AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-11-29 16:21:07 +00:00
Code Issues Packages Projects Releases Wiki Activity
fdbeb9c39e
glibc/malloc/arena.c
Mel Gorman c26efef979 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 12:14:14 +05:30

934 lines
27 KiB
C
Raw Blame 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
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
Lesser General Public License for more details.
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. */
#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
# 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. */
typedef struct _heap_info
{
mstate ar_ptr; /* Arena for this heap. */
struct _heap_info *prev; /* Previous heap. */
size_t size; /* Current size in bytes. */
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)
+ 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 { \
arena_lookup (ptr); \
arena_lock (ptr, size); \
} while (0)
#define arena_lookup(ptr) do { \
void *vptr = NULL; \
ptr = (mstate) tsd_getspecific (arena_key, vptr); \
} while (0)
#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) \
((heap_info *) ((unsigned long) (ptr) & ~(HEAP_MAX_SIZE - 1)))
#define arena_for_chunk(ptr) \
(chunk_non_main_arena (ptr) ? heap_for_ptr (ptr)->ar_ptr : &main_arena)
/**************************************************************************/
#ifndef NO_THREADS
/* atfork support. */
static void *(*save_malloc_hook)(size_t __size, const void *);
static void (*save_free_hook) (void *__ptr, const void *);
static void *save_arena;
# ifdef ATFORK_MEM
ATFORK_MEM;
# endif
/* Magic value for the thread-specific arena pointer when
malloc_atfork() is in use. */
# define ATFORK_ARENA_PTR ((void *) -1)
/* The following hooks are used while the `atfork' handling mechanism
is active. */
static void *
malloc_atfork (size_t sz, const void *caller)
{
void *vptr = NULL;
void *victim;
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
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;
p = mem2chunk (mem); /* do not bother to replicate free_check here */
if (chunk_is_mmapped (p)) /* release mmapped memory. */
{
munmap_chunk (p);
return;
}
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;
if (__malloc_initialized < 1)
return;
if (mutex_trylock (&list_lock))
{
void *my_arena;
tsd_getspecific (arena_key, my_arena);
if (my_arena == ATFORK_ARENA_PTR)
/* This is the same thread which already locks the global list.
Just bump the counter. */
goto out;
/* This thread has to wait its turn. */
(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. */
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;
if (__malloc_initialized < 1)
return;
if (--atfork_recursive_cntr != 0)
return;
tsd_setspecific (arena_key, save_arena);
__malloc_hook = save_malloc_hook;
__free_hook = save_free_hook;
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;
if (__malloc_initialized < 1)
return;
tsd_setspecific (arena_key, save_arena);
__malloc_hook = save_malloc_hook;
__free_hook = save_free_hook;
free_list = NULL;
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;
}
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)
&& (*current)[1] == 'A'
&& (*current)[2] == 'L'
&& (*current)[3] == 'L'
&& (*current)[4] == 'O'
&& (*current)[5] == 'C'
&& (*current)[6] == '_')
{
result = &(*current)[7];
/* Save current position for next visit. */
*position = ++current;
break;
}
++current;
}
return result;
}
#ifdef SHARED
static void *
__failing_morecore (ptrdiff_t d)
{
return (void *) MORECORE_FAILURE;
}
extern struct dl_open_hook *_dl_open_hook;
libc_hidden_proto (_dl_open_hook);
#endif
static void
ptmalloc_init (void)
{
if (__malloc_initialized >= 0)
return;
__malloc_initialized = 0;
#ifdef SHARED
/* In case this libc copy is in a non-default namespace, never use brk.
Likewise if dlopened from statically linked program. */
Dl_info di;
struct link_map *l;
if (_dl_open_hook != NULL
|| (_dl_addr (ptmalloc_init, &di, &l, NULL) != 0
&& l->l_ns != LM_ID_BASE))
__morecore = __failing_morecore;
#endif
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,
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
thread_atfork_static (ptmalloc_lock_all, ptmalloc_unlock_all, \
ptmalloc_unlock_all2)
#endif
/* Managing heaps and arenas (for concurrent threads) */
#if MALLOC_DEBUG > 1
/* Print the complete contents of a single heap to stderr. */
static void
dump_heap (heap_info *heap)
{
char *ptr;
mchunkptr p;
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
new_heap (size_t size, size_t top_pad)
{
size_t pagesize = GLRO (dl_pagesize);
char *p1, *p2;
unsigned long ul;
heap_info *h;
if (size + top_pad < HEAP_MIN_SIZE)
size = HEAP_MIN_SIZE;
else if (size + top_pad <= HEAP_MAX_SIZE)
size += top_pad;
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;
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;
}
}
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
{
/* 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;
}
}
}
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
grow_heap (heap_info *h, long diff)
{
size_t pagesize = GLRO (dl_pagesize);
long new_size;
diff = ALIGN_UP (diff, pagesize);
new_size = (long) h->size + diff;
if ((unsigned long) new_size > (unsigned long) HEAP_MAX_SIZE)
return -1;
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
shrink_heap (heap_info *h, long diff)
{
long new_size;
new_size = (long) h->size - diff;
if (new_size < (long) sizeof (*h))
return -1;
/* 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 ()))
{
if ((char *) MMAP ((char *) h + new_size, diff, PROT_NONE,
MAP_FIXED) == (char *) MAP_FAILED)
return -2;
h->mprotect_size = new_size;
}
else
__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) \
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
heap_trim (heap_info *heap, size_t pad)
{
mstate ar_ptr = heap->ar_ptr;
unsigned long pagesz = GLRO (dl_pagesize);
mchunkptr top_chunk = top (ar_ptr), p, bck, fwd;
heap_info *prev_heap;
long new_size, top_size, top_area, extra, prev_size, misalign;
/* Can this heap go away completely? */
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);*/
}
/* Uses similar logic for per-thread arenas as the main arena with systrim
by preserving the top pad and at least a page. */
top_size = chunksize (top_chunk);
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;
/* Try to shrink. */
if (shrink_heap (heap, extra) != 0)
return 0;
ar_ptr->system_mem -= extra;
arena_mem -= extra;
/* Success. Adjust top accordingly. */
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
_int_new_arena (size_t size)
{
mstate a;
heap_info *h;
char *ptr;
unsigned long misalign;
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. */
ptr = (char *) (a + 1);
misalign = (unsigned long) chunk2mem (ptr) & MALLOC_ALIGN_MASK;
if (misalign > 0)
ptr += MALLOC_ALIGNMENT - misalign;
top (a) = (mchunkptr) ptr;
set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);
LIBC_PROBE (memory_arena_new, 2, a, size);
tsd_setspecific (arena_key, (void *) a);
mutex_init (&a->mutex);
(void) mutex_lock (&a->mutex);
(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;
(void) mutex_unlock (&list_lock);
return a;
}
static mstate
get_free_list (void)
{
mstate result = free_list;
if (result != NULL)
{
(void) mutex_lock (&list_lock);
result = free_list;
if (result != NULL)
free_list = result->next_free;
(void) mutex_unlock (&list_lock);
if (result != NULL)
{
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
{
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);
(void) mutex_lock (&result->mutex);
out:
LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
tsd_setspecific (arena_key, (void *) result);
next_to_use = result->next;
return result;
}
static mstate
internal_function
arena_get2 (mstate a_tsd, size_t size, mstate avoid_arena)
{
mstate a;
static size_t narenas_limit;
a = get_free_list ();
if (a == NULL)
{
/* Nothing immediately available, so generate a new arena. */
if (narenas_limit == 0)
{
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);
}
}
repeat:;
size_t n = narenas;
/* NB: the following depends on the fact that (size_t)0 - 1 is a
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))
{
if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n))
goto repeat;
a = _int_new_arena (size);
if (__glibc_unlikely (a == NULL))
catomic_decrement (&narenas);
}
else
a = reused_arena (avoid_arena);
}
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);
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;
mstate a = tsd_getspecific (arena_key, vptr);
tsd_setspecific (arena_key, NULL);
if (a != NULL)
{
(void) mutex_lock (&list_lock);
a->next_free = free_list;
free_list = a;
(void) mutex_unlock (&list_lock);
}
}
text_set_element (__libc_thread_subfreeres, arena_thread_freeres);
/*
* Local variables:
* c-basic-offset: 2
* End:
*/
Reference in New Issue View Git Blame Copy Permalink