glibc/malloc/arena.c

934 lines
27 KiB
C
Raw Normal View History

/* Malloc implementation for multiple threads without lock contention.
Copyright (C) 2001-2015 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Wolfram Gloger <wg@malloc.de>, 2001.
The GNU C Library is free software; you can redistribute it and/or
2002-08-26 Brian Youmans <3diff@gnu.org> * crypt/crypt.c: Changed copying permission notice to Lesser GPL from Library GPL. * crypt/crypt_util.c: Likewise. * crypt/ufc.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-iteratephdr.c: Likewise. * iconv/iconvconfig.h: Likewise. * linuxthreads/Examples/ex10.c: Likewise. * linuxthreads/Examples/ex11.c: Likewise. * linuxthreads/Examples/ex13.c: Likewise. * linuxthreads/Examples/ex8.c: Likewise. * linuxthreads/Examples/ex9.c: Likewise. * linuxthreads/barrier.c: Likewise. * linuxthreads/events.c: Likewise. * linuxthreads/lockfile.c: Likewise. * linuxthreads/no-tsd.c: Likewise. * linuxthreads/pt-machine.c: Likewise. * linuxthreads/ptclock_gettime.c: Likewise. * linuxthreads/ptclock_settime.c: Likewise. * linuxthreads/rwlock.c: Likewise. * linuxthreads/sysdeps/alpha/pspinlock.c: Likewise. * linuxthreads/sysdeps/alpha/pt-machine.h: Likewise. * linuxthreads/sysdeps/arm/pspinlock.c: Likewise. * linuxthreads/sysdeps/arm/pt-machine.h: Likewise. * linuxthreads/sysdeps/cris/pspinlock.c: Likewise. * linuxthreads/sysdeps/cris/pt-machine.h: Likewise. * linuxthreads/sysdeps/hppa/pspinlock.c: Likewise. * linuxthreads/sysdeps/hppa/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/i686/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/pspinlock.c: Likewise. * linuxthreads/sysdeps/i386/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/useldt.h: Likewise. * linuxthreads/sysdeps/ia64/pspinlock.c: Likewise. * linuxthreads/sysdeps/ia64/pt-machine.h: Likewise. * linuxthreads/sysdeps/m68k/pspinlock.c: Likewise. * linuxthreads/sysdeps/m68k/pt-machine.h: Likewise. * linuxthreads/sysdeps/mips/pspinlock.c: Likewise. * linuxthreads/sysdeps/mips/pt-machine.h: Likewise. * linuxthreads/sysdeps/powerpc/pspinlock.c: Likewise. * linuxthreads/sysdeps/powerpc/pt-machine.h: Likewise. * linuxthreads/sysdeps/pthread/bits/initspin.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-lock.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-tsd.h: Likewise. * linuxthreads/sysdeps/pthread/getcpuclockid.c: Likewise. * linuxthreads/sysdeps/pthread/posix-timer.h: Likewise. * linuxthreads/sysdeps/pthread/timer_create.c: Likewise. * linuxthreads/sysdeps/pthread/timer_delete.c: Likewise. * linuxthreads/sysdeps/pthread/timer_getoverr.c: Likewise. * linuxthreads/sysdeps/pthread/timer_gettime.c: Likewise. * linuxthreads/sysdeps/pthread/timer_routines.c: Likewise. * linuxthreads/sysdeps/pthread/timer_settime.c: Likewise. * linuxthreads/sysdeps/pthread/tst-timer.c: Likewise. * linuxthreads/sysdeps/s390/pspinlock.c: Likewise. * linuxthreads/sysdeps/s390/s390-32/pt-machine.h: Likewise. * linuxthreads/sysdeps/s390/s390-64/pt-machine.h: Likewise. * linuxthreads/sysdeps/sh/pspinlock.c: Likewise. * linuxthreads/sysdeps/sh/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/sparcv9/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pt-machine.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/local_lim.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/posix_opt.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/sigthread.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/hppa/bits/initspin.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/i386/bits/posix_opt.h: Likewise. * linuxthreads/tststack.c: Likewise. * linuxthreads/unload.c: Likewise. * linuxthreads/weaks.c: Likewise. * linuxthreads/wrapsyscall.c: Likewise. * malloc/arena.c: Likewise. * malloc/hooks.c: Likewise. * malloc/malloc.c: Likewise. * posix/glob/Makefile.ami: Likewise. * posix/glob/Makefile.in: Likewise. * stdlib/gmp-impl.h: Likewise. * stdlib/gmp.h: Likewise. * sysdeps/generic/dl-iteratephdr-static.c: Likewise. * sysdeps/generic/strnlen.c: Likewise. * sysdeps/mach/hurd/powerpc/bits/sigcontext.h: Likewise. * sysdeps/mach/hurd/recvmsg.c: Likewise. * sysdeps/mach/hurd/sendmsg.c: Likewise. * sysdeps/mach/hurd/spawni.c: Likewise. * sysdeps/mach/powerpc/machine-sp.h: Likewise. * sysdeps/mach/powerpc/sysdep.h: Likewise. * sysdeps/mach/powerpc/thread_state.h: Likewise. * sysdeps/unix/bsd/bsd4.4/bits/socket.h: Likewise. * sysdeps/unix/sysv/linux/ia64/dl-iteratephdr-static.c: Likewise. * sysdeps/x86_64/gmp-mparam.h: Likewise.
2002-08-26 22:40:48 +00:00
modify it under the terms of the GNU Lesser General Public License as
published by the Free Software Foundation; either version 2.1 of the
License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
2002-08-26 Brian Youmans <3diff@gnu.org> * crypt/crypt.c: Changed copying permission notice to Lesser GPL from Library GPL. * crypt/crypt_util.c: Likewise. * crypt/ufc.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-iteratephdr.c: Likewise. * iconv/iconvconfig.h: Likewise. * linuxthreads/Examples/ex10.c: Likewise. * linuxthreads/Examples/ex11.c: Likewise. * linuxthreads/Examples/ex13.c: Likewise. * linuxthreads/Examples/ex8.c: Likewise. * linuxthreads/Examples/ex9.c: Likewise. * linuxthreads/barrier.c: Likewise. * linuxthreads/events.c: Likewise. * linuxthreads/lockfile.c: Likewise. * linuxthreads/no-tsd.c: Likewise. * linuxthreads/pt-machine.c: Likewise. * linuxthreads/ptclock_gettime.c: Likewise. * linuxthreads/ptclock_settime.c: Likewise. * linuxthreads/rwlock.c: Likewise. * linuxthreads/sysdeps/alpha/pspinlock.c: Likewise. * linuxthreads/sysdeps/alpha/pt-machine.h: Likewise. * linuxthreads/sysdeps/arm/pspinlock.c: Likewise. * linuxthreads/sysdeps/arm/pt-machine.h: Likewise. * linuxthreads/sysdeps/cris/pspinlock.c: Likewise. * linuxthreads/sysdeps/cris/pt-machine.h: Likewise. * linuxthreads/sysdeps/hppa/pspinlock.c: Likewise. * linuxthreads/sysdeps/hppa/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/i686/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/pspinlock.c: Likewise. * linuxthreads/sysdeps/i386/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/useldt.h: Likewise. * linuxthreads/sysdeps/ia64/pspinlock.c: Likewise. * linuxthreads/sysdeps/ia64/pt-machine.h: Likewise. * linuxthreads/sysdeps/m68k/pspinlock.c: Likewise. * linuxthreads/sysdeps/m68k/pt-machine.h: Likewise. * linuxthreads/sysdeps/mips/pspinlock.c: Likewise. * linuxthreads/sysdeps/mips/pt-machine.h: Likewise. * linuxthreads/sysdeps/powerpc/pspinlock.c: Likewise. * linuxthreads/sysdeps/powerpc/pt-machine.h: Likewise. * linuxthreads/sysdeps/pthread/bits/initspin.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-lock.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-tsd.h: Likewise. * linuxthreads/sysdeps/pthread/getcpuclockid.c: Likewise. * linuxthreads/sysdeps/pthread/posix-timer.h: Likewise. * linuxthreads/sysdeps/pthread/timer_create.c: Likewise. * linuxthreads/sysdeps/pthread/timer_delete.c: Likewise. * linuxthreads/sysdeps/pthread/timer_getoverr.c: Likewise. * linuxthreads/sysdeps/pthread/timer_gettime.c: Likewise. * linuxthreads/sysdeps/pthread/timer_routines.c: Likewise. * linuxthreads/sysdeps/pthread/timer_settime.c: Likewise. * linuxthreads/sysdeps/pthread/tst-timer.c: Likewise. * linuxthreads/sysdeps/s390/pspinlock.c: Likewise. * linuxthreads/sysdeps/s390/s390-32/pt-machine.h: Likewise. * linuxthreads/sysdeps/s390/s390-64/pt-machine.h: Likewise. * linuxthreads/sysdeps/sh/pspinlock.c: Likewise. * linuxthreads/sysdeps/sh/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/sparcv9/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pt-machine.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/local_lim.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/posix_opt.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/sigthread.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/hppa/bits/initspin.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/i386/bits/posix_opt.h: Likewise. * linuxthreads/tststack.c: Likewise. * linuxthreads/unload.c: Likewise. * linuxthreads/weaks.c: Likewise. * linuxthreads/wrapsyscall.c: Likewise. * malloc/arena.c: Likewise. * malloc/hooks.c: Likewise. * malloc/malloc.c: Likewise. * posix/glob/Makefile.ami: Likewise. * posix/glob/Makefile.in: Likewise. * stdlib/gmp-impl.h: Likewise. * stdlib/gmp.h: Likewise. * sysdeps/generic/dl-iteratephdr-static.c: Likewise. * sysdeps/generic/strnlen.c: Likewise. * sysdeps/mach/hurd/powerpc/bits/sigcontext.h: Likewise. * sysdeps/mach/hurd/recvmsg.c: Likewise. * sysdeps/mach/hurd/sendmsg.c: Likewise. * sysdeps/mach/hurd/spawni.c: Likewise. * sysdeps/mach/powerpc/machine-sp.h: Likewise. * sysdeps/mach/powerpc/sysdep.h: Likewise. * sysdeps/mach/powerpc/thread_state.h: Likewise. * sysdeps/unix/bsd/bsd4.4/bits/socket.h: Likewise. * sysdeps/unix/sysv/linux/ia64/dl-iteratephdr-static.c: Likewise. * sysdeps/x86_64/gmp-mparam.h: Likewise.
2002-08-26 22:40:48 +00:00
Lesser General Public License for more details.
2002-08-26 Brian Youmans <3diff@gnu.org> * crypt/crypt.c: Changed copying permission notice to Lesser GPL from Library GPL. * crypt/crypt_util.c: Likewise. * crypt/ufc.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-iteratephdr.c: Likewise. * iconv/iconvconfig.h: Likewise. * linuxthreads/Examples/ex10.c: Likewise. * linuxthreads/Examples/ex11.c: Likewise. * linuxthreads/Examples/ex13.c: Likewise. * linuxthreads/Examples/ex8.c: Likewise. * linuxthreads/Examples/ex9.c: Likewise. * linuxthreads/barrier.c: Likewise. * linuxthreads/events.c: Likewise. * linuxthreads/lockfile.c: Likewise. * linuxthreads/no-tsd.c: Likewise. * linuxthreads/pt-machine.c: Likewise. * linuxthreads/ptclock_gettime.c: Likewise. * linuxthreads/ptclock_settime.c: Likewise. * linuxthreads/rwlock.c: Likewise. * linuxthreads/sysdeps/alpha/pspinlock.c: Likewise. * linuxthreads/sysdeps/alpha/pt-machine.h: Likewise. * linuxthreads/sysdeps/arm/pspinlock.c: Likewise. * linuxthreads/sysdeps/arm/pt-machine.h: Likewise. * linuxthreads/sysdeps/cris/pspinlock.c: Likewise. * linuxthreads/sysdeps/cris/pt-machine.h: Likewise. * linuxthreads/sysdeps/hppa/pspinlock.c: Likewise. * linuxthreads/sysdeps/hppa/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/i686/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/pspinlock.c: Likewise. * linuxthreads/sysdeps/i386/pt-machine.h: Likewise. * linuxthreads/sysdeps/i386/useldt.h: Likewise. * linuxthreads/sysdeps/ia64/pspinlock.c: Likewise. * linuxthreads/sysdeps/ia64/pt-machine.h: Likewise. * linuxthreads/sysdeps/m68k/pspinlock.c: Likewise. * linuxthreads/sysdeps/m68k/pt-machine.h: Likewise. * linuxthreads/sysdeps/mips/pspinlock.c: Likewise. * linuxthreads/sysdeps/mips/pt-machine.h: Likewise. * linuxthreads/sysdeps/powerpc/pspinlock.c: Likewise. * linuxthreads/sysdeps/powerpc/pt-machine.h: Likewise. * linuxthreads/sysdeps/pthread/bits/initspin.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-lock.h: Likewise. * linuxthreads/sysdeps/pthread/bits/libc-tsd.h: Likewise. * linuxthreads/sysdeps/pthread/getcpuclockid.c: Likewise. * linuxthreads/sysdeps/pthread/posix-timer.h: Likewise. * linuxthreads/sysdeps/pthread/timer_create.c: Likewise. * linuxthreads/sysdeps/pthread/timer_delete.c: Likewise. * linuxthreads/sysdeps/pthread/timer_getoverr.c: Likewise. * linuxthreads/sysdeps/pthread/timer_gettime.c: Likewise. * linuxthreads/sysdeps/pthread/timer_routines.c: Likewise. * linuxthreads/sysdeps/pthread/timer_settime.c: Likewise. * linuxthreads/sysdeps/pthread/tst-timer.c: Likewise. * linuxthreads/sysdeps/s390/pspinlock.c: Likewise. * linuxthreads/sysdeps/s390/s390-32/pt-machine.h: Likewise. * linuxthreads/sysdeps/s390/s390-64/pt-machine.h: Likewise. * linuxthreads/sysdeps/sh/pspinlock.c: Likewise. * linuxthreads/sysdeps/sh/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc32/pt-machine.h: Likewise. * linuxthreads/sysdeps/sparc/sparc32/sparcv9/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pspinlock.c: Likewise. * linuxthreads/sysdeps/sparc/sparc64/pt-machine.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/local_lim.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/posix_opt.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/bits/sigthread.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/hppa/bits/initspin.h: Likewise. * linuxthreads/sysdeps/unix/sysv/linux/i386/bits/posix_opt.h: Likewise. * linuxthreads/tststack.c: Likewise. * linuxthreads/unload.c: Likewise. * linuxthreads/weaks.c: Likewise. * linuxthreads/wrapsyscall.c: Likewise. * malloc/arena.c: Likewise. * malloc/hooks.c: Likewise. * malloc/malloc.c: Likewise. * posix/glob/Makefile.ami: Likewise. * posix/glob/Makefile.in: Likewise. * stdlib/gmp-impl.h: Likewise. * stdlib/gmp.h: Likewise. * sysdeps/generic/dl-iteratephdr-static.c: Likewise. * sysdeps/generic/strnlen.c: Likewise. * sysdeps/mach/hurd/powerpc/bits/sigcontext.h: Likewise. * sysdeps/mach/hurd/recvmsg.c: Likewise. * sysdeps/mach/hurd/sendmsg.c: Likewise. * sysdeps/mach/hurd/spawni.c: Likewise. * sysdeps/mach/powerpc/machine-sp.h: Likewise. * sysdeps/mach/powerpc/sysdep.h: Likewise. * sysdeps/mach/powerpc/thread_state.h: Likewise. * sysdeps/unix/bsd/bsd4.4/bits/socket.h: Likewise. * sysdeps/unix/sysv/linux/ia64/dl-iteratephdr-static.c: Likewise. * sysdeps/x86_64/gmp-mparam.h: Likewise.
2002-08-26 22:40:48 +00:00
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; see the file COPYING.LIB. If
not, see <http://www.gnu.org/licenses/>. */
#include <stdbool.h>
/* Compile-time constants. */
2014-01-02 08:38:18 +00:00
#define HEAP_MIN_SIZE (32 * 1024)
#ifndef HEAP_MAX_SIZE
# ifdef DEFAULT_MMAP_THRESHOLD_MAX
# define HEAP_MAX_SIZE (2 * DEFAULT_MMAP_THRESHOLD_MAX)
# else
2014-01-02 08:38:18 +00:00
# define HEAP_MAX_SIZE (1024 * 1024) /* must be a power of two */
# endif
#endif
/* HEAP_MIN_SIZE and HEAP_MAX_SIZE limit the size of mmap()ed heaps
that are dynamically created for multi-threaded programs. The
maximum size must be a power of two, for fast determination of
which heap belongs to a chunk. It should be much larger than the
mmap threshold, so that requests with a size just below that
threshold can be fulfilled without creating too many heaps. */
/***************************************************************************/
#define top(ar_ptr) ((ar_ptr)->top)
/* A heap is a single contiguous memory region holding (coalesceable)
malloc_chunks. It is allocated with mmap() and always starts at an
address aligned to HEAP_MAX_SIZE. */
2014-01-02 08:38:18 +00:00
typedef struct _heap_info
{
mstate ar_ptr; /* Arena for this heap. */
struct _heap_info *prev; /* Previous heap. */
size_t size; /* Current size in bytes. */
2014-01-02 08:38:18 +00:00
size_t mprotect_size; /* Size in bytes that has been mprotected
PROT_READ|PROT_WRITE. */
/* Make sure the following data is properly aligned, particularly
that sizeof (heap_info) + 2 * SIZE_SZ is a multiple of
MALLOC_ALIGNMENT. */
char pad[-6 * SIZE_SZ & MALLOC_ALIGN_MASK];
} heap_info;
/* Get a compile-time error if the heap_info padding is not correct
to make alignment work as expected in sYSMALLOc. */
extern int sanity_check_heap_info_alignment[(sizeof (heap_info)
2014-01-02 08:38:18 +00:00
+ 2 * SIZE_SZ) % MALLOC_ALIGNMENT
? -1 : 1];
/* Thread specific data */
static tsd_key_t arena_key;
static mutex_t list_lock = MUTEX_INITIALIZER;
static size_t narenas = 1;
static mstate free_list;
/* Mapped memory in non-main arenas (reliable only for NO_THREADS). */
static unsigned long arena_mem;
/* Already initialized? */
int __malloc_initialized = -1;
/**************************************************************************/
/* arena_get() acquires an arena and locks the corresponding mutex.
First, try the one last locked successfully by this thread. (This
is the common case and handled with a macro for speed.) Then, loop
once over the circularly linked list of arenas. If no arena is
readily available, create a new one. In this latter case, `size'
is just a hint as to how much memory will be required immediately
in the new arena. */
#define arena_get(ptr, size) do { \
2014-01-02 08:38:18 +00:00
arena_lookup (ptr); \
arena_lock (ptr, size); \
} while (0)
#define arena_lookup(ptr) do { \
2014-01-02 08:38:18 +00:00
void *vptr = NULL; \
ptr = (mstate) tsd_getspecific (arena_key, vptr); \
} while (0)
2014-01-02 08:38:18 +00:00
#define arena_lock(ptr, size) do { \
if (ptr) \
(void) mutex_lock (&ptr->mutex); \
else \
ptr = arena_get2 (ptr, (size), NULL); \
} while (0)
/* find the heap and corresponding arena for a given ptr */
#define heap_for_ptr(ptr) \
2014-01-02 08:38:18 +00:00
((heap_info *) ((unsigned long) (ptr) & ~(HEAP_MAX_SIZE - 1)))
#define arena_for_chunk(ptr) \
2014-01-02 08:38:18 +00:00
(chunk_non_main_arena (ptr) ? heap_for_ptr (ptr)->ar_ptr : &main_arena)
/**************************************************************************/
#ifndef NO_THREADS
/* atfork support. */
2014-01-02 08:38:18 +00:00
static void *(*save_malloc_hook)(size_t __size, const void *);
2013-03-08 21:27:42 +00:00
static void (*save_free_hook) (void *__ptr, const void *);
static void *save_arena;
2014-01-02 08:38:18 +00:00
# ifdef ATFORK_MEM
ATFORK_MEM;
2014-01-02 08:38:18 +00:00
# endif
/* Magic value for the thread-specific arena pointer when
malloc_atfork() is in use. */
2014-01-02 08:38:18 +00:00
# define ATFORK_ARENA_PTR ((void *) -1)
/* The following hooks are used while the `atfork' handling mechanism
is active. */
2014-01-02 08:38:18 +00:00
static void *
malloc_atfork (size_t sz, const void *caller)
{
void *vptr = NULL;
void *victim;
2014-01-02 08:38:18 +00:00
tsd_getspecific (arena_key, vptr);
if (vptr == ATFORK_ARENA_PTR)
{
/* We are the only thread that may allocate at all. */
if (save_malloc_hook != malloc_check)
{
return _int_malloc (&main_arena, sz);
}
else
{
if (top_check () < 0)
return 0;
victim = _int_malloc (&main_arena, sz + 1);
return mem2mem_check (victim, sz);
}
}
else
{
/* Suspend the thread until the `atfork' handlers have completed.
By that time, the hooks will have been reset as well, so that
mALLOc() can be used again. */
(void) mutex_lock (&list_lock);
(void) mutex_unlock (&list_lock);
return __libc_malloc (sz);
}
}
static void
2014-01-02 08:38:18 +00:00
free_atfork (void *mem, const void *caller)
{
void *vptr = NULL;
mstate ar_ptr;
mchunkptr p; /* chunk corresponding to mem */
if (mem == 0) /* free(0) has no effect */
return;
2014-01-02 08:38:18 +00:00
p = mem2chunk (mem); /* do not bother to replicate free_check here */
2014-01-02 08:38:18 +00:00
if (chunk_is_mmapped (p)) /* release mmapped memory. */
{
munmap_chunk (p);
return;
}
2014-01-02 08:38:18 +00:00
ar_ptr = arena_for_chunk (p);
tsd_getspecific (arena_key, vptr);
_int_free (ar_ptr, p, vptr == ATFORK_ARENA_PTR);
}
/* Counter for number of times the list is locked by the same thread. */
static unsigned int atfork_recursive_cntr;
/* The following two functions are registered via thread_atfork() to
make sure that the mutexes remain in a consistent state in the
fork()ed version of a thread. Also adapt the malloc and free hooks
temporarily, because the `atfork' handler mechanism may use
malloc/free internally (e.g. in LinuxThreads). */
static void
ptmalloc_lock_all (void)
{
mstate ar_ptr;
2014-01-02 08:38:18 +00:00
if (__malloc_initialized < 1)
return;
2014-01-02 08:38:18 +00:00
if (mutex_trylock (&list_lock))
{
void *my_arena;
2014-01-02 08:38:18 +00:00
tsd_getspecific (arena_key, my_arena);
if (my_arena == ATFORK_ARENA_PTR)
2014-01-02 08:38:18 +00:00
/* This is the same thread which already locks the global list.
Just bump the counter. */
goto out;
/* This thread has to wait its turn. */
2014-01-02 08:38:18 +00:00
(void) mutex_lock (&list_lock);
}
for (ar_ptr = &main_arena;; )
{
(void) mutex_lock (&ar_ptr->mutex);
ar_ptr = ar_ptr->next;
if (ar_ptr == &main_arena)
break;
}
save_malloc_hook = __malloc_hook;
save_free_hook = __free_hook;
__malloc_hook = malloc_atfork;
__free_hook = free_atfork;
/* Only the current thread may perform malloc/free calls now. */
2014-01-02 08:38:18 +00:00
tsd_getspecific (arena_key, save_arena);
tsd_setspecific (arena_key, ATFORK_ARENA_PTR);
out:
++atfork_recursive_cntr;
}
static void
ptmalloc_unlock_all (void)
{
mstate ar_ptr;
2014-01-02 08:38:18 +00:00
if (__malloc_initialized < 1)
return;
2014-01-02 08:38:18 +00:00
if (--atfork_recursive_cntr != 0)
return;
2014-01-02 08:38:18 +00:00
tsd_setspecific (arena_key, save_arena);
__malloc_hook = save_malloc_hook;
__free_hook = save_free_hook;
2014-01-02 08:38:18 +00:00
for (ar_ptr = &main_arena;; )
{
(void) mutex_unlock (&ar_ptr->mutex);
ar_ptr = ar_ptr->next;
if (ar_ptr == &main_arena)
break;
}
(void) mutex_unlock (&list_lock);
}
# ifdef __linux__
/* In NPTL, unlocking a mutex in the child process after a
fork() is currently unsafe, whereas re-initializing it is safe and
does not leak resources. Therefore, a special atfork handler is
installed for the child. */
static void
ptmalloc_unlock_all2 (void)
{
mstate ar_ptr;
2014-01-02 08:38:18 +00:00
if (__malloc_initialized < 1)
return;
2014-01-02 08:38:18 +00:00
tsd_setspecific (arena_key, save_arena);
__malloc_hook = save_malloc_hook;
__free_hook = save_free_hook;
free_list = NULL;
2014-01-02 08:38:18 +00:00
for (ar_ptr = &main_arena;; )
{
mutex_init (&ar_ptr->mutex);
if (ar_ptr != save_arena)
{
ar_ptr->next_free = free_list;
free_list = ar_ptr;
}
ar_ptr = ar_ptr->next;
if (ar_ptr == &main_arena)
break;
}
2014-01-02 08:38:18 +00:00
mutex_init (&list_lock);
atfork_recursive_cntr = 0;
}
# else
# define ptmalloc_unlock_all2 ptmalloc_unlock_all
# endif
#endif /* !NO_THREADS */
/* Initialization routine. */
#include <string.h>
extern char **_environ;
static char *
internal_function
next_env_entry (char ***position)
{
char **current = *position;
char *result = NULL;
while (*current != NULL)
{
if (__builtin_expect ((*current)[0] == 'M', 0)
2014-01-02 08:38:18 +00:00
&& (*current)[1] == 'A'
&& (*current)[2] == 'L'
&& (*current)[3] == 'L'
&& (*current)[4] == 'O'
&& (*current)[5] == 'C'
&& (*current)[6] == '_')
{
result = &(*current)[7];
2014-01-02 08:38:18 +00:00
/* Save current position for next visit. */
*position = ++current;
2014-01-02 08:38:18 +00:00
break;
}
++current;
}
return result;
}
[BZ #77] Update. Add support for namespaces in the dynamic linker. * dlfcn/Makefile (libdl-routines): Add dlmopen. * dlfcn/Versions [libdl, GLIBC_2.3.4]: Add dlmopen. * dlfcn/dlfcn.h: Define Lmid_t, LM_ID_BASE, and LM_ID_NEWLM. Declare dlmopen. Document RTLD_DI_LMID. * dlfcn/dlinfo.c: Handle RTLD_DI_LMID. * dlfcn/dlmopen.c: New file. * dlfcn/dlopen.c: Pass new parameter to _dl_open. * dlfcn/dlopenold.c: Likewise. * elf/dl-addr.c: Adjust for removal of GL(dl_loaded). * elf/dl-caller.c: Likewise. * elf/dl-close.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-debug.c: Likewise. * elf/dl-lookup.c: Likewise. * elf/dl-sym.c: Likewise. * elf/dl-version.c: Likewise. * elf/do-lookup.h: Likewise. * elf/rtld.c: Likewise. * sysdeps/unix/sysv/linux/i386/dl-librecon.h: Likewise. * elf/dl-depsc: Likewise. Add new parameter to _dl_map_object. * elf/dl-fini.c: Call destructors in all namespaces. * elf/dl-iteratephdr.c: Compute total nloaded. Adjust for removal of GL(dl_loaded). * elf/dl-libc.c: Pass new parameter to _dl_open. Adjust for removal of GL(dl_loaded). * elf/dl-load.c (_dl_map_object_from_fd): Don't load ld.so a second time. Reuse the one from the main namespace in all others. Pass new parameter to _dl_new_object. Adjust for removal of GL(dl_loaded). * elf/dl-object.c: Take new parameter. Use it to initialize l_ns. Adjust for removal of GL(dl_loaded). * elf/dl-open.c (_dl_open): Take new parameter. Adjust for removal of GL(dl_loaded). * elf/dl-support.c: Replace global _dl_loaded etc variables with _dl_ns variable. * include/dlfcn.h: Adjust prototype of _dl_open. Define __LM_ID_CALLER. * include/link.h: Add l_real, l_ns, and l_direct_opencount elements. * sysdeps/generic/dl-tls.c: Bump TLS_STATIC_SURPLUS. Since libc is using TLS we need memory appropriate to the number of namespaces. * sysdeps/generic/ldsodefs.h (struct rtld_global): Replace _dl_loaded, _dl_nloaded, _dl_global_scope, _dl_main_searchlist, and _dl_global_scope_alloc with _dl_ns element. Define DL_NNS. Adjust prototypes of _dl_map_object and member in rtld_global_ro. * malloc/malloc.c: Include <dlfcn.h>. * malloc/arena.c (ptmalloc_init): If libc is not in primary namespace, never use brk. * elf/Makefile: Add rules to build and run tst-dlmopen1 and tst-dlmopen2. * elf/tst-dlmopen1.c: New file. * elf/tst-dlmopen1mod.c: New file. * elf/tst-dlmopen2.c: New file. * elf/dl-close.c: Improve reference counting by tracking direct loads. * elf/dl-lookup.c (add_dependency): Likewise. * elf/dl-open.c (dl_open_worker): Likewise. * elf/rtld.c (dl_main): Likewise. 2004-09-09 GOTO Masanori <gotom@debian.or.jp> [BZ #77] * elf/dl-close.c: Count down l_opencount to check not only for l_reldeps, but also l_initfini. 2004-10-13 Ulrich Drepper <drepper@redhat.com>
2004-10-14 02:08:23 +00:00
#ifdef SHARED
[BZ #77] Update. Add support for namespaces in the dynamic linker. * dlfcn/Makefile (libdl-routines): Add dlmopen. * dlfcn/Versions [libdl, GLIBC_2.3.4]: Add dlmopen. * dlfcn/dlfcn.h: Define Lmid_t, LM_ID_BASE, and LM_ID_NEWLM. Declare dlmopen. Document RTLD_DI_LMID. * dlfcn/dlinfo.c: Handle RTLD_DI_LMID. * dlfcn/dlmopen.c: New file. * dlfcn/dlopen.c: Pass new parameter to _dl_open. * dlfcn/dlopenold.c: Likewise. * elf/dl-addr.c: Adjust for removal of GL(dl_loaded). * elf/dl-caller.c: Likewise. * elf/dl-close.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-debug.c: Likewise. * elf/dl-lookup.c: Likewise. * elf/dl-sym.c: Likewise. * elf/dl-version.c: Likewise. * elf/do-lookup.h: Likewise. * elf/rtld.c: Likewise. * sysdeps/unix/sysv/linux/i386/dl-librecon.h: Likewise. * elf/dl-depsc: Likewise. Add new parameter to _dl_map_object. * elf/dl-fini.c: Call destructors in all namespaces. * elf/dl-iteratephdr.c: Compute total nloaded. Adjust for removal of GL(dl_loaded). * elf/dl-libc.c: Pass new parameter to _dl_open. Adjust for removal of GL(dl_loaded). * elf/dl-load.c (_dl_map_object_from_fd): Don't load ld.so a second time. Reuse the one from the main namespace in all others. Pass new parameter to _dl_new_object. Adjust for removal of GL(dl_loaded). * elf/dl-object.c: Take new parameter. Use it to initialize l_ns. Adjust for removal of GL(dl_loaded). * elf/dl-open.c (_dl_open): Take new parameter. Adjust for removal of GL(dl_loaded). * elf/dl-support.c: Replace global _dl_loaded etc variables with _dl_ns variable. * include/dlfcn.h: Adjust prototype of _dl_open. Define __LM_ID_CALLER. * include/link.h: Add l_real, l_ns, and l_direct_opencount elements. * sysdeps/generic/dl-tls.c: Bump TLS_STATIC_SURPLUS. Since libc is using TLS we need memory appropriate to the number of namespaces. * sysdeps/generic/ldsodefs.h (struct rtld_global): Replace _dl_loaded, _dl_nloaded, _dl_global_scope, _dl_main_searchlist, and _dl_global_scope_alloc with _dl_ns element. Define DL_NNS. Adjust prototypes of _dl_map_object and member in rtld_global_ro. * malloc/malloc.c: Include <dlfcn.h>. * malloc/arena.c (ptmalloc_init): If libc is not in primary namespace, never use brk. * elf/Makefile: Add rules to build and run tst-dlmopen1 and tst-dlmopen2. * elf/tst-dlmopen1.c: New file. * elf/tst-dlmopen1mod.c: New file. * elf/tst-dlmopen2.c: New file. * elf/dl-close.c: Improve reference counting by tracking direct loads. * elf/dl-lookup.c (add_dependency): Likewise. * elf/dl-open.c (dl_open_worker): Likewise. * elf/rtld.c (dl_main): Likewise. 2004-09-09 GOTO Masanori <gotom@debian.or.jp> [BZ #77] * elf/dl-close.c: Count down l_opencount to check not only for l_reldeps, but also l_initfini. 2004-10-13 Ulrich Drepper <drepper@redhat.com>
2004-10-14 02:08:23 +00:00
static void *
__failing_morecore (ptrdiff_t d)
{
return (void *) MORECORE_FAILURE;
}
Update. 2004-10-18 Jakub Jelinek <jakub@redhat.com> * elf/dl-libc.c (__libc_dlsym_private, __libc_register_dl_open_hook): New functions. (__libc_dlopen_mode): Call __libc_register_dl_open_hook and __libc_register_dlfcn_hook. * dlfcn/Makefile (routines, elide-routines.os): Set. Add rules to build and test tststatic2. * dlfcn/tststatic2.c: New test. * dlfcn/modstatic2.c: New test module. * dlfcn/dladdr.c: Call _dlfcn_hook from libdl.so if not NULL. Define __ prefixed routine in libc.a and in libdl.a just call it. * dlfcn/dladdr1.c: Likewise. * dlfcn/dlclose.c: Likewise. * dlfcn/dlerror.c: Likewise. * dlfcn/dlinfo.c: Likewise. * dlfcn/dlmopen.c: Likewise. * dlfcn/dlopen.c: Likewise. * dlfcn/dlopenold.c: Likewise. * dlfcn/dlsym.c: Likewise. * dlfcn/dlvsym.c: Likewise. * dlfcn/sdladdr.c: New file. * dlfcn/sdladdr1.c: New file. * dlfcn/sdlclose.c: New file. * dlfcn/sdlerror.c: New file. * dlfcn/sdlinfo.c: New file. * dlfcn/sdlopen.c: New file. * dlfcn/sdlsym.c: New file. * dlfcn/sdlvsym.c: New file. * dlfcn/Versions (libdl): Export _dlfcn_hook@GLIBC_PRIVATE. * include/dlfcn.h (DL_CALLER_DECL, DL_CALLER RETURN_ADDRESS): Define. (struct dlfcn_hook): New type. (_dlfcn_hook): New extern decl. (__dlopen, __dlclose, __dlsym, __dlerror, __dladdr, __dladdr1, __dlinfo, __dlmopen, __libc_dlsym_private, __libc_register_dl_open_hook, __libc_register_dlfcn_hook): New prototypes. (__dlvsym): Use DL_CALLER_DECL. * include/libc-symbols.h: Define libdl_hidden_proto and friends. * malloc/arena.c (_dl_open_hook): Extern decl. (ptmalloc_init): Don't call _dl_addr when dlopened from statically linked programs but don't use brk for them either.
2004-10-18 23:17:46 +00:00
extern struct dl_open_hook *_dl_open_hook;
libc_hidden_proto (_dl_open_hook);
* sysdeps/generic/libc-tls.c (__libc_setup_tls): Cope with zero ALIGN. * malloc/hooks.c [_LIBC && (USE___THREAD || (USE_TLS && !SHARED))] (malloc_starter, memalign_starter, free_starter): Don't define these. * malloc/malloc.c [_LIBC && (USE___THREAD || (USE_TLS && !SHARED))]: Don't declare them either. * malloc/arena.c (ptmalloc_init) [_LIBC && USE_TLS]: Don't call __pthread_initialize, so no need to set hooks to *_starter. (ptmalloc_init_minimal): New function, broken out of ptmalloc_init. [_LIBC && SHARED && USE_TLS && !USE___THREAD] (__libc_malloc_pthread_startup): New function. * malloc/Versions (libc: GLIBC_PRIVATE): New set, add that function. * malloc/hooks.c (memalign_starter): New function. * malloc/malloc.c: Declare it. * malloc/arena.c (save_memalign_hook): New variable. (ptmalloc_init): Set __memalign_hook to memalign_starter. * elf/dl-minimal.c (free): Clear the memory. (calloc): Just call malloc, knowing all memory it returns is cleared. * sysdeps/generic/dl-tls.c (allocate_dtv): Use calloc instead of malloc and memset; calloc can avoid the zeroing when redundant. (_dl_tls_setup): Likewise. * elf/dl-load.c (decompose_rpath): Likewise. * sysdeps/generic/libc-tls.c (__libc_setup_tls): Comment out memset call, since memory from sbrk at startup is already zero. * elf/rtld.c (_dl_start, dl_main): TLS_INIT_TP macro now returns an error string for failure, null for success. Update callers. * sysdeps/generic/libc-tls.c (__libc_setup_tls): Likewise. * elf/dl-load.c (_dl_map_object_from_fd): Likewise.
2002-12-06 11:15:07 +00:00
#endif
static void
ptmalloc_init (void)
{
2014-01-02 08:38:18 +00:00
if (__malloc_initialized >= 0)
return;
__malloc_initialized = 0;
#ifdef SHARED
Update. 2004-10-18 Jakub Jelinek <jakub@redhat.com> * elf/dl-libc.c (__libc_dlsym_private, __libc_register_dl_open_hook): New functions. (__libc_dlopen_mode): Call __libc_register_dl_open_hook and __libc_register_dlfcn_hook. * dlfcn/Makefile (routines, elide-routines.os): Set. Add rules to build and test tststatic2. * dlfcn/tststatic2.c: New test. * dlfcn/modstatic2.c: New test module. * dlfcn/dladdr.c: Call _dlfcn_hook from libdl.so if not NULL. Define __ prefixed routine in libc.a and in libdl.a just call it. * dlfcn/dladdr1.c: Likewise. * dlfcn/dlclose.c: Likewise. * dlfcn/dlerror.c: Likewise. * dlfcn/dlinfo.c: Likewise. * dlfcn/dlmopen.c: Likewise. * dlfcn/dlopen.c: Likewise. * dlfcn/dlopenold.c: Likewise. * dlfcn/dlsym.c: Likewise. * dlfcn/dlvsym.c: Likewise. * dlfcn/sdladdr.c: New file. * dlfcn/sdladdr1.c: New file. * dlfcn/sdlclose.c: New file. * dlfcn/sdlerror.c: New file. * dlfcn/sdlinfo.c: New file. * dlfcn/sdlopen.c: New file. * dlfcn/sdlsym.c: New file. * dlfcn/sdlvsym.c: New file. * dlfcn/Versions (libdl): Export _dlfcn_hook@GLIBC_PRIVATE. * include/dlfcn.h (DL_CALLER_DECL, DL_CALLER RETURN_ADDRESS): Define. (struct dlfcn_hook): New type. (_dlfcn_hook): New extern decl. (__dlopen, __dlclose, __dlsym, __dlerror, __dladdr, __dladdr1, __dlinfo, __dlmopen, __libc_dlsym_private, __libc_register_dl_open_hook, __libc_register_dlfcn_hook): New prototypes. (__dlvsym): Use DL_CALLER_DECL. * include/libc-symbols.h: Define libdl_hidden_proto and friends. * malloc/arena.c (_dl_open_hook): Extern decl. (ptmalloc_init): Don't call _dl_addr when dlopened from statically linked programs but don't use brk for them either.
2004-10-18 23:17:46 +00:00
/* In case this libc copy is in a non-default namespace, never use brk.
Likewise if dlopened from statically linked program. */
[BZ #77] Update. Add support for namespaces in the dynamic linker. * dlfcn/Makefile (libdl-routines): Add dlmopen. * dlfcn/Versions [libdl, GLIBC_2.3.4]: Add dlmopen. * dlfcn/dlfcn.h: Define Lmid_t, LM_ID_BASE, and LM_ID_NEWLM. Declare dlmopen. Document RTLD_DI_LMID. * dlfcn/dlinfo.c: Handle RTLD_DI_LMID. * dlfcn/dlmopen.c: New file. * dlfcn/dlopen.c: Pass new parameter to _dl_open. * dlfcn/dlopenold.c: Likewise. * elf/dl-addr.c: Adjust for removal of GL(dl_loaded). * elf/dl-caller.c: Likewise. * elf/dl-close.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-debug.c: Likewise. * elf/dl-lookup.c: Likewise. * elf/dl-sym.c: Likewise. * elf/dl-version.c: Likewise. * elf/do-lookup.h: Likewise. * elf/rtld.c: Likewise. * sysdeps/unix/sysv/linux/i386/dl-librecon.h: Likewise. * elf/dl-depsc: Likewise. Add new parameter to _dl_map_object. * elf/dl-fini.c: Call destructors in all namespaces. * elf/dl-iteratephdr.c: Compute total nloaded. Adjust for removal of GL(dl_loaded). * elf/dl-libc.c: Pass new parameter to _dl_open. Adjust for removal of GL(dl_loaded). * elf/dl-load.c (_dl_map_object_from_fd): Don't load ld.so a second time. Reuse the one from the main namespace in all others. Pass new parameter to _dl_new_object. Adjust for removal of GL(dl_loaded). * elf/dl-object.c: Take new parameter. Use it to initialize l_ns. Adjust for removal of GL(dl_loaded). * elf/dl-open.c (_dl_open): Take new parameter. Adjust for removal of GL(dl_loaded). * elf/dl-support.c: Replace global _dl_loaded etc variables with _dl_ns variable. * include/dlfcn.h: Adjust prototype of _dl_open. Define __LM_ID_CALLER. * include/link.h: Add l_real, l_ns, and l_direct_opencount elements. * sysdeps/generic/dl-tls.c: Bump TLS_STATIC_SURPLUS. Since libc is using TLS we need memory appropriate to the number of namespaces. * sysdeps/generic/ldsodefs.h (struct rtld_global): Replace _dl_loaded, _dl_nloaded, _dl_global_scope, _dl_main_searchlist, and _dl_global_scope_alloc with _dl_ns element. Define DL_NNS. Adjust prototypes of _dl_map_object and member in rtld_global_ro. * malloc/malloc.c: Include <dlfcn.h>. * malloc/arena.c (ptmalloc_init): If libc is not in primary namespace, never use brk. * elf/Makefile: Add rules to build and run tst-dlmopen1 and tst-dlmopen2. * elf/tst-dlmopen1.c: New file. * elf/tst-dlmopen1mod.c: New file. * elf/tst-dlmopen2.c: New file. * elf/dl-close.c: Improve reference counting by tracking direct loads. * elf/dl-lookup.c (add_dependency): Likewise. * elf/dl-open.c (dl_open_worker): Likewise. * elf/rtld.c (dl_main): Likewise. 2004-09-09 GOTO Masanori <gotom@debian.or.jp> [BZ #77] * elf/dl-close.c: Count down l_opencount to check not only for l_reldeps, but also l_initfini. 2004-10-13 Ulrich Drepper <drepper@redhat.com>
2004-10-14 02:08:23 +00:00
Dl_info di;
struct link_map *l;
Update. 2004-10-18 Jakub Jelinek <jakub@redhat.com> * elf/dl-libc.c (__libc_dlsym_private, __libc_register_dl_open_hook): New functions. (__libc_dlopen_mode): Call __libc_register_dl_open_hook and __libc_register_dlfcn_hook. * dlfcn/Makefile (routines, elide-routines.os): Set. Add rules to build and test tststatic2. * dlfcn/tststatic2.c: New test. * dlfcn/modstatic2.c: New test module. * dlfcn/dladdr.c: Call _dlfcn_hook from libdl.so if not NULL. Define __ prefixed routine in libc.a and in libdl.a just call it. * dlfcn/dladdr1.c: Likewise. * dlfcn/dlclose.c: Likewise. * dlfcn/dlerror.c: Likewise. * dlfcn/dlinfo.c: Likewise. * dlfcn/dlmopen.c: Likewise. * dlfcn/dlopen.c: Likewise. * dlfcn/dlopenold.c: Likewise. * dlfcn/dlsym.c: Likewise. * dlfcn/dlvsym.c: Likewise. * dlfcn/sdladdr.c: New file. * dlfcn/sdladdr1.c: New file. * dlfcn/sdlclose.c: New file. * dlfcn/sdlerror.c: New file. * dlfcn/sdlinfo.c: New file. * dlfcn/sdlopen.c: New file. * dlfcn/sdlsym.c: New file. * dlfcn/sdlvsym.c: New file. * dlfcn/Versions (libdl): Export _dlfcn_hook@GLIBC_PRIVATE. * include/dlfcn.h (DL_CALLER_DECL, DL_CALLER RETURN_ADDRESS): Define. (struct dlfcn_hook): New type. (_dlfcn_hook): New extern decl. (__dlopen, __dlclose, __dlsym, __dlerror, __dladdr, __dladdr1, __dlinfo, __dlmopen, __libc_dlsym_private, __libc_register_dl_open_hook, __libc_register_dlfcn_hook): New prototypes. (__dlvsym): Use DL_CALLER_DECL. * include/libc-symbols.h: Define libdl_hidden_proto and friends. * malloc/arena.c (_dl_open_hook): Extern decl. (ptmalloc_init): Don't call _dl_addr when dlopened from statically linked programs but don't use brk for them either.
2004-10-18 23:17:46 +00:00
if (_dl_open_hook != NULL
|| (_dl_addr (ptmalloc_init, &di, &l, NULL) != 0
2014-01-02 08:38:18 +00:00
&& l->l_ns != LM_ID_BASE))
[BZ #77] Update. Add support for namespaces in the dynamic linker. * dlfcn/Makefile (libdl-routines): Add dlmopen. * dlfcn/Versions [libdl, GLIBC_2.3.4]: Add dlmopen. * dlfcn/dlfcn.h: Define Lmid_t, LM_ID_BASE, and LM_ID_NEWLM. Declare dlmopen. Document RTLD_DI_LMID. * dlfcn/dlinfo.c: Handle RTLD_DI_LMID. * dlfcn/dlmopen.c: New file. * dlfcn/dlopen.c: Pass new parameter to _dl_open. * dlfcn/dlopenold.c: Likewise. * elf/dl-addr.c: Adjust for removal of GL(dl_loaded). * elf/dl-caller.c: Likewise. * elf/dl-close.c: Likewise. * elf/dl-conflict.c: Likewise. * elf/dl-debug.c: Likewise. * elf/dl-lookup.c: Likewise. * elf/dl-sym.c: Likewise. * elf/dl-version.c: Likewise. * elf/do-lookup.h: Likewise. * elf/rtld.c: Likewise. * sysdeps/unix/sysv/linux/i386/dl-librecon.h: Likewise. * elf/dl-depsc: Likewise. Add new parameter to _dl_map_object. * elf/dl-fini.c: Call destructors in all namespaces. * elf/dl-iteratephdr.c: Compute total nloaded. Adjust for removal of GL(dl_loaded). * elf/dl-libc.c: Pass new parameter to _dl_open. Adjust for removal of GL(dl_loaded). * elf/dl-load.c (_dl_map_object_from_fd): Don't load ld.so a second time. Reuse the one from the main namespace in all others. Pass new parameter to _dl_new_object. Adjust for removal of GL(dl_loaded). * elf/dl-object.c: Take new parameter. Use it to initialize l_ns. Adjust for removal of GL(dl_loaded). * elf/dl-open.c (_dl_open): Take new parameter. Adjust for removal of GL(dl_loaded). * elf/dl-support.c: Replace global _dl_loaded etc variables with _dl_ns variable. * include/dlfcn.h: Adjust prototype of _dl_open. Define __LM_ID_CALLER. * include/link.h: Add l_real, l_ns, and l_direct_opencount elements. * sysdeps/generic/dl-tls.c: Bump TLS_STATIC_SURPLUS. Since libc is using TLS we need memory appropriate to the number of namespaces. * sysdeps/generic/ldsodefs.h (struct rtld_global): Replace _dl_loaded, _dl_nloaded, _dl_global_scope, _dl_main_searchlist, and _dl_global_scope_alloc with _dl_ns element. Define DL_NNS. Adjust prototypes of _dl_map_object and member in rtld_global_ro. * malloc/malloc.c: Include <dlfcn.h>. * malloc/arena.c (ptmalloc_init): If libc is not in primary namespace, never use brk. * elf/Makefile: Add rules to build and run tst-dlmopen1 and tst-dlmopen2. * elf/tst-dlmopen1.c: New file. * elf/tst-dlmopen1mod.c: New file. * elf/tst-dlmopen2.c: New file. * elf/dl-close.c: Improve reference counting by tracking direct loads. * elf/dl-lookup.c (add_dependency): Likewise. * elf/dl-open.c (dl_open_worker): Likewise. * elf/rtld.c (dl_main): Likewise. 2004-09-09 GOTO Masanori <gotom@debian.or.jp> [BZ #77] * elf/dl-close.c: Count down l_opencount to check not only for l_reldeps, but also l_initfini. 2004-10-13 Ulrich Drepper <drepper@redhat.com>
2004-10-14 02:08:23 +00:00
__morecore = __failing_morecore;
#endif
2014-01-02 08:38:18 +00:00
tsd_key_create (&arena_key, NULL);
tsd_setspecific (arena_key, (void *) &main_arena);
thread_atfork (ptmalloc_lock_all, ptmalloc_unlock_all, ptmalloc_unlock_all2);
const char *s = NULL;
if (__glibc_likely (_environ != NULL))
{
char **runp = _environ;
char *envline;
while (__builtin_expect ((envline = next_env_entry (&runp)) != NULL,
2014-01-02 08:38:18 +00:00
0))
{
size_t len = strcspn (envline, "=");
if (envline[len] != '=')
/* This is a "MALLOC_" variable at the end of the string
without a '=' character. Ignore it since otherwise we
will access invalid memory below. */
continue;
switch (len)
{
case 6:
if (memcmp (envline, "CHECK_", 6) == 0)
s = &envline[7];
break;
case 8:
if (!__builtin_expect (__libc_enable_secure, 0))
{
if (memcmp (envline, "TOP_PAD_", 8) == 0)
__libc_mallopt (M_TOP_PAD, atoi (&envline[9]));
else if (memcmp (envline, "PERTURB_", 8) == 0)
__libc_mallopt (M_PERTURB, atoi (&envline[9]));
}
break;
case 9:
if (!__builtin_expect (__libc_enable_secure, 0))
{
if (memcmp (envline, "MMAP_MAX_", 9) == 0)
__libc_mallopt (M_MMAP_MAX, atoi (&envline[10]));
else if (memcmp (envline, "ARENA_MAX", 9) == 0)
__libc_mallopt (M_ARENA_MAX, atoi (&envline[10]));
}
break;
case 10:
if (!__builtin_expect (__libc_enable_secure, 0))
{
if (memcmp (envline, "ARENA_TEST", 10) == 0)
__libc_mallopt (M_ARENA_TEST, atoi (&envline[11]));
}
break;
case 15:
if (!__builtin_expect (__libc_enable_secure, 0))
{
if (memcmp (envline, "TRIM_THRESHOLD_", 15) == 0)
__libc_mallopt (M_TRIM_THRESHOLD, atoi (&envline[16]));
else if (memcmp (envline, "MMAP_THRESHOLD_", 15) == 0)
__libc_mallopt (M_MMAP_THRESHOLD, atoi (&envline[16]));
}
break;
default:
break;
}
}
}
if (s && s[0])
{
__libc_mallopt (M_CHECK_ACTION, (int) (s[0] - '0'));
if (check_action != 0)
__malloc_check_init ();
}
void (*hook) (void) = atomic_forced_read (__malloc_initialize_hook);
if (hook != NULL)
(*hook)();
__malloc_initialized = 1;
}
/* There are platforms (e.g. Hurd) with a link-time hook mechanism. */
#ifdef thread_atfork_static
2014-01-02 08:38:18 +00:00
thread_atfork_static (ptmalloc_lock_all, ptmalloc_unlock_all, \
ptmalloc_unlock_all2)
#endif
2014-01-02 08:38:18 +00:00
/* Managing heaps and arenas (for concurrent threads) */
#if MALLOC_DEBUG > 1
/* Print the complete contents of a single heap to stderr. */
static void
2014-01-02 08:38:18 +00:00
dump_heap (heap_info *heap)
{
char *ptr;
mchunkptr p;
2014-01-02 08:38:18 +00:00
fprintf (stderr, "Heap %p, size %10lx:\n", heap, (long) heap->size);
ptr = (heap->ar_ptr != (mstate) (heap + 1)) ?
(char *) (heap + 1) : (char *) (heap + 1) + sizeof (struct malloc_state);
p = (mchunkptr) (((unsigned long) ptr + MALLOC_ALIGN_MASK) &
~MALLOC_ALIGN_MASK);
for (;; )
{
fprintf (stderr, "chunk %p size %10lx", p, (long) p->size);
if (p == top (heap->ar_ptr))
{
fprintf (stderr, " (top)\n");
break;
}
else if (p->size == (0 | PREV_INUSE))
{
fprintf (stderr, " (fence)\n");
break;
}
fprintf (stderr, "\n");
p = next_chunk (p);
}
}
#endif /* MALLOC_DEBUG > 1 */
/* If consecutive mmap (0, HEAP_MAX_SIZE << 1, ...) calls return decreasing
addresses as opposed to increasing, new_heap would badly fragment the
address space. In that case remember the second HEAP_MAX_SIZE part
aligned to HEAP_MAX_SIZE from last mmap (0, HEAP_MAX_SIZE << 1, ...)
call (if it is already aligned) and try to reuse it next time. We need
no locking for it, as kernel ensures the atomicity for us - worst case
we'll call mmap (addr, HEAP_MAX_SIZE, ...) for some value of addr in
multiple threads, but only one will succeed. */
static char *aligned_heap_area;
/* Create a new heap. size is automatically rounded up to a multiple
of the page size. */
static heap_info *
internal_function
2014-01-02 08:38:18 +00:00
new_heap (size_t size, size_t top_pad)
{
size_t pagesize = GLRO (dl_pagesize);
char *p1, *p2;
unsigned long ul;
heap_info *h;
2014-01-02 08:38:18 +00:00
if (size + top_pad < HEAP_MIN_SIZE)
size = HEAP_MIN_SIZE;
2014-01-02 08:38:18 +00:00
else if (size + top_pad <= HEAP_MAX_SIZE)
size += top_pad;
2014-01-02 08:38:18 +00:00
else if (size > HEAP_MAX_SIZE)
return 0;
else
size = HEAP_MAX_SIZE;
size = ALIGN_UP (size, pagesize);
/* A memory region aligned to a multiple of HEAP_MAX_SIZE is needed.
No swap space needs to be reserved for the following large
mapping (on Linux, this is the case for all non-writable mappings
anyway). */
p2 = MAP_FAILED;
2014-01-02 08:38:18 +00:00
if (aligned_heap_area)
{
p2 = (char *) MMAP (aligned_heap_area, HEAP_MAX_SIZE, PROT_NONE,
MAP_NORESERVE);
aligned_heap_area = NULL;
if (p2 != MAP_FAILED && ((unsigned long) p2 & (HEAP_MAX_SIZE - 1)))
{
__munmap (p2, HEAP_MAX_SIZE);
p2 = MAP_FAILED;
}
}
2014-01-02 08:38:18 +00:00
if (p2 == MAP_FAILED)
{
p1 = (char *) MMAP (0, HEAP_MAX_SIZE << 1, PROT_NONE, MAP_NORESERVE);
if (p1 != MAP_FAILED)
{
p2 = (char *) (((unsigned long) p1 + (HEAP_MAX_SIZE - 1))
& ~(HEAP_MAX_SIZE - 1));
ul = p2 - p1;
if (ul)
__munmap (p1, ul);
else
aligned_heap_area = p2 + HEAP_MAX_SIZE;
__munmap (p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul);
}
else
2014-01-02 08:38:18 +00:00
{
/* Try to take the chance that an allocation of only HEAP_MAX_SIZE
is already aligned. */
p2 = (char *) MMAP (0, HEAP_MAX_SIZE, PROT_NONE, MAP_NORESERVE);
if (p2 == MAP_FAILED)
return 0;
if ((unsigned long) p2 & (HEAP_MAX_SIZE - 1))
{
__munmap (p2, HEAP_MAX_SIZE);
return 0;
}
}
}
2014-01-02 08:38:18 +00:00
if (__mprotect (p2, size, PROT_READ | PROT_WRITE) != 0)
{
__munmap (p2, HEAP_MAX_SIZE);
return 0;
}
h = (heap_info *) p2;
h->size = size;
h->mprotect_size = size;
LIBC_PROBE (memory_heap_new, 2, h, h->size);
return h;
}
/* Grow a heap. size is automatically rounded up to a
multiple of the page size. */
static int
2014-01-02 08:38:18 +00:00
grow_heap (heap_info *h, long diff)
{
size_t pagesize = GLRO (dl_pagesize);
long new_size;
diff = ALIGN_UP (diff, pagesize);
2014-01-02 08:38:18 +00:00
new_size = (long) h->size + diff;
if ((unsigned long) new_size > (unsigned long) HEAP_MAX_SIZE)
return -1;
2014-01-02 08:38:18 +00:00
if ((unsigned long) new_size > h->mprotect_size)
{
if (__mprotect ((char *) h + h->mprotect_size,
(unsigned long) new_size - h->mprotect_size,
PROT_READ | PROT_WRITE) != 0)
return -2;
h->mprotect_size = new_size;
}
h->size = new_size;
LIBC_PROBE (memory_heap_more, 2, h, h->size);
return 0;
}
/* Shrink a heap. */
static int
2014-01-02 08:38:18 +00:00
shrink_heap (heap_info *h, long diff)
{
long new_size;
2014-01-02 08:38:18 +00:00
new_size = (long) h->size - diff;
if (new_size < (long) sizeof (*h))
return -1;
2014-01-02 08:38:18 +00:00
/* Try to re-map the extra heap space freshly to save memory, and make it
inaccessible. See malloc-sysdep.h to know when this is true. */
if (__glibc_unlikely (check_may_shrink_heap ()))
{
2014-01-02 08:38:18 +00:00
if ((char *) MMAP ((char *) h + new_size, diff, PROT_NONE,
MAP_FIXED) == (char *) MAP_FAILED)
return -2;
h->mprotect_size = new_size;
}
else
2014-01-02 08:38:18 +00:00
__madvise ((char *) h + new_size, diff, MADV_DONTNEED);
/*fprintf(stderr, "shrink %p %08lx\n", h, new_size);*/
h->size = new_size;
LIBC_PROBE (memory_heap_less, 2, h, h->size);
return 0;
}
/* Delete a heap. */
#define delete_heap(heap) \
2014-01-02 08:38:18 +00:00
do { \
if ((char *) (heap) + HEAP_MAX_SIZE == aligned_heap_area) \
aligned_heap_area = NULL; \
__munmap ((char *) (heap), HEAP_MAX_SIZE); \
} while (0)
static int
internal_function
2014-01-02 08:38:18 +00:00
heap_trim (heap_info *heap, size_t pad)
{
mstate ar_ptr = heap->ar_ptr;
2014-01-02 08:38:18 +00:00
unsigned long pagesz = GLRO (dl_pagesize);
mchunkptr top_chunk = top (ar_ptr), p, bck, fwd;
heap_info *prev_heap;
malloc: Consistently apply trim_threshold to all heaps [BZ #17195] Trimming heaps is a balance between saving memory and the system overhead required to update page tables and discard allocated pages. The malloc option M_TRIM_THRESHOLD is a tunable that users are meant to use to decide where this balance point is but it is only applied to the main arena. For scalability reasons, glibc malloc has per-thread heaps but these are shrunk with madvise() if there is one page free at the top of the heap. In some circumstances this can lead to high system overhead if a thread has a control flow like while (data_to_process) { buf = malloc(large_size); do_stuff(); free(buf); } For a large size, the free() will call madvise (pagetable teardown, page free and TLB flush) every time followed immediately by a malloc (fault, kernel page alloc, zeroing and charge accounting). The kernel overhead can dominate such a workload. This patch allows the user to tune when madvise gets called by applying the trim threshold to the per-thread heaps and using similar logic to the main arena when deciding whether to shrink. Alternatively if the dynamic brk/mmap threshold gets adjusted then the new values will be obeyed by the per-thread heaps. Bug 17195 was a test case motivated by a problem encountered in scientific applications written in python that performance badly due to high page fault overhead. The basic operation of such a program was posted by Julian Taylor https://sourceware.org/ml/libc-alpha/2015-02/msg00373.html With this patch applied, the overhead is eliminated. All numbers in this report are in seconds and were recorded by running Julian's program 30 times. pyarray glibc madvise 2.21 v2 System min 1.81 ( 0.00%) 0.00 (100.00%) System mean 1.93 ( 0.00%) 0.02 ( 99.20%) System stddev 0.06 ( 0.00%) 0.01 ( 88.99%) System max 2.06 ( 0.00%) 0.03 ( 98.54%) Elapsed min 3.26 ( 0.00%) 2.37 ( 27.30%) Elapsed mean 3.39 ( 0.00%) 2.41 ( 28.84%) Elapsed stddev 0.14 ( 0.00%) 0.02 ( 82.73%) Elapsed max 4.05 ( 0.00%) 2.47 ( 39.01%) glibc madvise 2.21 v2 User 141.86 142.28 System 57.94 0.60 Elapsed 102.02 72.66 Note that almost a minutes worth of system time is eliminted and the program completes 28% faster on average. To illustrate the problem without python this is a basic test-case for the worst case scenario where every free is a madvise followed by a an alloc /* gcc bench-free.c -lpthread -o bench-free */ static int num = 1024; void __attribute__((noinline,noclone)) dostuff (void *p) { } void *worker (void *data) { int i; for (i = num; i--;) { void *m = malloc (48*4096); dostuff (m); free (m); } return NULL; } int main() { int i; pthread_t t; void *ret; if (pthread_create (&t, NULL, worker, NULL)) exit (2); if (pthread_join (t, &ret)) exit (3); return 0; } Before the patch, this resulted in 1024 calls to madvise. With the patch applied, madvise is called twice because the default trim threshold is high enough to avoid this. This a more complex case where there is a mix of frees. It's simply a different worker function for the test case above void *worker (void *data) { int i; int j = 0; void *free_index[num]; for (i = num; i--;) { void *m = malloc ((i % 58) *4096); dostuff (m); if (i % 2 == 0) { free (m); } else { free_index[j++] = m; } } for (; j >= 0; j--) { free(free_index[j]); } return NULL; } glibc 2.21 calls malloc 90305 times but with the patch applied, it's called 13438. Increasing the trim threshold will decrease the number of times it's called with the option of eliminating the overhead. ebizzy is meant to generate a workload resembling common web application server workloads. It is threaded with a large working set that at its core has an allocation, do_stuff, free loop that also hits this case. The primary metric of the benchmark is records processed per second. This is running on my desktop which is a single socket machine with an I7-4770 and 8 cores. Each thread count was run for 30 seconds. It was only run once as the performance difference is so high that the variation is insignificant. glibc 2.21 patch threads 1 10230 44114 threads 2 19153 84925 threads 4 34295 134569 threads 8 51007 183387 Note that the saving happens to be a concidence as the size allocated by ebizzy was less than the default threshold. If a different number of chunks were specified then it may also be necessary to tune the threshold to compensate This is roughly quadrupling the performance of this benchmark. The difference in system CPU usage illustrates why. ebizzy running 1 thread with glibc 2.21 10230 records/s 306904 real 30.00 s user 7.47 s sys 22.49 s 22.49 seconds was spent in the kernel for a workload runinng 30 seconds. With the patch applied ebizzy running 1 thread with patch applied 44126 records/s 1323792 real 30.00 s user 29.97 s sys 0.00 s system CPU usage was zero with the patch applied. strace shows that glibc running this workload calls madvise approximately 9000 times a second. With the patch applied madvise was called twice during the workload (or 0.06 times per second). 2015-02-10 Mel Gorman <mgorman@suse.de> [BZ #17195] * malloc/arena.c (free): Apply trim threshold to per-thread heaps as well as the main arena.
2015-04-02 06:44:14 +00:00
long new_size, top_size, top_area, extra, prev_size, misalign;
/* Can this heap go away completely? */
2014-01-02 08:38:18 +00:00
while (top_chunk == chunk_at_offset (heap, sizeof (*heap)))
{
prev_heap = heap->prev;
prev_size = prev_heap->size - (MINSIZE - 2 * SIZE_SZ);
p = chunk_at_offset (prev_heap, prev_size);
/* fencepost must be properly aligned. */
misalign = ((long) p) & MALLOC_ALIGN_MASK;
p = chunk_at_offset (prev_heap, prev_size - misalign);
assert (p->size == (0 | PREV_INUSE)); /* must be fencepost */
p = prev_chunk (p);
new_size = chunksize (p) + (MINSIZE - 2 * SIZE_SZ) + misalign;
assert (new_size > 0 && new_size < (long) (2 * MINSIZE));
if (!prev_inuse (p))
new_size += p->prev_size;
assert (new_size > 0 && new_size < HEAP_MAX_SIZE);
if (new_size + (HEAP_MAX_SIZE - prev_heap->size) < pad + MINSIZE + pagesz)
break;
ar_ptr->system_mem -= heap->size;
arena_mem -= heap->size;
LIBC_PROBE (memory_heap_free, 2, heap, heap->size);
delete_heap (heap);
heap = prev_heap;
if (!prev_inuse (p)) /* consolidate backward */
{
p = prev_chunk (p);
unlink (p, bck, fwd);
}
assert (((unsigned long) ((char *) p + new_size) & (pagesz - 1)) == 0);
assert (((char *) p + new_size) == ((char *) heap + heap->size));
top (ar_ptr) = top_chunk = p;
set_head (top_chunk, new_size | PREV_INUSE);
/*check_chunk(ar_ptr, top_chunk);*/
}
malloc: Consistently apply trim_threshold to all heaps [BZ #17195] Trimming heaps is a balance between saving memory and the system overhead required to update page tables and discard allocated pages. The malloc option M_TRIM_THRESHOLD is a tunable that users are meant to use to decide where this balance point is but it is only applied to the main arena. For scalability reasons, glibc malloc has per-thread heaps but these are shrunk with madvise() if there is one page free at the top of the heap. In some circumstances this can lead to high system overhead if a thread has a control flow like while (data_to_process) { buf = malloc(large_size); do_stuff(); free(buf); } For a large size, the free() will call madvise (pagetable teardown, page free and TLB flush) every time followed immediately by a malloc (fault, kernel page alloc, zeroing and charge accounting). The kernel overhead can dominate such a workload. This patch allows the user to tune when madvise gets called by applying the trim threshold to the per-thread heaps and using similar logic to the main arena when deciding whether to shrink. Alternatively if the dynamic brk/mmap threshold gets adjusted then the new values will be obeyed by the per-thread heaps. Bug 17195 was a test case motivated by a problem encountered in scientific applications written in python that performance badly due to high page fault overhead. The basic operation of such a program was posted by Julian Taylor https://sourceware.org/ml/libc-alpha/2015-02/msg00373.html With this patch applied, the overhead is eliminated. All numbers in this report are in seconds and were recorded by running Julian's program 30 times. pyarray glibc madvise 2.21 v2 System min 1.81 ( 0.00%) 0.00 (100.00%) System mean 1.93 ( 0.00%) 0.02 ( 99.20%) System stddev 0.06 ( 0.00%) 0.01 ( 88.99%) System max 2.06 ( 0.00%) 0.03 ( 98.54%) Elapsed min 3.26 ( 0.00%) 2.37 ( 27.30%) Elapsed mean 3.39 ( 0.00%) 2.41 ( 28.84%) Elapsed stddev 0.14 ( 0.00%) 0.02 ( 82.73%) Elapsed max 4.05 ( 0.00%) 2.47 ( 39.01%) glibc madvise 2.21 v2 User 141.86 142.28 System 57.94 0.60 Elapsed 102.02 72.66 Note that almost a minutes worth of system time is eliminted and the program completes 28% faster on average. To illustrate the problem without python this is a basic test-case for the worst case scenario where every free is a madvise followed by a an alloc /* gcc bench-free.c -lpthread -o bench-free */ static int num = 1024; void __attribute__((noinline,noclone)) dostuff (void *p) { } void *worker (void *data) { int i; for (i = num; i--;) { void *m = malloc (48*4096); dostuff (m); free (m); } return NULL; } int main() { int i; pthread_t t; void *ret; if (pthread_create (&t, NULL, worker, NULL)) exit (2); if (pthread_join (t, &ret)) exit (3); return 0; } Before the patch, this resulted in 1024 calls to madvise. With the patch applied, madvise is called twice because the default trim threshold is high enough to avoid this. This a more complex case where there is a mix of frees. It's simply a different worker function for the test case above void *worker (void *data) { int i; int j = 0; void *free_index[num]; for (i = num; i--;) { void *m = malloc ((i % 58) *4096); dostuff (m); if (i % 2 == 0) { free (m); } else { free_index[j++] = m; } } for (; j >= 0; j--) { free(free_index[j]); } return NULL; } glibc 2.21 calls malloc 90305 times but with the patch applied, it's called 13438. Increasing the trim threshold will decrease the number of times it's called with the option of eliminating the overhead. ebizzy is meant to generate a workload resembling common web application server workloads. It is threaded with a large working set that at its core has an allocation, do_stuff, free loop that also hits this case. The primary metric of the benchmark is records processed per second. This is running on my desktop which is a single socket machine with an I7-4770 and 8 cores. Each thread count was run for 30 seconds. It was only run once as the performance difference is so high that the variation is insignificant. glibc 2.21 patch threads 1 10230 44114 threads 2 19153 84925 threads 4 34295 134569 threads 8 51007 183387 Note that the saving happens to be a concidence as the size allocated by ebizzy was less than the default threshold. If a different number of chunks were specified then it may also be necessary to tune the threshold to compensate This is roughly quadrupling the performance of this benchmark. The difference in system CPU usage illustrates why. ebizzy running 1 thread with glibc 2.21 10230 records/s 306904 real 30.00 s user 7.47 s sys 22.49 s 22.49 seconds was spent in the kernel for a workload runinng 30 seconds. With the patch applied ebizzy running 1 thread with patch applied 44126 records/s 1323792 real 30.00 s user 29.97 s sys 0.00 s system CPU usage was zero with the patch applied. strace shows that glibc running this workload calls madvise approximately 9000 times a second. With the patch applied madvise was called twice during the workload (or 0.06 times per second). 2015-02-10 Mel Gorman <mgorman@suse.de> [BZ #17195] * malloc/arena.c (free): Apply trim threshold to per-thread heaps as well as the main arena.
2015-04-02 06:44:14 +00:00
/* Uses similar logic for per-thread arenas as the main arena with systrim
by preserving the top pad and at least a page. */
2014-01-02 08:38:18 +00:00
top_size = chunksize (top_chunk);
malloc: Consistently apply trim_threshold to all heaps [BZ #17195] Trimming heaps is a balance between saving memory and the system overhead required to update page tables and discard allocated pages. The malloc option M_TRIM_THRESHOLD is a tunable that users are meant to use to decide where this balance point is but it is only applied to the main arena. For scalability reasons, glibc malloc has per-thread heaps but these are shrunk with madvise() if there is one page free at the top of the heap. In some circumstances this can lead to high system overhead if a thread has a control flow like while (data_to_process) { buf = malloc(large_size); do_stuff(); free(buf); } For a large size, the free() will call madvise (pagetable teardown, page free and TLB flush) every time followed immediately by a malloc (fault, kernel page alloc, zeroing and charge accounting). The kernel overhead can dominate such a workload. This patch allows the user to tune when madvise gets called by applying the trim threshold to the per-thread heaps and using similar logic to the main arena when deciding whether to shrink. Alternatively if the dynamic brk/mmap threshold gets adjusted then the new values will be obeyed by the per-thread heaps. Bug 17195 was a test case motivated by a problem encountered in scientific applications written in python that performance badly due to high page fault overhead. The basic operation of such a program was posted by Julian Taylor https://sourceware.org/ml/libc-alpha/2015-02/msg00373.html With this patch applied, the overhead is eliminated. All numbers in this report are in seconds and were recorded by running Julian's program 30 times. pyarray glibc madvise 2.21 v2 System min 1.81 ( 0.00%) 0.00 (100.00%) System mean 1.93 ( 0.00%) 0.02 ( 99.20%) System stddev 0.06 ( 0.00%) 0.01 ( 88.99%) System max 2.06 ( 0.00%) 0.03 ( 98.54%) Elapsed min 3.26 ( 0.00%) 2.37 ( 27.30%) Elapsed mean 3.39 ( 0.00%) 2.41 ( 28.84%) Elapsed stddev 0.14 ( 0.00%) 0.02 ( 82.73%) Elapsed max 4.05 ( 0.00%) 2.47 ( 39.01%) glibc madvise 2.21 v2 User 141.86 142.28 System 57.94 0.60 Elapsed 102.02 72.66 Note that almost a minutes worth of system time is eliminted and the program completes 28% faster on average. To illustrate the problem without python this is a basic test-case for the worst case scenario where every free is a madvise followed by a an alloc /* gcc bench-free.c -lpthread -o bench-free */ static int num = 1024; void __attribute__((noinline,noclone)) dostuff (void *p) { } void *worker (void *data) { int i; for (i = num; i--;) { void *m = malloc (48*4096); dostuff (m); free (m); } return NULL; } int main() { int i; pthread_t t; void *ret; if (pthread_create (&t, NULL, worker, NULL)) exit (2); if (pthread_join (t, &ret)) exit (3); return 0; } Before the patch, this resulted in 1024 calls to madvise. With the patch applied, madvise is called twice because the default trim threshold is high enough to avoid this. This a more complex case where there is a mix of frees. It's simply a different worker function for the test case above void *worker (void *data) { int i; int j = 0; void *free_index[num]; for (i = num; i--;) { void *m = malloc ((i % 58) *4096); dostuff (m); if (i % 2 == 0) { free (m); } else { free_index[j++] = m; } } for (; j >= 0; j--) { free(free_index[j]); } return NULL; } glibc 2.21 calls malloc 90305 times but with the patch applied, it's called 13438. Increasing the trim threshold will decrease the number of times it's called with the option of eliminating the overhead. ebizzy is meant to generate a workload resembling common web application server workloads. It is threaded with a large working set that at its core has an allocation, do_stuff, free loop that also hits this case. The primary metric of the benchmark is records processed per second. This is running on my desktop which is a single socket machine with an I7-4770 and 8 cores. Each thread count was run for 30 seconds. It was only run once as the performance difference is so high that the variation is insignificant. glibc 2.21 patch threads 1 10230 44114 threads 2 19153 84925 threads 4 34295 134569 threads 8 51007 183387 Note that the saving happens to be a concidence as the size allocated by ebizzy was less than the default threshold. If a different number of chunks were specified then it may also be necessary to tune the threshold to compensate This is roughly quadrupling the performance of this benchmark. The difference in system CPU usage illustrates why. ebizzy running 1 thread with glibc 2.21 10230 records/s 306904 real 30.00 s user 7.47 s sys 22.49 s 22.49 seconds was spent in the kernel for a workload runinng 30 seconds. With the patch applied ebizzy running 1 thread with patch applied 44126 records/s 1323792 real 30.00 s user 29.97 s sys 0.00 s system CPU usage was zero with the patch applied. strace shows that glibc running this workload calls madvise approximately 9000 times a second. With the patch applied madvise was called twice during the workload (or 0.06 times per second). 2015-02-10 Mel Gorman <mgorman@suse.de> [BZ #17195] * malloc/arena.c (free): Apply trim threshold to per-thread heaps as well as the main arena.
2015-04-02 06:44:14 +00:00
top_area = top_size - MINSIZE - 1;
if (top_area <= pad)
return 0;
extra = ALIGN_DOWN(top_area - pad, pagesz);
if ((unsigned long) extra < mp_.trim_threshold)
return 0;
2014-01-02 08:38:18 +00:00
/* Try to shrink. */
2014-01-02 08:38:18 +00:00
if (shrink_heap (heap, extra) != 0)
return 0;
2014-01-02 08:38:18 +00:00
ar_ptr->system_mem -= extra;
arena_mem -= extra;
/* Success. Adjust top accordingly. */
2014-01-02 08:38:18 +00:00
set_head (top_chunk, (top_size - extra) | PREV_INUSE);
/*check_chunk(ar_ptr, top_chunk);*/
return 1;
}
/* Create a new arena with initial size "size". */
static mstate
2014-01-02 08:38:18 +00:00
_int_new_arena (size_t size)
{
mstate a;
heap_info *h;
char *ptr;
unsigned long misalign;
2014-01-02 08:38:18 +00:00
h = new_heap (size + (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT),
mp_.top_pad);
if (!h)
{
/* Maybe size is too large to fit in a single heap. So, just try
to create a minimally-sized arena and let _int_malloc() attempt
to deal with the large request via mmap_chunk(). */
h = new_heap (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT, mp_.top_pad);
if (!h)
return 0;
}
a = h->ar_ptr = (mstate) (h + 1);
malloc_init_state (a);
/*a->next = NULL;*/
a->system_mem = a->max_system_mem = h->size;
arena_mem += h->size;
/* Set up the top chunk, with proper alignment. */
2014-01-02 08:38:18 +00:00
ptr = (char *) (a + 1);
misalign = (unsigned long) chunk2mem (ptr) & MALLOC_ALIGN_MASK;
if (misalign > 0)
ptr += MALLOC_ALIGNMENT - misalign;
2014-01-02 08:38:18 +00:00
top (a) = (mchunkptr) ptr;
set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);
LIBC_PROBE (memory_arena_new, 2, a, size);
2014-01-02 08:38:18 +00:00
tsd_setspecific (arena_key, (void *) a);
mutex_init (&a->mutex);
(void) mutex_lock (&a->mutex);
2014-01-02 08:38:18 +00:00
(void) mutex_lock (&list_lock);
/* Add the new arena to the global list. */
a->next = main_arena.next;
atomic_write_barrier ();
main_arena.next = a;
2014-01-02 08:38:18 +00:00
(void) mutex_unlock (&list_lock);
return a;
}
static mstate
get_free_list (void)
{
mstate result = free_list;
if (result != NULL)
{
2014-01-02 08:38:18 +00:00
(void) mutex_lock (&list_lock);
result = free_list;
if (result != NULL)
2014-01-02 08:38:18 +00:00
free_list = result->next_free;
(void) mutex_unlock (&list_lock);
if (result != NULL)
2014-01-02 08:38:18 +00:00
{
LIBC_PROBE (memory_arena_reuse_free_list, 1, result);
(void) mutex_lock (&result->mutex);
tsd_setspecific (arena_key, (void *) result);
}
}
return result;
}
/* Lock and return an arena that can be reused for memory allocation.
Avoid AVOID_ARENA as we have already failed to allocate memory in
it and it is currently locked. */
static mstate
reused_arena (mstate avoid_arena)
{
mstate result;
static mstate next_to_use;
if (next_to_use == NULL)
next_to_use = &main_arena;
result = next_to_use;
do
{
2014-01-02 08:38:18 +00:00
if (!mutex_trylock (&result->mutex))
goto out;
result = result->next;
}
while (result != next_to_use);
/* Avoid AVOID_ARENA as we have already failed to allocate memory
in that arena and it is currently locked. */
if (result == avoid_arena)
result = result->next;
/* No arena available. Wait for the next in line. */
LIBC_PROBE (memory_arena_reuse_wait, 3, &result->mutex, result, avoid_arena);
2014-01-02 08:38:18 +00:00
(void) mutex_lock (&result->mutex);
2014-01-02 08:38:18 +00:00
out:
LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
2014-01-02 08:38:18 +00:00
tsd_setspecific (arena_key, (void *) result);
next_to_use = result->next;
return result;
}
static mstate
internal_function
2014-01-02 08:38:18 +00:00
arena_get2 (mstate a_tsd, size_t size, mstate avoid_arena)
{
mstate a;
2011-11-09 16:14:39 +00:00
static size_t narenas_limit;
a = get_free_list ();
if (a == NULL)
{
/* Nothing immediately available, so generate a new arena. */
if (narenas_limit == 0)
2014-01-02 08:38:18 +00:00
{
if (mp_.arena_max != 0)
narenas_limit = mp_.arena_max;
else if (narenas > mp_.arena_test)
{
int n = __get_nprocs ();
if (n >= 1)
narenas_limit = NARENAS_FROM_NCORES (n);
else
/* We have no information about the system. Assume two
cores. */
narenas_limit = NARENAS_FROM_NCORES (2);
}
}
2011-11-09 16:14:39 +00:00
repeat:;
size_t n = narenas;
2012-01-31 19:42:34 +00:00
/* NB: the following depends on the fact that (size_t)0 - 1 is a
2014-01-02 08:38:18 +00:00
very large number and that the underflow is OK. If arena_max
is set the value of arena_test is irrelevant. If arena_test
is set but narenas is not yet larger or equal to arena_test
narenas_limit is 0. There is no possibility for narenas to
be too big for the test to always fail since there is not
enough address space to create that many arenas. */
if (__glibc_unlikely (n <= narenas_limit - 1))
2014-01-02 08:38:18 +00:00
{
if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n))
goto repeat;
a = _int_new_arena (size);
if (__glibc_unlikely (a == NULL))
2014-01-02 08:38:18 +00:00
catomic_decrement (&narenas);
}
else
2014-01-02 08:38:18 +00:00
a = reused_arena (avoid_arena);
2011-11-09 16:14:39 +00:00
}
return a;
}
/* If we don't have the main arena, then maybe the failure is due to running
out of mmapped areas, so we can try allocating on the main arena.
Otherwise, it is likely that sbrk() has failed and there is still a chance
to mmap(), so try one of the other arenas. */
static mstate
arena_get_retry (mstate ar_ptr, size_t bytes)
{
LIBC_PROBE (memory_arena_retry, 2, bytes, ar_ptr);
2014-01-02 08:38:18 +00:00
if (ar_ptr != &main_arena)
{
(void) mutex_unlock (&ar_ptr->mutex);
ar_ptr = &main_arena;
(void) mutex_lock (&ar_ptr->mutex);
}
else
{
/* Grab ar_ptr->next prior to releasing its lock. */
mstate prev = ar_ptr->next ? ar_ptr : 0;
(void) mutex_unlock (&ar_ptr->mutex);
ar_ptr = arena_get2 (prev, bytes, ar_ptr);
}
return ar_ptr;
}
static void __attribute__ ((section ("__libc_thread_freeres_fn")))
arena_thread_freeres (void)
{
void *vptr = NULL;
2014-01-02 08:38:18 +00:00
mstate a = tsd_getspecific (arena_key, vptr);
tsd_setspecific (arena_key, NULL);
if (a != NULL)
{
2014-01-02 08:38:18 +00:00
(void) mutex_lock (&list_lock);
a->next_free = free_list;
free_list = a;
2014-01-02 08:38:18 +00:00
(void) mutex_unlock (&list_lock);
}
}
text_set_element (__libc_thread_subfreeres, arena_thread_freeres);
/*
* Local variables:
* c-basic-offset: 2
* End:
*/