Google Git
Sign in
chromium / chromium / src / 64c2848c0ca94a6f4ce822d80d2506587439eab4 / . / third_party / jemalloc / chromium / jemalloc.c
blob: 65eb0b331167f70dff6a462940ff8d2c66cac094 [file] [log] [blame]
/* -*- Mode: C; tab-width: 8; c-basic-offset: 8 -*- */
/* vim:set softtabstop=8 shiftwidth=8: */
/*-
* Copyright (C) 2006-2008 Jason Evans <jasone@FreeBSD.org>.
* 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.
*
* + 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 |
* | | | ... |
* | | | 480 |
* | | | 496 |
* | | | 512 |
* | |----------------+---------|
* | | Sub-page | 1 kB |
* | | | 2 kB |
* |=====================================|
* | 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.
*
*******************************************************************************
*/
/*
* NOTE(mbelshe): Added these defines to fit within chromium build system.
*/
#define MOZ_MEMORY_WINDOWS
#define MOZ_MEMORY
#define DONT_OVERRIDE_LIBC
/*
* 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.
*/
#ifndef MOZ_MEMORY_DEBUG
# define MALLOC_PRODUCTION
#endif
/*
* Use only one arena by default. Mozilla does not currently make extensive
* use of concurrent allocation, so the increased fragmentation associated with
* multiple arenas is not warranted.
*/
#define MOZ_MEMORY_NARENAS_DEFAULT_ONE
/*
* MALLOC_STATS enables statistics calculation, and is required for
* jemalloc_stats().
*/
#define MALLOC_STATS
#ifndef MALLOC_PRODUCTION
/*
* MALLOC_DEBUG enables assertions and other sanity checks, and disables
* inline functions.
*/
# define MALLOC_DEBUG
/* Memory filling (junk/zero). */
# define MALLOC_FILL
/* Allocation tracing. */
# ifndef MOZ_MEMORY_WINDOWS
# define MALLOC_UTRACE
# endif
/* Support optional abort() on OOM. */
# define MALLOC_XMALLOC
/* Support SYSV semantics. */
# define MALLOC_SYSV
#endif
/*
* MALLOC_VALIDATE causes malloc_usable_size() to perform some pointer
* validation. There are many possible errors that validation does not even
* attempt to detect.
*/
#define MALLOC_VALIDATE
/* Embed no-op macros that support memory allocation tracking via valgrind. */
#ifdef MOZ_VALGRIND
# define MALLOC_VALGRIND
#endif
#ifdef MALLOC_VALGRIND
# include <valgrind/valgrind.h>
#else
# define VALGRIND_MALLOCLIKE_BLOCK(addr, sizeB, rzB, is_zeroed)
# define VALGRIND_FREELIKE_BLOCK(addr, rzB)
#endif
/*
* 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 */
#if (!defined(MOZ_MEMORY_WINDOWS) && !defined(MOZ_MEMORY_DARWIN))
/*
* MALLOC_PAGEFILE causes all mmap()ed memory to be backed by temporary
* files, so that if a chunk is mapped, it is guaranteed to be swappable.
* This avoids asynchronous OOM failures that are due to VM over-commit.
*
* XXX OS X over-commits, so we should probably use mmap() instead of
* vm_allocate(), so that MALLOC_PAGEFILE works.
*/
#define MALLOC_PAGEFILE
#endif
#ifdef MALLOC_PAGEFILE
/* Write size when initializing a page file. */
# define MALLOC_PAGEFILE_WRITE_SIZE 512
#endif
#ifdef MOZ_MEMORY_LINUX
#define _GNU_SOURCE /* For mremap(2). */
#define issetugid() 0
#if 0 /* Enable in order to test decommit code on Linux. */
# define MALLOC_DECOMMIT
#endif
#endif
#ifndef MOZ_MEMORY_WINCE
#include <sys/types.h>
#include <errno.h>
#include <stdlib.h>
#endif
#include <limits.h>
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#ifdef MOZ_MEMORY_WINDOWS
#ifndef MOZ_MEMORY_WINCE
//#include <cruntime.h>
//#include <internal.h>
#include <io.h>
#else
#include <cmnintrin.h>
#include <crtdefs.h>
#define SIZE_MAX UINT_MAX
#endif
#include <windows.h>
#pragma warning( disable: 4267 4996 4146 )
#define false FALSE
#define true TRUE
#define inline __inline
#define SIZE_T_MAX SIZE_MAX
#define STDERR_FILENO 2
#define PATH_MAX MAX_PATH
#define vsnprintf _vsnprintf
#ifndef NO_TLS
static unsigned long tlsIndex = 0xffffffff;
#endif
#define __thread
#ifdef MOZ_MEMORY_WINCE
#define _pthread_self() GetCurrentThreadId()
#else
#define _pthread_self() __threadid()
#endif
#define issetugid() 0
#ifndef MOZ_MEMORY_WINCE
/* use MSVC intrinsics */
#pragma intrinsic(_BitScanForward)
static __forceinline int
ffs(int x)
{
unsigned long i;
if (_BitScanForward(&i, x) != 0)
return (i + 1);
return (0);
}
/* Implement getenv without using malloc */
static char mozillaMallocOptionsBuf[64];
#define getenv xgetenv
static char *
getenv(const char *name)
{
if (GetEnvironmentVariableA(name, (LPSTR)&mozillaMallocOptionsBuf,
sizeof(mozillaMallocOptionsBuf)) > 0)
return (mozillaMallocOptionsBuf);
return (NULL);
}
#else /* WIN CE */
#define ENOMEM 12
#define EINVAL 22
static __forceinline int
ffs(int x)
{
return 32 - _CountLeadingZeros((-x) & x);
}
#endif
typedef unsigned char uint8_t;
typedef unsigned uint32_t;
typedef unsigned long long uint64_t;
typedef unsigned long long uintmax_t;
typedef long ssize_t;
#define MALLOC_DECOMMIT
#endif
#ifndef MOZ_MEMORY_WINDOWS
#ifndef MOZ_MEMORY_SOLARIS
#include <sys/cdefs.h>
#endif
#ifndef __DECONST
# define __DECONST(type, var) ((type)(uintptr_t)(const void *)(var))
#endif
#ifndef MOZ_MEMORY
__FBSDID("$FreeBSD: head/lib/libc/stdlib/malloc.c 180599 2008-07-18 19:35:44Z jasone $");
#include "libc_private.h"
#ifdef MALLOC_DEBUG
# define _LOCK_DEBUG
#endif
#include "spinlock.h"
#include "namespace.h"
#endif
#include <sys/mman.h>
#ifndef MADV_FREE
# define MADV_FREE MADV_DONTNEED
#endif
#ifndef MAP_NOSYNC
# define MAP_NOSYNC 0
#endif
#include <sys/param.h>
#ifndef MOZ_MEMORY
#include <sys/stddef.h>
#endif
#include <sys/time.h>
#include <sys/types.h>
#ifndef MOZ_MEMORY_SOLARIS
#include <sys/sysctl.h>
#endif
#include <sys/uio.h>
#ifndef MOZ_MEMORY
#include <sys/ktrace.h> /* Must come after several other sys/ includes. */
#include <machine/atomic.h>
#include <machine/cpufunc.h>
#include <machine/vmparam.h>
#endif
#include <errno.h>
#include <limits.h>
#ifndef SIZE_T_MAX
# define SIZE_T_MAX SIZE_MAX
#endif
#include <pthread.h>
#ifdef MOZ_MEMORY_DARWIN
#define _pthread_self pthread_self
#define _pthread_mutex_init pthread_mutex_init
#define _pthread_mutex_trylock pthread_mutex_trylock
#define _pthread_mutex_lock pthread_mutex_lock
#define _pthread_mutex_unlock pthread_mutex_unlock
#endif
#include <sched.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#ifndef MOZ_MEMORY_DARWIN
#include <strings.h>
#endif
#include <unistd.h>
#ifdef MOZ_MEMORY_DARWIN
#include <libkern/OSAtomic.h>
#include <mach/mach_error.h>
#include <mach/mach_init.h>
#include <mach/vm_map.h>
#include <malloc/malloc.h>
#endif
#ifndef MOZ_MEMORY
#include "un-namespace.h"
#endif
#endif
#include "jemalloc.h"
#undef bool
#define bool jemalloc_bool
#ifdef MOZ_MEMORY_DARWIN
static const bool __isthreaded = true;
#endif
#if defined(MOZ_MEMORY_SOLARIS) && defined(MAP_ALIGN) && !defined(JEMALLOC_NEVER_USES_MAP_ALIGN)
#define JEMALLOC_USES_MAP_ALIGN /* Required on Solaris 10. Might improve performance elsewhere. */
#endif
#if defined(MOZ_MEMORY_WINCE) && !defined(MOZ_MEMORY_WINCE6)
#define JEMALLOC_USES_MAP_ALIGN /* Required for Windows CE < 6 */
#endif
#define __DECONST(type, var) ((type)(uintptr_t)(const void *)(var))
#include "qr.h"
#include "ql.h"
#ifdef MOZ_MEMORY_WINDOWS
/* MSVC++ does not support C99 variable-length arrays. */
# define RB_NO_C99_VARARRAYS
#endif
#include "rb.h"
#ifdef MALLOC_DEBUG
/* Disable inlining to make debugging easier. */
#ifdef inline
#undef inline
#endif
# define inline
#endif
/* Size of stack-allocated buffer passed to strerror_r(). */
#define STRERROR_BUF 64
/* Minimum alignment of allocations is 2^QUANTUM_2POW_MIN bytes. */
# define QUANTUM_2POW_MIN 4
#ifdef MOZ_MEMORY_SIZEOF_PTR_2POW
# define SIZEOF_PTR_2POW MOZ_MEMORY_SIZEOF_PTR_2POW
#else
# define SIZEOF_PTR_2POW 2
#endif
#define PIC
#ifndef MOZ_MEMORY_DARWIN
static const bool __isthreaded = true;
#else
# define NO_TLS
#endif
#if 0
#ifdef __i386__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 2
# define CPU_SPINWAIT __asm__ volatile("pause")
#endif
#ifdef __ia64__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 3
#endif
#ifdef __alpha__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 3
# define NO_TLS
#endif
#ifdef __sparc64__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 3
# define NO_TLS
#endif
#ifdef __amd64__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 3
# define CPU_SPINWAIT __asm__ volatile("pause")
#endif
#ifdef __arm__
# define QUANTUM_2POW_MIN 3
# define SIZEOF_PTR_2POW 2
# define NO_TLS
#endif
#ifdef __mips__
# define QUANTUM_2POW_MIN 3
# define SIZEOF_PTR_2POW 2
# define NO_TLS
#endif
#ifdef __powerpc__
# define QUANTUM_2POW_MIN 4
# define SIZEOF_PTR_2POW 2
#endif
#endif
#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_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.
*/
#if defined(MOZ_MEMORY_WINCE) && !defined(MOZ_MEMORY_WINCE6)
#define CHUNK_2POW_DEFAULT 21
#else
#define CHUNK_2POW_DEFAULT 20
#endif
/* Maximum number of dirty pages per arena. */
#define DIRTY_MAX_DEFAULT (1U << 10)
/* Default reserve chunks. */
#define RESERVE_MIN_2POW_DEFAULT 1
/*
* Default range (in chunks) between reserve_min and reserve_max, in addition
* to the mandatory one chunk per arena.
*/
#ifdef MALLOC_PAGEFILE
# define RESERVE_RANGE_2POW_DEFAULT 5
#else
# define RESERVE_RANGE_2POW_DEFAULT 0
#endif
/*
* Maximum size of L1 cache line. This is used to avoid cache line aliasing,
* so over-estimates are okay (up to a point), but under-estimates will
* negatively affect performance.
*/
#define CACHELINE_2POW 6
#define CACHELINE ((size_t)(1U << CACHELINE_2POW))
/* Smallest size class to support. */
#define TINY_MIN_2POW 1
/*
* 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 SMALL_MAX_2POW_DEFAULT 9
#define SMALL_MAX_DEFAULT (1U << SMALL_MAX_2POW_DEFAULT)
/*
* 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_2POW 15
#define RUN_MAX_SMALL (1U << RUN_MAX_SMALL_2POW)
/*
* 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_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.
*/
#if defined(MOZ_MEMORY_WINDOWS)
#define malloc_mutex_t CRITICAL_SECTION
#define malloc_spinlock_t CRITICAL_SECTION
#elif defined(MOZ_MEMORY_DARWIN)
typedef struct {
OSSpinLock lock;
} malloc_mutex_t;
typedef struct {
OSSpinLock lock;
} malloc_spinlock_t;
#elif defined(MOZ_MEMORY)
typedef pthread_mutex_t malloc_mutex_t;
typedef pthread_mutex_t malloc_spinlock_t;
#else
/* XXX these should #ifdef these for freebsd (and linux?) only */
typedef struct {
spinlock_t lock;
} malloc_mutex_t;
typedef malloc_spinlock_t malloc_mutex_t;
#endif
/* Set to true once the allocator has been initialized. */
static bool malloc_initialized = false;
#if defined(MOZ_MEMORY_WINDOWS)
/* No init lock for Windows. */
#elif defined(MOZ_MEMORY_DARWIN)
static malloc_mutex_t init_lock = {OS_SPINLOCK_INIT};
#elif defined(MOZ_MEMORY_LINUX)
static malloc_mutex_t init_lock = PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP;
#elif defined(MOZ_MEMORY)
static malloc_mutex_t init_lock = PTHREAD_MUTEX_INITIALIZER;
#else
static malloc_mutex_t init_lock = {_SPINLOCK_INITIALIZER};
#endif
/******************************************************************************/
/*
* 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;
/* 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;
#ifdef MALLOC_DECOMMIT
/*
* Total number of decommit/commit operations, and total number of
* pages decommitted.
*/
uint64_t ndecommit;
uint64_t ncommit;
uint64_t decommitted;
#endif
/* 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 {
/* Linkage for the size/address-ordered tree. */
rb_node(extent_node_t) link_szad;
/* 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;
/******************************************************************************/
/*
* Radix tree data structures.
*/
#ifdef MALLOC_VALIDATE
/*
* Size of each radix tree node (must be a power of 2). This impacts tree
* depth.
*/
# if (SIZEOF_PTR == 4)
# define MALLOC_RTREE_NODESIZE (1U << 14)
# else
# define MALLOC_RTREE_NODESIZE CACHELINE
# endif
typedef struct malloc_rtree_s malloc_rtree_t;
struct malloc_rtree_s {
malloc_spinlock_t lock;
void **root;
unsigned height;
unsigned level2bits[1]; /* Dynamically sized. */
};
#endif
/******************************************************************************/
/*
* Reserve data structures.
*/
/* Callback registration. */
typedef struct reserve_reg_s reserve_reg_t;
struct reserve_reg_s {
/* Linkage for list of all registered callbacks. */
ql_elm(reserve_reg_t) link;
/* Callback function pointer. */
reserve_cb_t *cb;
/* Opaque application data pointer. */
void *ctx;
/*
* Sequence number of condition notification most recently sent to this
* callback.
*/
uint64_t seq;
};
/******************************************************************************/
/*
* 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):
*
* ???????? ???????? ????---- --ckdzla
*
* ? : 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.
* c : decommitted?
* 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
* [cdzla] : bit set
*
* Unallocated:
* ssssssss ssssssss ssss---- --c-----
* 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;
#ifdef MALLOC_DECOMMIT
#define CHUNK_MAP_DECOMMITTED ((size_t)0x20U)
#endif
#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. */
#ifdef MOZ_MEMORY
malloc_spinlock_t lock;
#else
pthread_mutex_t lock;
#endif
#ifdef MALLOC_STATS
arena_stats_t stats;
#endif
/*
* Chunk allocation sequence number, used to detect races with other
* threads during chunk allocation, and then discard unnecessary chunks.
*/
uint64_t chunk_seq;
/* 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. */
};
/******************************************************************************/
/*
* 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. */
static size_t bin_maxclass; /* Max size class for bins. */
static unsigned ntbins; /* Number of (2^n)-spaced tiny bins. */
static unsigned nqbins; /* Number of quantum-spaced bins. */
static unsigned nsbins; /* Number of (2^n)-spaced sub-page bins. */
static size_t small_min;
static size_t small_max;
/* Various quantum-related settings. */
static size_t quantum;
static size_t quantum_mask; /* (quantum - 1). */
/* 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.
*/
#ifdef MALLOC_VALIDATE
static malloc_rtree_t *chunk_rtree;
#endif
/* 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_STATS
/* Huge allocation statistics. */
static uint64_t huge_nmalloc;
static uint64_t huge_ndalloc;
static size_t huge_allocated;
#endif
/****************/
/*
* Memory reserve.
*/
#ifdef MALLOC_PAGEFILE
static char pagefile_templ[PATH_MAX];
#endif
/* Protects reserve-related data structures. */
static malloc_mutex_t reserve_mtx;
/*
* Bounds on acceptable reserve size, and current reserve size. Reserve
* depletion may cause (reserve_cur < reserve_min).
*/
static size_t reserve_min;
static size_t reserve_cur;
static size_t reserve_max;
/* List of registered callbacks. */
static ql_head(reserve_reg_t) reserve_regs;
/*
* Condition notification sequence number, used to determine whether all
* registered callbacks have been notified of the most current condition.
*/
static uint64_t reserve_seq;
/*
* Trees of chunks currently in the memory reserve. Depending on function,
* different tree orderings are needed, which is why there are two trees with
* the same contents.
*/
static extent_tree_t reserve_chunks_szad;
static extent_tree_t reserve_chunks_ad;
/****************************/
/*
* 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;
#ifdef MALLOC_DECOMMIT
static void *base_next_decommitted;
#endif
static void *base_past_addr; /* Addr immediately past base_pages. */
static extent_node_t *base_nodes;
static reserve_reg_t *base_reserve_regs;
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;
static unsigned narenas_2pow;
#ifndef NO_TLS
# ifdef MALLOC_BALANCE
static unsigned narenas_2pow;
# else
static unsigned next_arena;
# endif
#endif
#ifdef MOZ_MEMORY
static malloc_spinlock_t arenas_lock; /* Protects arenas initialization. */
#else
static pthread_mutex_t arenas_lock; /* Protects arenas initialization. */
#endif
#ifndef NO_TLS
/*
* Map of pthread_self() --> arenas[???], used for selecting an arena to use
* for allocations.
*/
#ifndef MOZ_MEMORY_WINDOWS
static __thread arena_t *arenas_map;
#endif
#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;
#ifdef MALLOC_FILL
static bool opt_junk = true;
#endif
#else
static bool opt_abort = false;
#ifdef MALLOC_FILL
static bool opt_junk = false;
#endif
#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_quantum_2pow = QUANTUM_2POW_MIN;
static size_t opt_small_max_2pow = SMALL_MAX_2POW_DEFAULT;
static size_t opt_chunk_2pow = CHUNK_2POW_DEFAULT;
static int opt_reserve_min_lshift = 0;
static int opt_reserve_range_lshift = 0;
#ifdef MALLOC_PAGEFILE
static bool opt_pagefile = false;
#endif
#ifdef MALLOC_UTRACE
static bool opt_utrace = false;
#endif
#ifdef MALLOC_SYSV
static bool opt_sysv = false;
#endif
#ifdef MALLOC_XMALLOC
static bool opt_xmalloc = false;
#endif
#ifdef MALLOC_FILL
static bool opt_zero = false;
#endif
static int opt_narenas_lshift = 0;
#ifdef MALLOC_UTRACE
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)); \
}
#else
#define UTRACE(a, b, c)
#endif
/******************************************************************************/
/*
* Begin function prototypes for non-inline static functions.
*/
static char *umax2s(uintmax_t x, char *s);
static bool malloc_mutex_init(malloc_mutex_t *mutex);
static bool malloc_spin_init(malloc_spinlock_t *lock);
static void wrtmessage(const char *p1, const char *p2, const char *p3,
const char *p4);
#ifdef MALLOC_STATS
#ifdef MOZ_MEMORY_DARWIN
/* Avoid namespace collision with OS X's malloc APIs. */
#define malloc_printf moz_malloc_printf
#endif
static void malloc_printf(const char *format, ...);
#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);
static reserve_reg_t *base_reserve_reg_alloc(void);
static void base_reserve_reg_dealloc(reserve_reg_t *reg);
#ifdef MALLOC_STATS
static void stats_print(arena_t *arena);
#endif
static void *pages_map(void *addr, size_t size, int pfd);
static void pages_unmap(void *addr, size_t size);
static void *chunk_alloc_mmap(size_t size, bool pagefile);
#ifdef MALLOC_PAGEFILE
static int pagefile_init(size_t size);
static void pagefile_close(int pfd);
#endif
static void *chunk_recycle_reserve(size_t size, bool zero);
static void *chunk_alloc(size_t size, bool zero, bool pagefile);
static extent_node_t *chunk_dealloc_reserve(void *chunk, size_t size);
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 void arena_chunk_init(arena_t *arena, arena_chunk_t *chunk);
static void arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk);
static arena_run_t *arena_run_alloc(arena_t *arena, arena_bin_t *bin,
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
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);
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);
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);
#ifndef MOZ_MEMORY_WINDOWS
static
#endif
bool malloc_init_hard(void);
static void reserve_shrink(void);
static uint64_t reserve_notify(reserve_cnd_t cnd, size_t size, uint64_t seq);
static uint64_t reserve_crit(size_t size, const char *fname, uint64_t seq);
static void reserve_fail(size_t size, const char *fname);
void _malloc_prefork(void);
void _malloc_postfork(void);
/*
* End function prototypes.
*/
/******************************************************************************/
/*
* umax2s() provides minimal integer printing functionality, which is
* especially useful for situations where allocation in vsnprintf() calls would
* potentially cause deadlock.
*/
#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]);
}
static void
wrtmessage(const char *p1, const char *p2, const char *p3, const char *p4)
{
#ifdef MOZ_MEMORY_WINCE
wchar_t buf[1024];
#define WRT_PRINT(s) \
MultiByteToWideChar(CP_ACP, 0, s, -1, buf, 1024); \
OutputDebugStringW(buf)
WRT_PRINT(p1);
WRT_PRINT(p2);
WRT_PRINT(p3);
WRT_PRINT(p4);
#else
#if defined(MOZ_MEMORY) && !defined(MOZ_MEMORY_WINDOWS)
#define _write write
#endif
_write(STDERR_FILENO, p1, (unsigned int) strlen(p1));
_write(STDERR_FILENO, p2, (unsigned int) strlen(p2));
_write(STDERR_FILENO, p3, (unsigned int) strlen(p3));
_write(STDERR_FILENO, p4, (unsigned int) strlen(p4));
#endif
}
#define _malloc_message malloc_message
void (*_malloc_message)(const char *p1, const char *p2, const char *p3,
const char *p4) = wrtmessage;
#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
/******************************************************************************/
/*
* 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 bool
malloc_mutex_init(malloc_mutex_t *mutex)
{
#if defined(MOZ_MEMORY_WINCE)
InitializeCriticalSection(mutex);
#elif defined(MOZ_MEMORY_WINDOWS)
// XXXMB
//if (__isthreaded)
// if (! __crtInitCritSecAndSpinCount(mutex, _CRT_SPINCOUNT))
// return (true);
if (!InitializeCriticalSectionAndSpinCount(mutex, 4000))
return true;
#elif defined(MOZ_MEMORY_DARWIN)
mutex->lock = OS_SPINLOCK_INIT;
#elif defined(MOZ_MEMORY_LINUX)
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);
#elif defined(MOZ_MEMORY)
if (pthread_mutex_init(mutex, NULL) != 0)
return (true);
#else
static const spinlock_t lock = _SPINLOCK_INITIALIZER;
mutex->lock = lock;
#endif
return (false);
}
static inline void
malloc_mutex_lock(malloc_mutex_t *mutex)
{
#if defined(MOZ_MEMORY_WINDOWS)
EnterCriticalSection(mutex);
#elif defined(MOZ_MEMORY_DARWIN)
OSSpinLockLock(&mutex->lock);
#elif defined(MOZ_MEMORY)
pthread_mutex_lock(mutex);
#else
if (__isthreaded)
_SPINLOCK(&mutex->lock);
#endif
}
static inline void
malloc_mutex_unlock(malloc_mutex_t *mutex)
{
#if defined(MOZ_MEMORY_WINDOWS)
LeaveCriticalSection(mutex);
#elif defined(MOZ_MEMORY_DARWIN)
OSSpinLockUnlock(&mutex->lock);
#elif defined(MOZ_MEMORY)
pthread_mutex_unlock(mutex);
#else
if (__isthreaded)
_SPINUNLOCK(&mutex->lock);
#endif
}
static bool
malloc_spin_init(malloc_spinlock_t *lock)
{
#if defined(MOZ_MEMORY_WINCE)
InitializeCriticalSection(lock);
#elif defined(MOZ_MEMORY_WINDOWS)
// XXXMB
//if (__isthreaded)
// if (! __crtInitCritSecAndSpinCount(lock, _CRT_SPINCOUNT))
// return (true);
#elif defined(MOZ_MEMORY_DARWIN)
lock->lock = OS_SPINLOCK_INIT;
#elif defined(MOZ_MEMORY_LINUX)
pthread_mutexattr_t attr;
if (pthread_mutexattr_init(&attr) != 0)
return (true);
pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ADAPTIVE_NP);
if (pthread_mutex_init(lock, &attr) != 0) {
pthread_mutexattr_destroy(&attr);
return (true);
}
pthread_mutexattr_destroy(&attr);
#elif defined(MOZ_MEMORY)
if (pthread_mutex_init(lock, NULL) != 0)
return (true);
#else
lock->lock = _SPINLOCK_INITIALIZER;
#endif
return (false);
}
static inline void
malloc_spin_lock(malloc_spinlock_t *lock)
{
#if defined(MOZ_MEMORY_WINDOWS)
EnterCriticalSection(lock);
#elif defined(MOZ_MEMORY_DARWIN)
OSSpinLockLock(&lock->lock);
#elif defined(MOZ_MEMORY)
pthread_mutex_lock(lock);
#else
if (__isthreaded)
_SPINLOCK(&lock->lock);
#endif
}
static inline void
malloc_spin_unlock(malloc_spinlock_t *lock)
{
#if defined(MOZ_MEMORY_WINDOWS)
LeaveCriticalSection(lock);
#elif defined(MOZ_MEMORY_DARWIN)
OSSpinLockUnlock(&lock->lock);
#elif defined(MOZ_MEMORY)
pthread_mutex_unlock(lock);
#else
if (__isthreaded)
_SPINUNLOCK(&lock->lock);
#endif
}
/*
* 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.
*/
#if defined(MOZ_MEMORY) && !defined(MOZ_MEMORY_DARWIN)
# define malloc_spin_init malloc_mutex_init
# define malloc_spin_lock malloc_mutex_lock
# define malloc_spin_unlock malloc_mutex_unlock
#endif
#ifndef MOZ_MEMORY
/*
* 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);
}
#endif
/*
* 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 cacheline multiple that is >= s. */
#define CACHELINE_CEILING(s) \
(((s) + (CACHELINE - 1)) & ~(CACHELINE - 1))
/* Return the smallest quantum multiple that is >= a. */
#define QUANTUM_CEILING(a) \
(((a) + quantum_mask) & ~quantum_mask)
/* Return the smallest pagesize multiple that is >= s. */
#define PAGE_CEILING(s) \
(((s) + pagesize_mask) & ~pagesize_mask)
/* 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);
}
#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_UTRACE
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 ("<jemalloc>");
}
#ifdef MALLOC_STATS
/*
* Print to stderr in such a way as to (hopefully) avoid memory allocation.
*/
static void
malloc_printf(const char *format, ...)
{
#ifndef WINCE
char buf[4096];
va_list ap;
va_start(ap, format);
vsnprintf(buf, sizeof(buf), format, ap);
va_end(ap);
_malloc_message(buf, "", "", "");
#endif
}
#endif
/******************************************************************************/
#ifdef MALLOC_DECOMMIT
static inline void
pages_decommit(void *addr, size_t size)
{
#ifdef MOZ_MEMORY_WINDOWS
VirtualFree(addr, size, MEM_DECOMMIT);
#else
if (mmap(addr, size, PROT_NONE, MAP_FIXED | MAP_PRIVATE | MAP_ANON, -1,
0) == MAP_FAILED)
abort();
#endif
}
static inline void
pages_commit(void *addr, size_t size)
{
# ifdef MOZ_MEMORY_WINDOWS
VirtualAlloc(addr, size, MEM_COMMIT, PAGE_READWRITE);
# else
if (mmap(addr, size, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_PRIVATE |
MAP_ANON, -1, 0) == MAP_FAILED)
abort();
# endif
}
#endif
static bool
base_pages_alloc_mmap(size_t minsize)
{
bool ret;
size_t csize;
#ifdef MALLOC_DECOMMIT
size_t pminsize;
#endif
int pfd;
assert(minsize != 0);
csize = CHUNK_CEILING(minsize);
#ifdef MALLOC_PAGEFILE
if (opt_pagefile) {
pfd = pagefile_init(csize);
if (pfd == -1)
return (true);
} else
#endif
pfd = -1;
base_pages = pages_map(NULL, csize, pfd);
if (base_pages == NULL) {
ret = true;
goto RETURN;
}
base_next_addr = base_pages;
base_past_addr = (void *)((uintptr_t)base_pages + csize);
#ifdef MALLOC_DECOMMIT
/*
* Leave enough pages for minsize committed, since otherwise they would
* have to be immediately recommitted.
*/
pminsize = PAGE_CEILING(minsize);
base_next_decommitted = (void *)((uintptr_t)base_pages + pminsize);
if (pminsize < csize)
pages_decommit(base_next_decommitted, csize - pminsize);
#endif
#ifdef MALLOC_STATS
base_mapped += csize;
#endif
ret = false;
RETURN:
#ifdef MALLOC_PAGEFILE
if (pfd != -1)
pagefile_close(pfd);
#endif
return (false);
}
static bool
base_pages_alloc(size_t minsize)
{
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);
#ifdef MALLOC_DECOMMIT
/* Make sure enough pages are committed for the new allocation. */
if ((uintptr_t)base_next_addr > (uintptr_t)base_next_decommitted) {
void *pbase_next_addr =
(void *)(PAGE_CEILING((uintptr_t)base_next_addr));
pages_commit(base_next_decommitted, (uintptr_t)pbase_next_addr -
(uintptr_t)base_next_decommitted);
base_next_decommitted = pbase_next_addr;
}
#endif
malloc_mutex_unlock(&base_mtx);
VALGRIND_MALLOCLIKE_BLOCK(ret, size, 0, false);
return (ret);
}
static void *
base_calloc(size_t number, size_t size)
{
void *ret;
ret = base_alloc(number * size);
#ifdef MALLOC_VALGRIND
if (ret != NULL) {
VALGRIND_FREELIKE_BLOCK(ret, 0);
VALGRIND_MALLOCLIKE_BLOCK(ret, size, 0, true);
}
#endif
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;
VALGRIND_FREELIKE_BLOCK(ret, 0);
VALGRIND_MALLOCLIKE_BLOCK(ret, sizeof(extent_node_t), 0, false);
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);
VALGRIND_FREELIKE_BLOCK(node, 0);
VALGRIND_MALLOCLIKE_BLOCK(node, sizeof(extent_node_t *), 0, false);
*(extent_node_t **)node = base_nodes;
base_nodes = node;
malloc_mutex_unlock(&base_mtx);
}
static reserve_reg_t *
base_reserve_reg_alloc(void)
{
reserve_reg_t *ret;
malloc_mutex_lock(&base_mtx);
if (base_reserve_regs != NULL) {
ret = base_reserve_regs;
base_reserve_regs = *(reserve_reg_t **)ret;
VALGRIND_FREELIKE_BLOCK(ret, 0);
VALGRIND_MALLOCLIKE_BLOCK(ret, sizeof(reserve_reg_t), 0, false);
malloc_mutex_unlock(&base_mtx);
} else {
malloc_mutex_unlock(&base_mtx);
ret = (reserve_reg_t *)base_alloc(sizeof(reserve_reg_t));
}
return (ret);
}
static void
base_reserve_reg_dealloc(reserve_reg_t *reg)
{
malloc_mutex_lock(&base_mtx);
VALGRIND_FREELIKE_BLOCK(reg, 0);
VALGRIND_MALLOCLIKE_BLOCK(reg, sizeof(reserve_reg_t *), 0, false);
*(reserve_reg_t **)reg = base_reserve_regs;
base_reserve_regs = reg;
malloc_mutex_unlock(&base_mtx);
}
/******************************************************************************/
#ifdef MALLOC_STATS
static void
stats_print(arena_t *arena)
{
unsigned i, gap_start;
#ifdef MOZ_MEMORY_WINDOWS
malloc_printf("dirty: %Iu page%s dirty, %I64u sweep%s,"
" %I64u madvise%s, %I64u 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");
# ifdef MALLOC_DECOMMIT
malloc_printf("decommit: %I64u decommit%s, %I64u commit%s,"
" %I64u page%s decommitted\n",
arena->stats.ndecommit, (arena->stats.ndecommit == 1) ? "" : "s",
arena->stats.ncommit, (arena->stats.ncommit == 1) ? "" : "s",
arena->stats.decommitted,
(arena->stats.decommitted == 1) ? "" : "s");
# endif
malloc_printf(" allocated nmalloc ndalloc\n");
malloc_printf("small: %12Iu %12I64u %12I64u\n",
arena->stats.allocated_small, arena->stats.nmalloc_small,
arena->stats.ndalloc_small);
malloc_printf("large: %12Iu %12I64u %12I64u\n",
arena->stats.allocated_large, arena->stats.nmalloc_large,
arena->stats.ndalloc_large);
malloc_printf("total: %12Iu %12I64u %12I64u\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: %12Iu\n", arena->stats.mapped);
#else
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");
# ifdef MALLOC_DECOMMIT
malloc_printf("decommit: %llu decommit%s, %llu commit%s,"
" %llu page%s decommitted\n",
arena->stats.ndecommit, (arena->stats.ndecommit == 1) ? "" : "s",
arena->stats.ncommit, (arena->stats.ncommit == 1) ? "" : "s",
arena->stats.decommitted,
(arena->stats.decommitted == 1) ? "" : "s");
# endif
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);
#endif
malloc_printf("bins: bin size regs pgs requests newruns"
" reruns maxruns curruns\n");
for (i = 0, gap_start = UINT_MAX; i < ntbins + nqbins + nsbins; i++) {
if (arena->bins[i].stats.nrequests == 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(
#if defined(MOZ_MEMORY_WINDOWS)
"%13u %1s %4u %4u %3u %9I64u %9I64u"
" %9I64u %7u %7u\n",
#else
"%13u %1s %4u %4u %3u %9llu %9llu"
" %9llu %7lu %7lu\n",
#endif
i,
i < ntbins ? "T" : i < ntbins + nqbins ? "Q" : "S",
arena->bins[i].reg_size,
arena->bins[i].nregs,
arena->bins[i].run_size >> pagesize_2pow,
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.
*/
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)
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.
*/
#ifdef MOZ_MEMORY_WINDOWS
#ifdef MOZ_MEMORY_WINCE
#define ALIGN_ADDR2OFFSET(al, ad) \
((uintptr_t)ad & (al - 1))
static void *
pages_map_align(size_t size, int pfd, size_t alignment)
{
void *ret;
int offset;
if (size % alignment)
size += (alignment - (size % alignment));
assert(size >= alignment);
ret = pages_map(NULL, size, pfd);
offset = ALIGN_ADDR2OFFSET(alignment, ret);
if (offset) {
/* try to over allocate by the ammount we're offset */
void *tmp;
pages_unmap(ret, size);
tmp = VirtualAlloc(NULL, size + alignment - offset,
MEM_RESERVE, PAGE_NOACCESS);
if (offset == ALIGN_ADDR2OFFSET(alignment, tmp))
ret = VirtualAlloc((void*)((intptr_t)tmp + alignment
- offset), size, MEM_COMMIT,
PAGE_READWRITE);
else
VirtualFree(tmp, 0, MEM_RELEASE);
offset = ALIGN_ADDR2OFFSET(alignment, ret);
if (offset) {
/* over allocate to ensure we have an aligned region */
ret = VirtualAlloc(NULL, size + alignment, MEM_RESERVE,
PAGE_NOACCESS);
offset = ALIGN_ADDR2OFFSET(alignment, ret);
ret = VirtualAlloc((void*)((intptr_t)ret +
alignment - offset),
size, MEM_COMMIT, PAGE_READWRITE);
}
}
return (ret);
}
#endif
static void *
pages_map(void *addr, size_t size, int pfd)
{
void *ret = NULL;
#if defined(MOZ_MEMORY_WINCE) && !defined(MOZ_MEMORY_WINCE6)
void *va_ret;
assert(addr == NULL);
va_ret = VirtualAlloc(addr, size, MEM_RESERVE, PAGE_NOACCESS);
if (va_ret)
ret = VirtualAlloc(va_ret, size, MEM_COMMIT, PAGE_READWRITE);
assert(va_ret == ret);
#else
ret = VirtualAlloc(addr, size, MEM_COMMIT | MEM_RESERVE,
PAGE_READWRITE);
#endif
return (ret);
}
static void
pages_unmap(void *addr, size_t size)
{
if (VirtualFree(addr, 0, MEM_RELEASE) == 0) {
#if defined(MOZ_MEMORY_WINCE) && !defined(MOZ_MEMORY_WINCE6)
if (GetLastError() == ERROR_INVALID_PARAMETER) {
MEMORY_BASIC_INFORMATION info;
VirtualQuery(addr, &info, sizeof(info));
if (VirtualFree(info.AllocationBase, 0, MEM_RELEASE))
return;
}
#endif
_malloc_message(_getprogname(),
": (malloc) Error in VirtualFree()\n", "", "");
if (opt_abort)
abort();
}
}
#elif (defined(MOZ_MEMORY_DARWIN))
static void *
pages_map(void *addr, size_t size, int pfd)
{
void *ret;
kern_return_t err;
int flags;
if (addr != NULL) {
ret = addr;
flags = 0;
} else
flags = VM_FLAGS_ANYWHERE;
err = vm_allocate((vm_map_t)mach_task_self(), (vm_address_t *)&ret,
(vm_size_t)size, flags);
if (err != KERN_SUCCESS)
ret = NULL;
assert(ret == NULL || (addr == NULL && ret != addr)
|| (addr != NULL && ret == addr));
return (ret);
}
static void
pages_unmap(void *addr, size_t size)
{
kern_return_t err;
err = vm_deallocate((vm_map_t)mach_task_self(), (vm_address_t)addr,
(vm_size_t)size);
if (err != KERN_SUCCESS) {
malloc_message(_getprogname(),
": (malloc) Error in vm_deallocate(): ",
mach_error_string(err), "\n");
if (opt_abort)
abort();
}
}
#define VM_COPY_MIN (pagesize << 5)
static inline void
pages_copy(void *dest, const void *src, size_t n)
{
assert((void *)((uintptr_t)dest & ~pagesize_mask) == dest);
assert(n >= VM_COPY_MIN);
assert((void *)((uintptr_t)src & ~pagesize_mask) == src);
vm_copy(mach_task_self(), (vm_address_t)src, (vm_size_t)n,
(vm_address_t)dest);
}
#else /* MOZ_MEMORY_DARWIN */
#ifdef JEMALLOC_USES_MAP_ALIGN
static void *
pages_map_align(size_t size, int pfd, size_t alignment)
{
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.
*/
#ifdef MALLOC_PAGEFILE
if (pfd != -1) {
ret = mmap((void *)alignment, size, PROT_READ | PROT_WRITE, MAP_PRIVATE |
MAP_NOSYNC | MAP_ALIGN, pfd, 0);
} else
#endif
{
ret = mmap((void *)alignment, size, PROT_READ | PROT_WRITE, MAP_PRIVATE |
MAP_NOSYNC | MAP_ALIGN | MAP_ANON, -1, 0);
}
assert(ret != NULL);
if (ret == MAP_FAILED)
ret = NULL;
return (ret);
}
#endif
static void *
pages_map(void *addr, size_t size, int pfd)
{
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.
*/
#ifdef MALLOC_PAGEFILE
if (pfd != -1) {
ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE |
MAP_NOSYNC, pfd, 0);
} else
#endif
{
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();
}
}
#endif
#ifdef MALLOC_VALIDATE
static inline malloc_rtree_t *
malloc_rtree_new(unsigned bits)
{
malloc_rtree_t *ret;
unsigned bits_per_level, height, i;
bits_per_level = ffs(pow2_ceil((MALLOC_RTREE_NODESIZE /
sizeof(void *)))) - 1;
height = bits / bits_per_level;
if (height * bits_per_level != bits)
height++;
assert(height * bits_per_level >= bits);
ret = (malloc_rtree_t*)base_calloc(1, sizeof(malloc_rtree_t) + (sizeof(unsigned) *
(height - 1)));
if (ret == NULL)
return (NULL);
malloc_spin_init(&ret->lock);
ret->height = height;
if (bits_per_level * height > bits)
ret->level2bits[0] = bits % bits_per_level;
else
ret->level2bits[0] = bits_per_level;
for (i = 1; i < height; i++)
ret->level2bits[i] = bits_per_level;
ret->root = (void**)base_calloc(1, sizeof(void *) << ret->level2bits[0]);
if (ret->root == NULL) {
/*
* We leak the rtree here, since there's no generic base
* deallocation.
*/
return (NULL);
}
return (ret);
}
/* The least significant bits of the key are ignored. */
static inline void *
malloc_rtree_get(malloc_rtree_t *rtree, uintptr_t key)
{
void *ret;
uintptr_t subkey;
unsigned i, lshift, height, bits;
void **node, **child;
malloc_spin_lock(&rtree->lock);
for (i = lshift = 0, height = rtree->height, node = rtree->root;
i < height - 1;
i++, lshift += bits, node = child) {
bits = rtree->level2bits[i];
subkey = (key << lshift) >> ((SIZEOF_PTR << 3) - bits);
child = (void**)node[subkey];
if (child == NULL) {
malloc_spin_unlock(&rtree->lock);
return (NULL);
}
}
/* node is a leaf, so it contains values rather than node pointers. */
bits = rtree->level2bits[i];
subkey = (key << lshift) >> ((SIZEOF_PTR << 3) - bits);
ret = node[subkey];
malloc_spin_unlock(&rtree->lock);
return (ret);
}
static inline bool
malloc_rtree_set(malloc_rtree_t *rtree, uintptr_t key, void *val)
{
uintptr_t subkey;
unsigned i, lshift, height, bits;
void **node, **child;
malloc_spin_lock(&rtree->lock);
for (i = lshift = 0, height = rtree->height, node = rtree->root;
i < height - 1;
i++, lshift += bits, node = child) {
bits = rtree->level2bits[i];
subkey = (key << lshift) >> ((SIZEOF_PTR << 3) - bits);
child = (void**)node[subkey];
if (child == NULL) {
child = (void**)base_calloc(1, sizeof(void *) <<
rtree->level2bits[i+1]);
if (child == NULL) {
malloc_spin_unlock(&rtree->lock);
return (true);
}
node[subkey] = child;
}
}
/* node is a leaf, so it contains values rather than node pointers. */
bits = rtree->level2bits[i];
subkey = (key << lshift) >> ((SIZEOF_PTR << 3) - bits);
node[subkey] = val;
malloc_spin_unlock(&rtree->lock);
return (false);
}
#endif
static void *
chunk_alloc_mmap(size_t size, bool pagefile)
{
void *ret;
#ifndef JEMALLOC_USES_MAP_ALIGN
size_t offset;
#endif
int pfd;
#ifdef MALLOC_PAGEFILE
if (opt_pagefile && pagefile) {
pfd = pagefile_init(size);
if (pfd == -1)
return (NULL);
} else
#endif
pfd = -1;
/*
* Windows requires that there be a 1:1 mapping between VM
* allocation/deallocation operations. Therefore, take care here to
* acquire the final result via one mapping operation. This means
* unmapping any preliminary result that is not correctly aligned.
*
* The MALLOC_PAGEFILE code also benefits from this mapping algorithm,
* since it reduces the number of page files.
*/
#ifdef JEMALLOC_USES_MAP_ALIGN
ret = pages_map_align(size, pfd, chunksize);
#else
ret = pages_map(NULL, size, pfd);
if (ret == NULL)
goto RETURN;
offset = CHUNK_ADDR2OFFSET(ret);
if (offset != 0) {
/* Deallocate, then try to allocate at (ret + size - offset). */
pages_unmap(ret, size);
ret = pages_map((void *)((uintptr_t)ret + size - offset), size,
pfd);
while (ret == NULL) {
/*
* Over-allocate in order to map a memory region that
* is definitely large enough.
*/
ret = pages_map(NULL, size + chunksize, -1);
if (ret == NULL)
goto RETURN;
/*
* Deallocate, then allocate the correct size, within
* the over-sized mapping.
*/
offset = CHUNK_ADDR2OFFSET(ret);
pages_unmap(ret, size + chunksize);
if (offset == 0)
ret = pages_map(ret, size, pfd);
else {
ret = pages_map((void *)((uintptr_t)ret +
chunksize - offset), size, pfd);
}
/*
* Failure here indicates a race with another thread, so
* try again.
*/
}
}
RETURN:
#endif
#ifdef MALLOC_PAGEFILE
if (pfd != -1)
pagefile_close(pfd);
#endif
#ifdef MALLOC_STATS
if (ret != NULL)
stats_chunks.nchunks += (size / chunksize);
#endif
return (ret);
}
#ifdef MALLOC_PAGEFILE
static int
pagefile_init(size_t size)
{
int ret;
size_t i;
char pagefile_path[PATH_MAX];
char zbuf[MALLOC_PAGEFILE_WRITE_SIZE];
/*
* Create a temporary file, then immediately unlink it so that it will
* not persist.
*/
strcpy(pagefile_path, pagefile_templ);
ret = mkstemp(pagefile_path);
if (ret == -1)
return (ret);
if (unlink(pagefile_path)) {
char buf[STRERROR_BUF];
strerror_r(errno, buf, sizeof(buf));
_malloc_message(_getprogname(), ": (malloc) Error in unlink(\"",
pagefile_path, "\"):");
_malloc_message(buf, "\n", "", "");
if (opt_abort)
abort();
}
/*
* Write sequential zeroes to the file in order to assure that disk
* space is committed, with minimal fragmentation. It would be
* sufficient to write one zero per disk block, but that potentially
* results in more system calls, for no real gain.
*/
memset(zbuf, 0, sizeof(zbuf));
for (i = 0; i < size; i += sizeof(zbuf)) {
if (write(ret, zbuf, sizeof(zbuf)) != sizeof(zbuf)) {
if (errno != ENOSPC) {
char buf[STRERROR_BUF];
strerror_r(errno, buf, sizeof(buf));
_malloc_message(_getprogname(),
": (malloc) Error in write(): ", buf, "\n");
if (opt_abort)
abort();
}
pagefile_close(ret);
return (-1);
}
}
return (ret);
}
static void
pagefile_close(int pfd)
{
if (close(pfd)) {
char buf[STRERROR_BUF];
strerror_r(errno, buf, sizeof(buf));
_malloc_message(_getprogname(),
": (malloc) Error in close(): ", buf, "\n");
if (opt_abort)
abort();
}
}
#endif
static void *
chunk_recycle_reserve(size_t size, bool zero)
{
extent_node_t *node, key;
#ifdef MALLOC_DECOMMIT
if (size != chunksize)
return (NULL);
#endif
key.addr = NULL;
key.size = size;
malloc_mutex_lock(&reserve_mtx);
node = extent_tree_szad_nsearch(&reserve_chunks_szad, &key);
if (node != NULL) {
void *ret = node->addr;
/* Remove node from the tree. */
extent_tree_szad_remove(&reserve_chunks_szad, node);
#ifndef MALLOC_DECOMMIT
if (node->size == size) {
#else
assert(node->size == size);
#endif
extent_tree_ad_remove(&reserve_chunks_ad, node);
base_node_dealloc(node);
#ifndef MALLOC_DECOMMIT
} else {
/*
* Insert the remainder of node's address range as a
* smaller chunk. Its position within reserve_chunks_ad
* does not change.
*/
assert(node->size > size);
node->addr = (void *)((uintptr_t)node->addr + size);
node->size -= size;
extent_tree_szad_insert(&reserve_chunks_szad, node);
}
#endif
reserve_cur -= size;
/*
* Try to replenish the reserve if this allocation depleted it.
*/
#ifndef MALLOC_DECOMMIT
if (reserve_cur < reserve_min) {
size_t diff = reserve_min - reserve_cur;
#else
while (reserve_cur < reserve_min) {
# define diff chunksize
#endif
void *chunk;
malloc_mutex_unlock(&reserve_mtx);
chunk = chunk_alloc_mmap(diff, true);
malloc_mutex_lock(&reserve_mtx);
if (chunk == NULL) {
uint64_t seq = 0;
do {
seq = reserve_notify(RESERVE_CND_LOW,
size, seq);
if (seq == 0)
goto MALLOC_OUT;
} while (reserve_cur < reserve_min);
} else {
extent_node_t *node;
node = chunk_dealloc_reserve(chunk, diff);
if (node == NULL) {
uint64_t seq = 0;
pages_unmap(chunk, diff);
do {
seq = reserve_notify(
RESERVE_CND_LOW, size, seq);
if (seq == 0)
goto MALLOC_OUT;
} while (reserve_cur < reserve_min);
}
}
}
MALLOC_OUT:
malloc_mutex_unlock(&reserve_mtx);
#ifdef MALLOC_DECOMMIT
pages_commit(ret, size);
# undef diff
#else
if (zero)
memset(ret, 0, size);
#endif
return (ret);
}
malloc_mutex_unlock(&reserve_mtx);
return (NULL);
}
static void *
chunk_alloc(size_t size, bool zero, bool pagefile)
{
void *ret;
assert(size != 0);
assert((size & chunksize_mask) == 0);
ret = chunk_recycle_reserve(size, zero);
if (ret != NULL)
goto RETURN;
ret = chunk_alloc_mmap(size, pagefile);
if (ret != NULL) {
goto RETURN;
}
/* All strategies for allocation failed. */
ret = NULL;
RETURN:
#ifdef MALLOC_STATS
if (ret != NULL)
stats_chunks.curchunks += (size / chunksize);
if (stats_chunks.curchunks > stats_chunks.highchunks)
stats_chunks.highchunks = stats_chunks.curchunks;
#endif
#ifdef MALLOC_VALIDATE
if (ret != NULL) {
if (malloc_rtree_set(chunk_rtree, (uintptr_t)ret, ret)) {
chunk_dealloc(ret, size);
return (NULL);
}
}
#endif
assert(CHUNK_ADDR2BASE(ret) == ret);
return (ret);
}
static extent_node_t *
chunk_dealloc_reserve(void *chunk, size_t size)
{
extent_node_t *node;
#ifdef MALLOC_DECOMMIT
if (size != chunksize)
return (NULL);
#else
extent_node_t *prev, key;
key.addr = (void *)((uintptr_t)chunk + size);
node = extent_tree_ad_nsearch(&reserve_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 reserve_chunks_ad, so only
* remove/insert from/into reserve_chunks_szad.
*/
extent_tree_szad_remove(&reserve_chunks_szad, node);
node->addr = chunk;
node->size += size;
extent_tree_szad_insert(&reserve_chunks_szad, node);
} else {
#endif
/* Coalescing forward failed, so insert a new node. */
node = base_node_alloc();
if (node == NULL)
return (NULL);
node->addr = chunk;
node->size = size;
extent_tree_ad_insert(&reserve_chunks_ad, node);
extent_tree_szad_insert(&reserve_chunks_szad, node);
#ifndef MALLOC_DECOMMIT
}
/* Try to coalesce backward. */
prev = extent_tree_ad_prev(&reserve_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 reserve_chunks_ad, so only
* remove/insert node from/into reserve_chunks_szad.
*/
extent_tree_szad_remove(&reserve_chunks_szad, prev);
extent_tree_ad_remove(&reserve_chunks_ad, prev);
extent_tree_szad_remove(&reserve_chunks_szad, node);
node->addr = prev->addr;
node->size += prev->size;
extent_tree_szad_insert(&reserve_chunks_szad, node);
base_node_dealloc(prev);
}
#endif
#ifdef MALLOC_DECOMMIT
pages_decommit(chunk, size);
#else
madvise(chunk, size, MADV_FREE);
#endif
reserve_cur += size;
if (reserve_cur > reserve_max)
reserve_shrink();
return (node);
}
static void
chunk_dealloc_mmap(void *chunk, size_t size)
{
pages_unmap(chunk, size);
}
static void
chunk_dealloc(void *chunk, size_t size)
{
extent_node_t *node;
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_VALIDATE
malloc_rtree_set(chunk_rtree, (uintptr_t)chunk, NULL);
#endif
/* Try to merge chunk into the reserve. */
malloc_mutex_lock(&reserve_mtx);
node = chunk_dealloc_reserve(chunk, size);
malloc_mutex_unlock(&reserve_mtx);
if (node == NULL)
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]);
}
# ifdef MOZ_MEMORY_WINDOWS
ret = (arena_t*)TlsGetValue(tlsIndex);
# else
ret = arenas_map;
# endif
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];
#ifdef MOZ_MEMORY_WINDOWS
TlsSetValue(tlsIndex, ret);
#else
arenas_map = ret;
#endif
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)
{
/*
* 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 * size_invs[(D >> QUANTUM_2POW_MIN) - 3]) >> SIZE_INV_SHIFT
*/
#define SIZE_INV_SHIFT 21
#define SIZE_INV(s) (((1U << SIZE_INV_SHIFT) / (s << QUANTUM_2POW_MIN)) + 1)
static const unsigned size_invs[] = {
SIZE_INV(3),