (true); + } + /* + * Zero the array. In practice, this should always be pre-zeroed, + * since it was just mmap()ed, but let's be sure. + */ + memset(arenas, 0, sizeof(arena_t *) * narenas); + + /* + * Initialize one arena here. The rest are lazily created in + * choose_arena_hard(). + */ + arenas_extend(0); + if (arenas[0] == NULL) { + malloc_mutex_unlock(&init_lock); + return (true); + } +#ifndef NO_TLS + /* + * Assign the initial arena to the initial thread, in order to avoid + * spurious creation of an extra arena if the application switches to + * threaded mode. + */ + arenas_map = arenas[0]; +#endif + /* + * Seed here for the initial thread, since choose_arena_hard() is only + * called for other threads. The seed value doesn't really matter. + */ +#ifdef MALLOC_BALANCE + SPRN(balance, 42); +#endif + + malloc_spin_init(&arenas_lock); + + malloc_initialized = true; + malloc_mutex_unlock(&init_lock); + return (false); +} + +/* + * End general internal functions. + */ +/******************************************************************************/ +/* + * Begin malloc(3)-compatible functions. + */ + +void * +malloc(size_t size) +{ + void *ret; + + if (malloc_init()) { + ret = NULL; + goto RETURN; + } + + if (size == 0) { + if (opt_sysv == false) + size = 1; + else { + ret = NULL; + goto RETURN; + } + } + + ret = imalloc(size); + +RETURN: + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in malloc(): out of memory\n", "", + ""); + abort(); + } + errno = ENOMEM; + } + + UTRACE(0, size, ret); + return (ret); +} + +int +posix_memalign(void **memptr, size_t alignment, size_t size) +{ + int ret; + void *result; + + if (malloc_init()) + result = NULL; + else { + /* Make sure that alignment is a large enough power of 2. */ + if (((alignment - 1) & alignment) != 0 + || alignment < sizeof(void *)) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in posix_memalign(): " + "invalid alignment\n", "", ""); + abort(); + } + result = NULL; + ret = EINVAL; + goto RETURN; + } + + result = ipalloc(alignment, size); + } + + if (result == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in posix_memalign(): out of memory\n", + "", ""); + abort(); + } + ret = ENOMEM; + goto RETURN; + } + + *memptr = result; + ret = 0; + +RETURN: + UTRACE(0, size, result); + return (ret); +} + +void * +calloc(size_t num, size_t size) +{ + void *ret; + size_t num_size; + + if (malloc_init()) { + num_size = 0; + ret = NULL; + goto RETURN; + } + + num_size = num * size; + if (num_size == 0) { + if ((opt_sysv == false) && ((num == 0) || (size == 0))) + num_size = 1; + else { + ret = NULL; + goto RETURN; + } + /* + * Try to avoid division here. We know that it isn't possible to + * overflow during multiplication if neither operand uses any of the + * most significant half of the bits in a size_t. + */ + } else if (((num | size) & (SIZE_T_MAX << (sizeof(size_t) << 2))) + && (num_size / size != num)) { + /* size_t overflow. */ + ret = NULL; + goto RETURN; + } + + ret = icalloc(num_size); + +RETURN: + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in calloc(): out of memory\n", "", + ""); + abort(); + } + errno = ENOMEM; + } + + UTRACE(0, num_size, ret); + return (ret); +} + +void * +realloc(void *ptr, size_t size) +{ + void *ret; + + if (size == 0) { + if (opt_sysv == false) + size = 1; + else { + if (ptr != NULL) + idalloc(ptr); + ret = NULL; + goto RETURN; + } + } + + if (ptr != NULL) { + assert(malloc_initialized); + + ret = iralloc(ptr, size); + + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in realloc(): out of " + "memory\n", "", ""); + abort(); + } + errno = ENOMEM; + } + } else { + if (malloc_init()) + ret = NULL; + else + ret = imalloc(size); + + if (ret == NULL) { + if (opt_xmalloc) { + _malloc_message(_getprogname(), + ": (malloc) Error in realloc(): out of " + "memory\n", "", ""); + abort(); + } + errno = ENOMEM; + } + } + +RETURN: + UTRACE(ptr, size, ret); + return (ret); +} + +void +free(void *ptr) +{ + + UTRACE(ptr, 0, 0); + if (ptr != NULL) { + assert(malloc_initialized); + + idalloc(ptr); + } +} + +/* + * End malloc(3)-compatible functions. + */ +/******************************************************************************/ +/* + * Begin non-standard functions. + */ + +size_t +malloc_usable_size(const void *ptr) +{ + + assert(ptr != NULL); + + return (isalloc(ptr)); +} + +/* + * End non-standard functions. + */ +/******************************************************************************/ +/* + * Begin library-private functions. + */ + +/******************************************************************************/ +/* + * Begin thread cache. + */ + +/* + * We provide an unpublished interface in order to receive notifications from + * the pthreads library whenever a thread exits. This allows us to clean up + * thread caches. + */ +void +_malloc_thread_cleanup(void) +{ + +#ifdef MALLOC_MAG + if (mag_rack != NULL) { + assert(mag_rack != (void *)-1); + mag_rack_destroy(mag_rack); +#ifdef MALLOC_DEBUG + mag_rack = (void *)-1; +#endif + } +#endif +} + +/* + * The following functions are used by threading libraries for protection of + * malloc during fork(). These functions are only called if the program is + * running in threaded mode, so there is no need to check whether the program + * is threaded here. + */ + +void +_malloc_prefork(void) +{ + unsigned i; + + /* Acquire all mutexes in a safe order. */ + + malloc_spin_lock(&arenas_lock); + for (i = 0; i < narenas; i++) { + if (arenas[i] != NULL) + malloc_spin_lock(&arenas[i]->lock); + } + malloc_spin_unlock(&arenas_lock); + + malloc_mutex_lock(&base_mtx); + + malloc_mutex_lock(&huge_mtx); + +#ifdef MALLOC_DSS + malloc_mutex_lock(&dss_mtx); +#endif +} + +void +_malloc_postfork(void) +{ + unsigned i; + + /* Release all mutexes, now that fork() has completed. */ + +#ifdef MALLOC_DSS + malloc_mutex_unlock(&dss_mtx); +#endif + + malloc_mutex_unlock(&huge_mtx); + + malloc_mutex_unlock(&base_mtx); + + malloc_spin_lock(&arenas_lock); + for (i = 0; i < narenas; i++) { + if (arenas[i] != NULL) + malloc_spin_unlock(&arenas[i]->lock); + } + malloc_spin_unlock(&arenas_lock); +} + +/* + * End library-private functions. + */ +/******************************************************************************/ diff --git a/varnish-cache/lib/libjemalloc/rb.h b/varnish-cache/lib/libjemalloc/rb.h new file mode 100644 index 00000000..acfe203d --- /dev/null +++ b/varnish-cache/lib/libjemalloc/rb.h @@ -0,0 +1,946 @@ +/****************************************************************************** + * + * Copyright (C) 2008 Jason Evans . + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice(s), this list of conditions and the following disclaimer + * unmodified other than the allowable addition of one or more + * copyright notices. + * 2. Redistributions in binary form must reproduce the above copyright + * notice(s), this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE + * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR + * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, + * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE + * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, + * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + ****************************************************************************** + * + * cpp macro implementation of left-leaning red-black trees. + * + * Usage: + * + * (Optional, see assert(3).) + * #define NDEBUG + * + * (Required.) + * #include + * #include + * ... + * + * All operations are done non-recursively. Parent pointers are not used, and + * color bits are stored in the least significant bit of right-child pointers, + * thus making node linkage as compact as is possible for red-black trees. + * + * Some macros use a comparison function pointer, which is expected to have the + * following prototype: + * + * int (a_cmp *)(a_type *a_node, a_type *a_other); + * ^^^^^^ + * or a_key + * + * Interpretation of comparision function return values: + * + * -1 : a_node < a_other + * 0 : a_node == a_other + * 1 : a_node > a_other + * + * In all cases, the a_node or a_key macro argument is the first argument to the + * comparison function, which makes it possible to write comparison functions + * that treat the first argument specially. + * + ******************************************************************************/ + +#ifndef RB_H_ +#define RB_H_ + +//__FBSDID("$FreeBSD: head/lib/libc/stdlib/rb.h 178995 2008-05-14 18:33:13Z jasone $"); + +/* Node structure. */ +#define rb_node(a_type) \ +struct { \ + a_type *rbn_left; \ + a_type *rbn_right_red; \ +} + +/* Root structure. */ +#define rb_tree(a_type) \ +struct { \ + a_type *rbt_root; \ + a_type rbt_nil; \ +} + +/* Left accessors. */ +#define rbp_left_get(a_type, a_field, a_node) \ + ((a_node)->a_field.rbn_left) +#define rbp_left_set(a_type, a_field, a_node, a_left) do { \ + (a_node)->a_field.rbn_left = a_left; \ +} while (0) + +/* Right accessors. */ +#define rbp_right_get(a_type, a_field, a_node) \ + ((a_type *) (((intptr_t) (a_node)->a_field.rbn_right_red) \ + & ((ssize_t)-2))) +#define rbp_right_set(a_type, a_field, a_node, a_right) do { \ + (a_node)->a_field.rbn_right_red = (a_type *) (((uintptr_t) a_right) \ + | (((uintptr_t) (a_node)->a_field.rbn_right_red) & ((size_t)1))); \ +} while (0) + +/* Color accessors. */ +#define rbp_red_get(a_type, a_field, a_node) \ + ((bool) (((uintptr_t) (a_node)->a_field.rbn_right_red) \ + & ((size_t)1))) +#define rbp_color_set(a_type, a_field, a_node, a_red) do { \ + (a_node)->a_field.rbn_right_red = (a_type *) ((((intptr_t) \ + (a_node)->a_field.rbn_right_red) & ((ssize_t)-2)) \ + | ((ssize_t)a_red)); \ +} while (0) +#define rbp_red_set(a_type, a_field, a_node) do { \ + (a_node)->a_field.rbn_right_red = (a_type *) (((uintptr_t) \ + (a_node)->a_field.rbn_right_red) | ((size_t)1)); \ +} while (0) +#define rbp_black_set(a_type, a_field, a_node) do { \ + (a_node)->a_field.rbn_right_red = (a_type *) (((intptr_t) \ + (a_node)->a_field.rbn_right_red) & ((ssize_t)-2)); \ +} while (0) + +/* Node initializer. */ +#define rbp_node_new(a_type, a_field, a_tree, a_node) do { \ + rbp_left_set(a_type, a_field, (a_node), &(a_tree)->rbt_nil); \ + rbp_right_set(a_type, a_field, (a_node), &(a_tree)->rbt_nil); \ + rbp_red_set(a_type, a_field, (a_node)); \ +} while (0) + +/* Tree initializer. */ +#define rb_new(a_type, a_field, a_tree) do { \ + (a_tree)->rbt_root = &(a_tree)->rbt_nil; \ + rbp_node_new(a_type, a_field, a_tree, &(a_tree)->rbt_nil); \ + rbp_black_set(a_type, a_field, &(a_tree)->rbt_nil); \ +} while (0) + +/* Tree operations. */ +#define rbp_black_height(a_type, a_field, a_tree, r_height) do { \ + a_type *rbp_bh_t; \ + for (rbp_bh_t = (a_tree)->rbt_root, (r_height) = 0; \ + rbp_bh_t != &(a_tree)->rbt_nil; \ + rbp_bh_t = rbp_left_get(a_type, a_field, rbp_bh_t)) { \ + if (rbp_red_get(a_type, a_field, rbp_bh_t) == false) { \ + (r_height)++; \ + } \ + } \ +} while (0) + +#define rbp_first(a_type, a_field, a_tree, a_root, r_node) do { \ + for ((r_node) = (a_root); \ + rbp_left_get(a_type, a_field, (r_node)) != &(a_tree)->rbt_nil; \ + (r_node) = rbp_left_get(a_type, a_field, (r_node))) { \ + } \ +} while (0) + +#define rbp_last(a_type, a_field, a_tree, a_root, r_node) do { \ + for ((r_node) = (a_root); \ + rbp_right_get(a_type, a_field, (r_node)) != &(a_tree)->rbt_nil; \ + (r_node) = rbp_right_get(a_type, a_field, (r_node))) { \ + } \ +} while (0) + +#define rbp_next(a_type, a_field, a_cmp, a_tree, a_node, r_node) do { \ + if (rbp_right_get(a_type, a_field, (a_node)) \ + != &(a_tree)->rbt_nil) { \ + rbp_first(a_type, a_field, a_tree, rbp_right_get(a_type, \ + a_field, (a_node)), (r_node)); \ + } else { \ + a_type *rbp_n_t = (a_tree)->rbt_root; \ + assert(rbp_n_t != &(a_tree)->rbt_nil); \ + (r_node) = &(a_tree)->rbt_nil; \ + while (true) { \ + int rbp_n_cmp = (a_cmp)((a_node), rbp_n_t); \ + if (rbp_n_cmp < 0) { \ + (r_node) = rbp_n_t; \ + rbp_n_t = rbp_left_get(a_type, a_field, rbp_n_t); \ + } else if (rbp_n_cmp > 0) { \ + rbp_n_t = rbp_right_get(a_type, a_field, rbp_n_t); \ + } else { \ + break; \ + } \ + assert(rbp_n_t != &(a_tree)->rbt_nil); \ + } \ + } \ +} while (0) + +#define rbp_prev(a_type, a_field, a_cmp, a_tree, a_node, r_node) do { \ + if (rbp_left_get(a_type, a_field, (a_node)) != &(a_tree)->rbt_nil) {\ + rbp_last(a_type, a_field, a_tree, rbp_left_get(a_type, \ + a_field, (a_node)), (r_node)); \ + } else { \ + a_type *rbp_p_t = (a_tree)->rbt_root; \ + assert(rbp_p_t != &(a_tree)->rbt_nil); \ + (r_node) = &(a_tree)->rbt_nil; \ + while (true) { \ + int rbp_p_cmp = (a_cmp)((a_node), rbp_p_t); \ + if (rbp_p_cmp < 0) { \ + rbp_p_t = rbp_left_get(a_type, a_field, rbp_p_t); \ + } else if (rbp_p_cmp > 0) { \ + (r_node) = rbp_p_t; \ + rbp_p_t = rbp_right_get(a_type, a_field, rbp_p_t); \ + } else { \ + break; \ + } \ + assert(rbp_p_t != &(a_tree)->rbt_nil); \ + } \ + } \ +} while (0) + +#define rb_first(a_type, a_field, a_tree, r_node) do { \ + rbp_first(a_type, a_field, a_tree, (a_tree)->rbt_root, (r_node)); \ + if ((r_node) == &(a_tree)->rbt_nil) { \ + (r_node) = NULL; \ + } \ +} while (0) + +#define rb_last(a_type, a_field, a_tree, r_node) do { \ + rbp_last(a_type, a_field, a_tree, (a_tree)->rbt_root, r_node); \ + if ((r_node) == &(a_tree)->rbt_nil) { \ + (r_node) = NULL; \ + } \ +} while (0) + +#define rb_next(a_type, a_field, a_cmp, a_tree, a_node, r_node) do { \ + rbp_next(a_type, a_field, a_cmp, a_tree, (a_node), (r_node)); \ + if ((r_node) == &(a_tree)->rbt_nil) { \ + (r_node) = NULL; \ + } \ +} while (0) + +#define rb_prev(a_type, a_field, a_cmp, a_tree, a_node, r_node) do { \ + rbp_prev(a_type, a_field, a_cmp, a_tree, (a_node), (r_node)); \ + if ((r_node) == &(a_tree)->rbt_nil) { \ + (r_node) = NULL; \ + } \ +} while (0) + +#define rb_search(a_type, a_field, a_cmp, a_tree, a_key, r_node) do { \ + int rbp_se_cmp; \ + (r_node) = (a_tree)->rbt_root; \ + while ((r_node) != &(a_tree)->rbt_nil \ + && (rbp_se_cmp = (a_cmp)((a_key), (r_node))) != 0) { \ + if (rbp_se_cmp < 0) { \ + (r_node) = rbp_left_get(a_type, a_field, (r_node)); \ + } else { \ + (r_node) = rbp_right_get(a_type, a_field, (r_node)); \ + } \ + } \ + if ((r_node) == &(a_tree)->rbt_nil) { \ + (r_node) = NULL; \ + } \ +} while (0) + +/* + * Find a match if it exists. Otherwise, find the next greater node, if one + * exists. + */ +#define rb_nsearch(a_type, a_field, a_cmp, a_tree, a_key, r_node) do { \ + a_type *rbp_ns_t = (a_tree)->rbt_root; \ + (r_node) = NULL; \ + while (rbp_ns_t != &(a_tree)->rbt_nil) { \ + int rbp_ns_cmp = (a_cmp)((a_key), rbp_ns_t); \ + if (rbp_ns_cmp < 0) { \ + (r_node) = rbp_ns_t; \ + rbp_ns_t = rbp_left_get(a_type, a_field, rbp_ns_t); \ + } else if (rbp_ns_cmp > 0) { \ + rbp_ns_t = rbp_right_get(a_type, a_field, rbp_ns_t); \ + } else { \ + (r_node) = rbp_ns_t; \ + break; \ + } \ + } \ +} while (0) + +/* + * Find a match if it exists. Otherwise, find the previous lesser node, if one + * exists. + */ +#define rb_psearch(a_type, a_field, a_cmp, a_tree, a_key, r_node) do { \ + a_type *rbp_ps_t = (a_tree)->rbt_root; \ + (r_node) = NULL; \ + while (rbp_ps_t != &(a_tree)->rbt_nil) { \ + int rbp_ps_cmp = (a_cmp)((a_key), rbp_ps_t); \ + if (rbp_ps_cmp < 0) { \ + rbp_ps_t = rbp_left_get(a_type, a_field, rbp_ps_t); \ + } else if (rbp_ps_cmp > 0) { \ + (r_node) = rbp_ps_t; \ + rbp_ps_t = rbp_right_get(a_type, a_field, rbp_ps_t); \ + } else { \ + (r_node) = rbp_ps_t; \ + break; \ + } \ + } \ +} while (0) + +#define rbp_rotate_left(a_type, a_field, a_node, r_node) do { \ + (r_node) = rbp_right_get(a_type, a_field, (a_node)); \ + rbp_right_set(a_type, a_field, (a_node), \ + rbp_left_get(a_type, a_field, (r_node))); \ + rbp_left_set(a_type, a_field, (r_node), (a_node)); \ +} while (0) + +#define rbp_rotate_right(a_type, a_field, a_node, r_node) do { \ + (r_node) = rbp_left_get(a_type, a_field, (a_node)); \ + rbp_left_set(a_type, a_field, (a_node), \ + rbp_right_get(a_type, a_field, (r_node))); \ + rbp_right_set(a_type, a_field, (r_node), (a_node)); \ +} while (0) + +#define rbp_lean_left(a_type, a_field, a_node, r_node) do { \ + bool rbp_ll_red; \ + rbp_rotate_left(a_type, a_field, (a_node), (r_node)); \ + rbp_ll_red = rbp_red_get(a_type, a_field, (a_node)); \ + rbp_color_set(a_type, a_field, (r_node), rbp_ll_red); \ + rbp_red_set(a_type, a_field, (a_node)); \ +} while (0) + +#define rbp_lean_right(a_type, a_field, a_node, r_node) do { \ + bool rbp_lr_red; \ + rbp_rotate_right(a_type, a_field, (a_node), (r_node)); \ + rbp_lr_red = rbp_red_get(a_type, a_field, (a_node)); \ + rbp_color_set(a_type, a_field, (r_node), rbp_lr_red); \ + rbp_red_set(a_type, a_field, (a_node)); \ +} while (0) + +#define rbp_move_red_left(a_type, a_field, a_node, r_node) do { \ + a_type *rbp_mrl_t, *rbp_mrl_u; \ + rbp_mrl_t = rbp_left_get(a_type, a_field, (a_node)); \ + rbp_red_set(a_type, a_field, rbp_mrl_t); \ + rbp_mrl_t = rbp_right_get(a_type, a_field, (a_node)); \ + rbp_mrl_u = rbp_left_get(a_type, a_field, rbp_mrl_t); \ + if (rbp_red_get(a_type, a_field, rbp_mrl_u)) { \ + rbp_rotate_right(a_type, a_field, rbp_mrl_t, rbp_mrl_u); \ + rbp_right_set(a_type, a_field, (a_node), rbp_mrl_u); \ + rbp_rotate_left(a_type, a_field, (a_node), (r_node)); \ + rbp_mrl_t = rbp_right_get(a_type, a_field, (a_node)); \ + if (rbp_red_get(a_type, a_field, rbp_mrl_t)) { \ + rbp_black_set(a_type, a_field, rbp_mrl_t); \ + rbp_red_set(a_type, a_field, (a_node)); \ + rbp_rotate_left(a_type, a_field, (a_node), rbp_mrl_t); \ + rbp_left_set(a_type, a_field, (r_node), rbp_mrl_t); \ + } else { \ + rbp_black_set(a_type, a_field, (a_node)); \ + } \ + } else { \ + rbp_red_set(a_type, a_field, (a_node)); \ + rbp_rotate_left(a_type, a_field, (a_node), (r_node)); \ + } \ +} while (0) + +#define rbp_move_red_right(a_type, a_field, a_node, r_node) do { \ + a_type *rbp_mrr_t; \ + rbp_mrr_t = rbp_left_get(a_type, a_field, (a_node)); \ + if (rbp_red_get(a_type, a_field, rbp_mrr_t)) { \ + a_type *rbp_mrr_u, *rbp_mrr_v; \ + rbp_mrr_u = rbp_right_get(a_type, a_field, rbp_mrr_t); \ + rbp_mrr_v = rbp_left_get(a_type, a_field, rbp_mrr_u); \ + if (rbp_red_get(a_type, a_field, rbp_mrr_v)) { \ + rbp_color_set(a_type, a_field, rbp_mrr_u, \ + rbp_red_get(a_type, a_field, (a_node))); \ + rbp_black_set(a_type, a_field, rbp_mrr_v); \ + rbp_rotate_left(a_type, a_field, rbp_mrr_t, rbp_mrr_u); \ + rbp_left_set(a_type, a_field, (a_node), rbp_mrr_u); \ + rbp_rotate_right(a_type, a_field, (a_node), (r_node)); \ + rbp_rotate_left(a_type, a_field, (a_node), rbp_mrr_t); \ + rbp_right_set(a_type, a_field, (r_node), rbp_mrr_t); \ + } else { \ + rbp_color_set(a_type, a_field, rbp_mrr_t, \ + rbp_red_get(a_type, a_field, (a_node))); \ + rbp_red_set(a_type, a_field, rbp_mrr_u); \ + rbp_rotate_right(a_type, a_field, (a_node), (r_node)); \ + rbp_rotate_left(a_type, a_field, (a_node), rbp_mrr_t); \ + rbp_right_set(a_type, a_field, (r_node), rbp_mrr_t); \ + } \ + rbp_red_set(a_type, a_field, (a_node)); \ + } else { \ + rbp_red_set(a_type, a_field, rbp_mrr_t); \ + rbp_mrr_t = rbp_left_get(a_type, a_field, rbp_mrr_t); \ + if (rbp_red_get(a_type, a_field, rbp_mrr_t)) { \ + rbp_black_set(a_type, a_field, rbp_mrr_t); \ + rbp_rotate_right(a_type, a_field, (a_node), (r_node)); \ + rbp_rotate_left(a_type, a_field, (a_node), rbp_mrr_t); \ + rbp_right_set(a_type, a_field, (r_node), rbp_mrr_t); \ + } else { \ + rbp_rotate_left(a_type, a_field, (a_node), (r_node)); \ + } \ + } \ +} while (0) + +#define rb_insert(a_type, a_field, a_cmp, a_tree, a_node) do { \ + a_type rbp_i_s; \ + a_type *rbp_i_g, *rbp_i_p, *rbp_i_c, *rbp_i_t, *rbp_i_u; \ + int rbp_i_cmp = 0; \ + rbp_i_g = &(a_tree)->rbt_nil; \ + rbp_left_set(a_type, a_field, &rbp_i_s, (a_tree)->rbt_root); \ + rbp_right_set(a_type, a_field, &rbp_i_s, &(a_tree)->rbt_nil); \ + rbp_black_set(a_type, a_field, &rbp_i_s); \ + rbp_i_p = &rbp_i_s; \ + rbp_i_c = (a_tree)->rbt_root; \ + /* Iteratively search down the tree for the insertion point, */\ + /* splitting 4-nodes as they are encountered. At the end of each */\ + /* iteration, rbp_i_g->rbp_i_p->rbp_i_c is a 3-level path down */\ + /* the tree, assuming a sufficiently deep tree. */\ + while (rbp_i_c != &(a_tree)->rbt_nil) { \ + rbp_i_t = rbp_left_get(a_type, a_field, rbp_i_c); \ + rbp_i_u = rbp_left_get(a_type, a_field, rbp_i_t); \ + if (rbp_red_get(a_type, a_field, rbp_i_t) \ + && rbp_red_get(a_type, a_field, rbp_i_u)) { \ + /* rbp_i_c is the top of a logical 4-node, so split it. */\ + /* This iteration does not move down the tree, due to the */\ + /* disruptiveness of node splitting. */\ + /* */\ + /* Rotate right. */\ + rbp_rotate_right(a_type, a_field, rbp_i_c, rbp_i_t); \ + /* Pass red links up one level. */\ + rbp_i_u = rbp_left_get(a_type, a_field, rbp_i_t); \ + rbp_black_set(a_type, a_field, rbp_i_u); \ + if (rbp_left_get(a_type, a_field, rbp_i_p) == rbp_i_c) { \ + rbp_left_set(a_type, a_field, rbp_i_p, rbp_i_t); \ + rbp_i_c = rbp_i_t; \ + } else { \ + /* rbp_i_c was the right child of rbp_i_p, so rotate */\ + /* left in order to maintain the left-leaning */\ + /* invariant. */\ + assert(rbp_right_get(a_type, a_field, rbp_i_p) \ + == rbp_i_c); \ + rbp_right_set(a_type, a_field, rbp_i_p, rbp_i_t); \ + rbp_lean_left(a_type, a_field, rbp_i_p, rbp_i_u); \ + if (rbp_left_get(a_type, a_field, rbp_i_g) == rbp_i_p) {\ + rbp_left_set(a_type, a_field, rbp_i_g, rbp_i_u); \ + } else { \ + assert(rbp_right_get(a_type, a_field, rbp_i_g) \ + == rbp_i_p); \ + rbp_right_set(a_type, a_field, rbp_i_g, rbp_i_u); \ + } \ + rbp_i_p = rbp_i_u; \ + rbp_i_cmp = (a_cmp)((a_node), rbp_i_p); \ + if (rbp_i_cmp < 0) { \ + rbp_i_c = rbp_left_get(a_type, a_field, rbp_i_p); \ + } else { \ + assert(rbp_i_cmp > 0); \ + rbp_i_c = rbp_right_get(a_type, a_field, rbp_i_p); \ + } \ + continue; \ + } \ + } \ + rbp_i_g = rbp_i_p; \ + rbp_i_p = rbp_i_c; \ + rbp_i_cmp = (a_cmp)((a_node), rbp_i_c); \ + if (rbp_i_cmp < 0) { \ + rbp_i_c = rbp_left_get(a_type, a_field, rbp_i_c); \ + } else { \ + assert(rbp_i_cmp > 0); \ + rbp_i_c = rbp_right_get(a_type, a_field, rbp_i_c); \ + } \ + } \ + /* rbp_i_p now refers to the node under which to insert. */\ + rbp_node_new(a_type, a_field, a_tree, (a_node)); \ + if (rbp_i_cmp > 0) { \ + rbp_right_set(a_type, a_field, rbp_i_p, (a_node)); \ + rbp_lean_left(a_type, a_field, rbp_i_p, rbp_i_t); \ + if (rbp_left_get(a_type, a_field, rbp_i_g) == rbp_i_p) { \ + rbp_left_set(a_type, a_field, rbp_i_g, rbp_i_t); \ + } else if (rbp_right_get(a_type, a_field, rbp_i_g) == rbp_i_p) {\ + rbp_right_set(a_type, a_field, rbp_i_g, rbp_i_t); \ + } \ + } else { \ + rbp_left_set(a_type, a_field, rbp_i_p, (a_node)); \ + } \ + /* Update the root and make sure that it is black. */\ + (a_tree)->rbt_root = rbp_left_get(a_type, a_field, &rbp_i_s); \ + rbp_black_set(a_type, a_field, (a_tree)->rbt_root); \ +} while (0) + +#define rb_remove(a_type, a_field, a_cmp, a_tree, a_node) do { \ + a_type rbp_r_s; \ + a_type *rbp_r_p, *rbp_r_c, *rbp_r_xp, *rbp_r_t, *rbp_r_u; \ + int rbp_r_cmp; \ + rbp_left_set(a_type, a_field, &rbp_r_s, (a_tree)->rbt_root); \ + rbp_right_set(a_type, a_field, &rbp_r_s, &(a_tree)->rbt_nil); \ + rbp_black_set(a_type, a_field, &rbp_r_s); \ + rbp_r_p = &rbp_r_s; \ + rbp_r_c = (a_tree)->rbt_root; \ + rbp_r_xp = &(a_tree)->rbt_nil; \ + /* Iterate down the tree, but always transform 2-nodes to 3- or */\ + /* 4-nodes in order to maintain the invariant that the current */\ + /* node is not a 2-node. This allows simple deletion once a leaf */\ + /* is reached. Handle the root specially though, since there may */\ + /* be no way to convert it from a 2-node to a 3-node. */\ + rbp_r_cmp = (a_cmp)((a_node), rbp_r_c); \ + if (rbp_r_cmp < 0) { \ + rbp_r_t = rbp_left_get(a_type, a_field, rbp_r_c); \ + rbp_r_u = rbp_left_get(a_type, a_field, rbp_r_t); \ + if (rbp_red_get(a_type, a_field, rbp_r_t) == false \ + && rbp_red_get(a_type, a_field, rbp_r_u) == false) { \ + /* Apply standard transform to prepare for left move. */\ + rbp_move_red_left(a_type, a_field, rbp_r_c, rbp_r_t); \ + rbp_black_set(a_type, a_field, rbp_r_t); \ + rbp_left_set(a_type, a_field, rbp_r_p, rbp_r_t); \ + rbp_r_c = rbp_r_t; \ + } else { \ + /* Move left. */\ + rbp_r_p = rbp_r_c; \ + rbp_r_c = rbp_left_get(a_type, a_field, rbp_r_c); \ + } \ + } else { \ + if (rbp_r_cmp == 0) { \ + assert((a_node) == rbp_r_c); \ + if (rbp_right_get(a_type, a_field, rbp_r_c) \ + == &(a_tree)->rbt_nil) { \ + /* Delete root node (which is also a leaf node). */\ + if (rbp_left_get(a_type, a_field, rbp_r_c) \ + != &(a_tree)->rbt_nil) { \ + rbp_lean_right(a_type, a_field, rbp_r_c, rbp_r_t); \ + rbp_right_set(a_type, a_field, rbp_r_t, \ + &(a_tree)->rbt_nil); \ + } else { \ + rbp_r_t = &(a_tree)->rbt_nil; \ + } \ + rbp_left_set(a_type, a_field, rbp_r_p, rbp_r_t); \ + } else { \ + /* This is the node we want to delete, but we will */\ + /* instead swap it with its successor and delete the */\ + /* successor. Record enough information to do the */\ + /* swap later. rbp_r_xp is the a_node's parent. */\ + rbp_r_xp = rbp_r_p; \ + rbp_r_cmp = 1; /* Note that deletion is incomplete. */\ + } \ + } \ + if (rbp_r_cmp == 1) { \ + if (rbp_red_get(a_type, a_field, rbp_left_get(a_type, \ + a_field, rbp_right_get(a_type, a_field, rbp_r_c))) \ + == false) { \ + rbp_r_t = rbp_left_get(a_type, a_field, rbp_r_c); \ + if (rbp_red_get(a_type, a_field, rbp_r_t)) { \ + /* Standard transform. */\ + rbp_move_red_right(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + } else { \ + /* Root-specific transform. */\ + rbp_red_set(a_type, a_field, rbp_r_c); \ + rbp_r_u = rbp_left_get(a_type, a_field, rbp_r_t); \ + if (rbp_red_get(a_type, a_field, rbp_r_u)) { \ + rbp_black_set(a_type, a_field, rbp_r_u); \ + rbp_rotate_right(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + rbp_rotate_left(a_type, a_field, rbp_r_c, \ + rbp_r_u); \ + rbp_right_set(a_type, a_field, rbp_r_t, \ + rbp_r_u); \ + } else { \ + rbp_red_set(a_type, a_field, rbp_r_t); \ + rbp_rotate_left(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + } \ + } \ + rbp_left_set(a_type, a_field, rbp_r_p, rbp_r_t); \ + rbp_r_c = rbp_r_t; \ + } else { \ + /* Move right. */\ + rbp_r_p = rbp_r_c; \ + rbp_r_c = rbp_right_get(a_type, a_field, rbp_r_c); \ + } \ + } \ + } \ + if (rbp_r_cmp != 0) { \ + while (true) { \ + assert(rbp_r_p != &(a_tree)->rbt_nil); \ + rbp_r_cmp = (a_cmp)((a_node), rbp_r_c); \ + if (rbp_r_cmp < 0) { \ + rbp_r_t = rbp_left_get(a_type, a_field, rbp_r_c); \ + if (rbp_r_t == &(a_tree)->rbt_nil) { \ + /* rbp_r_c now refers to the successor node to */\ + /* relocate, and rbp_r_xp/a_node refer to the */\ + /* context for the relocation. */\ + if (rbp_left_get(a_type, a_field, rbp_r_xp) \ + == (a_node)) { \ + rbp_left_set(a_type, a_field, rbp_r_xp, \ + rbp_r_c); \ + } else { \ + assert(rbp_right_get(a_type, a_field, \ + rbp_r_xp) == (a_node)); \ + rbp_right_set(a_type, a_field, rbp_r_xp, \ + rbp_r_c); \ + } \ + rbp_left_set(a_type, a_field, rbp_r_c, \ + rbp_left_get(a_type, a_field, (a_node))); \ + rbp_right_set(a_type, a_field, rbp_r_c, \ + rbp_right_get(a_type, a_field, (a_node))); \ + rbp_color_set(a_type, a_field, rbp_r_c, \ + rbp_red_get(a_type, a_field, (a_node))); \ + if (rbp_left_get(a_type, a_field, rbp_r_p) \ + == rbp_r_c) { \ + rbp_left_set(a_type, a_field, rbp_r_p, \ + &(a_tree)->rbt_nil); \ + } else { \ + assert(rbp_right_get(a_type, a_field, rbp_r_p) \ + == rbp_r_c); \ + rbp_right_set(a_type, a_field, rbp_r_p, \ + &(a_tree)->rbt_nil); \ + } \ + break; \ + } \ + rbp_r_u = rbp_left_get(a_type, a_field, rbp_r_t); \ + if (rbp_red_get(a_type, a_field, rbp_r_t) == false \ + && rbp_red_get(a_type, a_field, rbp_r_u) == false) { \ + rbp_move_red_left(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + if (rbp_left_get(a_type, a_field, rbp_r_p) \ + == rbp_r_c) { \ + rbp_left_set(a_type, a_field, rbp_r_p, rbp_r_t);\ + } else { \ + rbp_right_set(a_type, a_field, rbp_r_p, \ + rbp_r_t); \ + } \ + rbp_r_c = rbp_r_t; \ + } else { \ + rbp_r_p = rbp_r_c; \ + rbp_r_c = rbp_left_get(a_type, a_field, rbp_r_c); \ + } \ + } else { \ + /* Check whether to delete this node (it has to be */\ + /* the correct node and a leaf node). */\ + if (rbp_r_cmp == 0) { \ + assert((a_node) == rbp_r_c); \ + if (rbp_right_get(a_type, a_field, rbp_r_c) \ + == &(a_tree)->rbt_nil) { \ + /* Delete leaf node. */\ + if (rbp_left_get(a_type, a_field, rbp_r_c) \ + != &(a_tree)->rbt_nil) { \ + rbp_lean_right(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + rbp_right_set(a_type, a_field, rbp_r_t, \ + &(a_tree)->rbt_nil); \ + } else { \ + rbp_r_t = &(a_tree)->rbt_nil; \ + } \ + if (rbp_left_get(a_type, a_field, rbp_r_p) \ + == rbp_r_c) { \ + rbp_left_set(a_type, a_field, rbp_r_p, \ + rbp_r_t); \ + } else { \ + rbp_right_set(a_type, a_field, rbp_r_p, \ + rbp_r_t); \ + } \ + break; \ + } else { \ + /* This is the node we want to delete, but we */\ + /* will instead swap it with its successor */\ + /* and delete the successor. Record enough */\ + /* information to do the swap later. */\ + /* rbp_r_xp is a_node's parent. */\ + rbp_r_xp = rbp_r_p; \ + } \ + } \ + rbp_r_t = rbp_right_get(a_type, a_field, rbp_r_c); \ + rbp_r_u = rbp_left_get(a_type, a_field, rbp_r_t); \ + if (rbp_red_get(a_type, a_field, rbp_r_u) == false) { \ + rbp_move_red_right(a_type, a_field, rbp_r_c, \ + rbp_r_t); \ + if (rbp_left_get(a_type, a_field, rbp_r_p) \ + == rbp_r_c) { \ + rbp_left_set(a_type, a_field, rbp_r_p, rbp_r_t);\ + } else { \ + rbp_right_set(a_type, a_field, rbp_r_p, \ + rbp_r_t); \ + } \ + rbp_r_c = rbp_r_t; \ + } else { \ + rbp_r_p = rbp_r_c; \ + rbp_r_c = rbp_right_get(a_type, a_field, rbp_r_c); \ + } \ + } \ + } \ + } \ + /* Update root. */\ + (a_tree)->rbt_root = rbp_left_get(a_type, a_field, &rbp_r_s); \ +} while (0) + +/* + * The rb_wrap() macro provides a convenient way to wrap functions around the + * cpp macros. The main benefits of wrapping are that 1) repeated macro + * expansion can cause code bloat, especially for rb_{insert,remove)(), and + * 2) type, linkage, comparison functions, etc. need not be specified at every + * call point. + */ + +#define rb_wrap(a_attr, a_prefix, a_tree_type, a_type, a_field, a_cmp) \ +a_attr void \ +a_prefix##new(a_tree_type *tree) { \ + rb_new(a_type, a_field, tree); \ +} \ +a_attr a_type * \ +a_prefix##first(a_tree_type *tree) { \ + a_type *ret; \ + rb_first(a_type, a_field, tree, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##last(a_tree_type *tree) { \ + a_type *ret; \ + rb_last(a_type, a_field, tree, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##next(a_tree_type *tree, a_type *node) { \ + a_type *ret; \ + rb_next(a_type, a_field, a_cmp, tree, node, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##prev(a_tree_type *tree, a_type *node) { \ + a_type *ret; \ + rb_prev(a_type, a_field, a_cmp, tree, node, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##search(a_tree_type *tree, a_type *key) { \ + a_type *ret; \ + rb_search(a_type, a_field, a_cmp, tree, key, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##nsearch(a_tree_type *tree, a_type *key) { \ + a_type *ret; \ + rb_nsearch(a_type, a_field, a_cmp, tree, key, ret); \ + return (ret); \ +} \ +a_attr a_type * \ +a_prefix##psearch(a_tree_type *tree, a_type *key) { \ + a_type *ret; \ + rb_psearch(a_type, a_field, a_cmp, tree, key, ret); \ + return (ret); \ +} \ +a_attr void \ +a_prefix##insert(a_tree_type *tree, a_type *node) { \ + rb_insert(a_type, a_field, a_cmp, tree, node); \ +} \ +a_attr void \ +a_prefix##remove(a_tree_type *tree, a_type *node) { \ + rb_remove(a_type, a_field, a_cmp, tree, node); \ +} + +/* + * The iterators simulate recursion via an array of pointers that store the + * current path. This is critical to performance, since a series of calls to + * rb_{next,prev}() would require time proportional to (n lg n), whereas this + * implementation only requires time proportional to (n). + * + * Since the iterators cache a path down the tree, any tree modification may + * cause the cached path to become invalid. In order to continue iteration, + * use something like the following sequence: + * + * { + * a_type *node, *tnode; + * + * rb_foreach_begin(a_type, a_field, a_tree, node) { + * ... + * rb_next(a_type, a_field, a_cmp, a_tree, node, tnode); + * rb_remove(a_type, a_field, a_cmp, a_tree, node); + * rb_foreach_next(a_type, a_field, a_cmp, a_tree, tnode); + * ... + * } rb_foreach_end(a_type, a_field, a_tree, node) + * } + * + * Note that this idiom is not advised if every iteration modifies the tree, + * since in that case there is no algorithmic complexity improvement over a + * series of rb_{next,prev}() calls, thus making the setup overhead wasted + * effort. + */ + +#define rb_foreach_begin(a_type, a_field, a_tree, a_var) { \ + /* Compute the maximum possible tree depth (3X the black height). */\ + unsigned rbp_f_height; \ + rbp_black_height(a_type, a_field, a_tree, rbp_f_height); \ + rbp_f_height *= 3; \ + { \ + /* Initialize the path to contain the left spine. */\ + a_type *rbp_f_path[rbp_f_height]; \ + a_type *rbp_f_node; \ + bool rbp_f_synced = false; \ + unsigned rbp_f_depth = 0; \ + if ((a_tree)->rbt_root != &(a_tree)->rbt_nil) { \ + rbp_f_path[rbp_f_depth] = (a_tree)->rbt_root; \ + rbp_f_depth++; \ + while ((rbp_f_node = rbp_left_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1])) != &(a_tree)->rbt_nil) { \ + rbp_f_path[rbp_f_depth] = rbp_f_node; \ + rbp_f_depth++; \ + } \ + } \ + /* While the path is non-empty, iterate. */\ + while (rbp_f_depth > 0) { \ + (a_var) = rbp_f_path[rbp_f_depth-1]; + +/* Only use if modifying the tree during iteration. */ +#define rb_foreach_next(a_type, a_field, a_cmp, a_tree, a_node) \ + /* Re-initialize the path to contain the path to a_node. */\ + rbp_f_depth = 0; \ + if (a_node != NULL) { \ + if ((a_tree)->rbt_root != &(a_tree)->rbt_nil) { \ + rbp_f_path[rbp_f_depth] = (a_tree)->rbt_root; \ + rbp_f_depth++; \ + rbp_f_node = rbp_f_path[0]; \ + while (true) { \ + int rbp_f_cmp = (a_cmp)((a_node), \ + rbp_f_path[rbp_f_depth-1]); \ + if (rbp_f_cmp < 0) { \ + rbp_f_node = rbp_left_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1]); \ + } else if (rbp_f_cmp > 0) { \ + rbp_f_node = rbp_right_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1]); \ + } else { \ + break; \ + } \ + assert(rbp_f_node != &(a_tree)->rbt_nil); \ + rbp_f_path[rbp_f_depth] = rbp_f_node; \ + rbp_f_depth++; \ + } \ + } \ + } \ + rbp_f_synced = true; + +#define rb_foreach_end(a_type, a_field, a_tree, a_var) \ + if (rbp_f_synced) { \ + rbp_f_synced = false; \ + continue; \ + } \ + /* Find the successor. */\ + if ((rbp_f_node = rbp_right_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1])) != &(a_tree)->rbt_nil) { \ + /* The successor is the left-most node in the right */\ + /* subtree. */\ + rbp_f_path[rbp_f_depth] = rbp_f_node; \ + rbp_f_depth++; \ + while ((rbp_f_node = rbp_left_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1])) != &(a_tree)->rbt_nil) { \ + rbp_f_path[rbp_f_depth] = rbp_f_node; \ + rbp_f_depth++; \ + } \ + } else { \ + /* The successor is above the current node. Unwind */\ + /* until a left-leaning edge is removed from the */\ + /* path, or the path is empty. */\ + for (rbp_f_depth--; rbp_f_depth > 0; rbp_f_depth--) { \ + if (rbp_left_get(a_type, a_field, \ + rbp_f_path[rbp_f_depth-1]) \ + == rbp_f_path[rbp_f_depth]) { \ + break; \ + } \ + } \ + } \ + } \ + } \ +} + +#define rb_foreach_reverse_begin(a_type, a_field, a_tree, a_var) { \ + /* Compute the maximum possible tree depth (3X the black height). */\ + unsigned rbp_fr_height; \ + rbp_black_height(a_type, a_field, a_tree, rbp_fr_height); \ + rbp_fr_height *= 3; \ + { \ + /* Initialize the path to contain the right spine. */\ + a_type *rbp_fr_path[rbp_fr_height]; \ + a_type *rbp_fr_node; \ + bool rbp_fr_synced = false; \ + unsigned rbp_fr_depth = 0; \ + if ((a_tree)->rbt_root != &(a_tree)->rbt_nil) { \ + rbp_fr_path[rbp_fr_depth] = (a_tree)->rbt_root; \ + rbp_fr_depth++; \ + while ((rbp_fr_node = rbp_right_get(a_type, a_field, \ + rbp_fr_path[rbp_fr_depth-1])) != &(a_tree)->rbt_nil) { \ + rbp_fr_path[rbp_fr_depth] = rbp_fr_node; \ + rbp_fr_depth++; \ + } \ + } \ + /* While the path is non-empty, iterate. */\ + while (rbp_fr_depth > 0) { \ + (a_var) = rbp_fr_path[rbp_fr_depth-1]; + +/* Only use if modifying the tree during iteration. */ +#define rb_foreach_reverse_prev(a_type, a_field, a_cmp, a_tree, a_node) \ + /* Re-initialize the path to contain the path to a_node. */\ + rbp_fr_depth = 0; \ + if (a_node != NULL) { \ + if ((a_tree)->rbt_root != &(a_tree)->rbt_nil) { \ + rbp_fr_path[rbp_fr_depth] = (a_tree)->rbt_root; \ + rbp_fr_depth++; \ + rbp_fr_node = rbp_fr_path[0]; \ + while (true) { \ + int rbp_fr_cmp = (a_cmp)((a_node), \ + rbp_fr_path[rbp_fr_depth-1]); \ + if (rbp_fr_cmp < 0) { \ + rbp_fr_node = rbp_left_get(a_type, a_field, \ + rbp_fr_path[rbp_fr_depth-1]); \ + } else if (rbp_fr_cmp > 0) { \ + rbp_fr_node = rbp_right_get(a_type, a_field,\ + rbp_fr_path[rbp_fr_depth-1]); \ + } else { \ + break; \ + } \ + assert(rbp_fr_node != &(a_tree)->rbt_nil); \ + rbp_fr_path[rbp_fr_depth] = rbp_fr_node; \ + rbp_fr_depth++; \ + } \ + } \ + } \ + rbp_fr_synced = true; + +#define rb_foreach_reverse_end(a_type, a_field, a_tree, a_var) \ + if (rbp_fr_synced) { \ + rbp_fr_synced = false; \ + continue; \ + } \ + if (rbp_fr_depth == 0) { \ + /* rb_foreach_reverse_sync() was called with a NULL */\ + /* a_node. */\ + break; \ + } \ + /* Find the predecessor. */\ + if ((rbp_fr_node = rbp_left_get(a_type, a_field, \ + rbp_fr_path[rbp_fr_depth-1])) != &(a_tree)->rbt_nil) { \ + /* The predecessor is the right-most node in the left */\ + /* subtree. */\ + rbp_fr_path[rbp_fr_depth] = rbp_fr_node; \ + rbp_fr_depth++; \ + while ((rbp_fr_node = rbp_right_get(a_type, a_field, \ + rbp_fr_path[rbp_fr_depth-1])) != &(a_tree)->rbt_nil) {\ + rbp_fr_path[rbp_fr_depth] = rbp_fr_node; \ + rbp_fr_depth++; \ + } \ + } else { \ + /* The predecessor is above the current node. Unwind */\ + /* until a right-leaning edge is removed from the */\ + /* path, or the path is empty. */\ + for (rbp_fr_depth--; rbp_fr_depth > 0; rbp_fr_depth--) {\ + if (rbp_right_get(a_type, a_field, \ + rbp_fr_path[rbp_fr_depth-1]) \ + == rbp_fr_path[rbp_fr_depth]) { \ + break; \ + } \ + } \ + } \ + } \ + } \ +} + +#endif /* RB_H_ */