/* Malloc implementation for multiple threads without lock contention.
- Copyright (C) 2001-2012 Free Software Foundation, Inc.
+ Copyright (C) 2001-2021 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Wolfram Gloger <wg@malloc.de>, 2001.
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/>. */
+ not, see <https://www.gnu.org/licenses/>. */
#include <stdbool.h>
-/* Get the implementation for check_may_shrink_heap. */
-#include <malloc-sysdep.h>
+#if HAVE_TUNABLES
+# define TUNABLE_NAMESPACE malloc
+#endif
+#include <elf/dl-tunables.h>
/* Compile-time constants. */
-#define HEAP_MIN_SIZE (32*1024)
+#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 */
+# define HEAP_MAX_SIZE (1024 * 1024) /* must be a power of two */
# endif
#endif
mmap threshold, so that requests with a size just below that
threshold can be fulfilled without creating too many heaps. */
-
-#ifndef THREAD_STATS
-#define THREAD_STATS 0
-#endif
-
-/* If THREAD_STATS is non-zero, some statistics on mutex locking are
- computed. */
-
/***************************************************************************/
#define top(ar_ptr) ((ar_ptr)->top)
malloc_chunks. It is allocated with mmap() and always starts at an
address aligned to HEAP_MAX_SIZE. */
-typedef struct _heap_info {
+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. */
+ 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. */
/* 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];
+ + 2 * SIZE_SZ) % MALLOC_ALIGNMENT
+ ? -1 : 1];
-/* Thread specific data */
+/* Thread specific data. */
-static tsd_key_t arena_key;
-static mutex_t list_lock = MUTEX_INITIALIZER;
-#ifdef PER_THREAD
+static __thread mstate thread_arena attribute_tls_model_ie;
+
+/* Arena free list. free_list_lock synchronizes access to the
+ free_list variable below, and the next_free and attached_threads
+ members of struct malloc_state objects. No other locks must be
+ acquired after free_list_lock has been acquired. */
+
+__libc_lock_define_initialized (static, free_list_lock);
static size_t narenas = 1;
static mstate free_list;
-#endif
-#if THREAD_STATS
-static int stat_n_heaps;
-#define THREAD_STAT(x) x
-#else
-#define THREAD_STAT(x) do ; while(0)
-#endif
+/* list_lock prevents concurrent writes to the next member of struct
+ malloc_state objects.
+
+ Read access to the next member is supposed to synchronize with the
+ atomic_write_barrier and the write to the next member in
+ _int_new_arena. This suffers from data races; see the FIXME
+ comments in _int_new_arena and reused_arena.
-/* Mapped memory in non-main arenas (reliable only for NO_THREADS). */
-static unsigned long arena_mem;
+ list_lock also prevents concurrent forks. At the time list_lock is
+ acquired, no arena lock must have been acquired, but it is
+ permitted to acquire arena locks subsequently, while list_lock is
+ acquired. */
+__libc_lock_define_initialized (static, list_lock);
/* Already initialized? */
int __malloc_initialized = -1;
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)
-
-#ifdef PER_THREAD
-# define arena_lock(ptr, size) do { \
- if(ptr) \
- (void)mutex_lock(&ptr->mutex); \
- else \
- ptr = arena_get2(ptr, (size), NULL); \
-} while(0)
-#else
-# define arena_lock(ptr, size) do { \
- if(ptr && !mutex_trylock(&ptr->mutex)) { \
- THREAD_STAT(++(ptr->stat_lock_direct)); \
- } else \
- ptr = arena_get2(ptr, (size), NULL); \
-} while(0)
-#endif
+ ptr = thread_arena; \
+ arena_lock (ptr, size); \
+ } while (0)
+
+#define arena_lock(ptr, size) do { \
+ if (ptr) \
+ __libc_lock_lock (ptr->mutex); \
+ else \
+ ptr = arena_get2 ((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)))
+ ((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)
+ (chunk_main_arena (ptr) ? &main_arena : heap_for_ptr (ptr)->ar_ptr)
/**************************************************************************/
-#ifndef NO_THREADS
-
/* atfork support. */
-static __malloc_ptr_t (*save_malloc_hook) (size_t __size,
- const __malloc_ptr_t);
-static void (*save_free_hook) (__malloc_ptr_t __ptr,
- const __malloc_ptr_t);
-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. */
+/* The following three functions are called around fork from a
+ multi-threaded process. We do not use the general fork handler
+ mechanism to make sure that our handlers are the last ones being
+ called, so that other fork handlers can use the malloc
+ subsystem. */
-static void*
-malloc_atfork(size_t sz, const void *caller)
+void
+__malloc_fork_lock_parent (void)
{
- 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 */
+ if (__malloc_initialized < 1)
return;
- p = mem2chunk(mem); /* do not bother to replicate free_check here */
+ /* We do not acquire free_list_lock here because we completely
+ reconstruct free_list in __malloc_fork_unlock_child. */
- if (chunk_is_mmapped(p)) /* release mmapped memory. */
- {
- munmap_chunk(p);
- return;
- }
+ __libc_lock_lock (list_lock);
- ar_ptr = arena_for_chunk(p);
- tsd_getspecific(arena_key, vptr);
- _int_free(ar_ptr, p, vptr == ATFORK_ARENA_PTR);
+ for (mstate ar_ptr = &main_arena;; )
+ {
+ __libc_lock_lock (ar_ptr->mutex);
+ ar_ptr = ar_ptr->next;
+ if (ar_ptr == &main_arena)
+ break;
+ }
}
-
-/* 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)
+void
+__malloc_fork_unlock_parent (void)
{
- mstate ar_ptr;
-
- if(__malloc_initialized < 1)
+ if (__malloc_initialized < 1)
return;
- if (mutex_trylock(&list_lock))
+
+ for (mstate ar_ptr = &main_arena;; )
{
- 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);
+ __libc_lock_unlock (ar_ptr->mutex);
+ ar_ptr = ar_ptr->next;
+ if (ar_ptr == &main_arena)
+ break;
}
- 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;
+ __libc_lock_unlock (list_lock);
}
-static void
-ptmalloc_unlock_all (void)
+void
+__malloc_fork_unlock_child (void)
{
- mstate ar_ptr;
-
- if(__malloc_initialized < 1)
+ 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__
+ /* Push all arenas to the free list, except thread_arena, which is
+ attached to the current thread. */
+ __libc_lock_init (free_list_lock);
+ if (thread_arena != NULL)
+ thread_arena->attached_threads = 1;
+ free_list = NULL;
+ for (mstate ar_ptr = &main_arena;; )
+ {
+ __libc_lock_init (ar_ptr->mutex);
+ if (ar_ptr != thread_arena)
+ {
+ /* This arena is no longer attached to any thread. */
+ ar_ptr->attached_threads = 0;
+ ar_ptr->next_free = free_list;
+ free_list = ar_ptr;
+ }
+ ar_ptr = ar_ptr->next;
+ if (ar_ptr == &main_arena)
+ break;
+ }
-/* 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. */
+ __libc_lock_init (list_lock);
+}
+#if HAVE_TUNABLES
static void
-ptmalloc_unlock_all2 (void)
+TUNABLE_CALLBACK (set_mallopt_check) (tunable_val_t *valp)
{
- mstate ar_ptr;
-
- if(__malloc_initialized < 1)
- return;
- tsd_setspecific(arena_key, save_arena);
- __malloc_hook = save_malloc_hook;
- __free_hook = save_free_hook;
-#ifdef PER_THREAD
- free_list = NULL;
-#endif
- for(ar_ptr = &main_arena;;) {
- mutex_init(&ar_ptr->mutex);
-#ifdef PER_THREAD
- if (ar_ptr != save_arena) {
- ar_ptr->next_free = free_list;
- free_list = ar_ptr;
- }
-#endif
- ar_ptr = ar_ptr->next;
- if(ar_ptr == &main_arena) break;
- }
- mutex_init(&list_lock);
- atfork_recursive_cntr = 0;
+ int32_t value = (int32_t) valp->numval;
+ if (value != 0)
+ __malloc_check_init ();
}
-# else
-
-# define ptmalloc_unlock_all2 ptmalloc_unlock_all
-
-# endif
-
-#endif /* !NO_THREADS */
+# define TUNABLE_CALLBACK_FNDECL(__name, __type) \
+static inline int do_ ## __name (__type value); \
+static void \
+TUNABLE_CALLBACK (__name) (tunable_val_t *valp) \
+{ \
+ __type value = (__type) (valp)->numval; \
+ do_ ## __name (value); \
+}
+TUNABLE_CALLBACK_FNDECL (set_mmap_threshold, size_t)
+TUNABLE_CALLBACK_FNDECL (set_mmaps_max, int32_t)
+TUNABLE_CALLBACK_FNDECL (set_top_pad, size_t)
+TUNABLE_CALLBACK_FNDECL (set_perturb_byte, int32_t)
+TUNABLE_CALLBACK_FNDECL (set_trim_threshold, size_t)
+TUNABLE_CALLBACK_FNDECL (set_arena_max, size_t)
+TUNABLE_CALLBACK_FNDECL (set_arena_test, size_t)
+#if USE_TCACHE
+TUNABLE_CALLBACK_FNDECL (set_tcache_max, size_t)
+TUNABLE_CALLBACK_FNDECL (set_tcache_count, size_t)
+TUNABLE_CALLBACK_FNDECL (set_tcache_unsorted_limit, size_t)
+#endif
+TUNABLE_CALLBACK_FNDECL (set_mxfast, size_t)
+#else
/* Initialization routine. */
#include <string.h>
extern char **_environ;
static char *
-internal_function
next_env_entry (char ***position)
{
char **current = *position;
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];
+ && (*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;
+ /* Save current position for next visit. */
+ *position = ++current;
- break;
- }
+ break;
+ }
++current;
}
return result;
}
+#endif
#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
+#if USE_TCACHE
+static void tcache_key_initialize (void);
+#endif
+
static void
ptmalloc_init (void)
{
- if(__malloc_initialized >= 0) return;
+ if (__malloc_initialized >= 0)
+ return;
+
__malloc_initialized = 0;
+#if USE_TCACHE
+ tcache_key_initialize ();
+#endif
+
+#ifdef USE_MTAG
+ if ((TUNABLE_GET_FULL (glibc, mem, tagging, int32_t, NULL) & 1) != 0)
+ {
+ /* If the tunable says that we should be using tagged memory
+ and that morecore does not support tagged regions, then
+ disable it. */
+ if (__MTAG_SBRK_UNTAGGED)
+ __always_fail_morecore = true;
+
+ mtag_enabled = true;
+ mtag_mmap_flags = __MTAG_MMAP_FLAGS;
+ }
+#endif
+
#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;
+ /* In case this libc copy is in a non-default namespace, never use
+ brk. Likewise if dlopened from statically linked program. The
+ generic sbrk implementation also enforces this, but it is not
+ used on Hurd. */
+ if (!__libc_initial)
+ __always_fail_morecore = true;
#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);
+ thread_arena = &main_arena;
+
+ malloc_init_state (&main_arena);
+
+#if HAVE_TUNABLES
+ TUNABLE_GET (check, int32_t, TUNABLE_CALLBACK (set_mallopt_check));
+ TUNABLE_GET (top_pad, size_t, TUNABLE_CALLBACK (set_top_pad));
+ TUNABLE_GET (perturb, int32_t, TUNABLE_CALLBACK (set_perturb_byte));
+ TUNABLE_GET (mmap_threshold, size_t, TUNABLE_CALLBACK (set_mmap_threshold));
+ TUNABLE_GET (trim_threshold, size_t, TUNABLE_CALLBACK (set_trim_threshold));
+ TUNABLE_GET (mmap_max, int32_t, TUNABLE_CALLBACK (set_mmaps_max));
+ TUNABLE_GET (arena_max, size_t, TUNABLE_CALLBACK (set_arena_max));
+ TUNABLE_GET (arena_test, size_t, TUNABLE_CALLBACK (set_arena_test));
+# if USE_TCACHE
+ TUNABLE_GET (tcache_max, size_t, TUNABLE_CALLBACK (set_tcache_max));
+ TUNABLE_GET (tcache_count, size_t, TUNABLE_CALLBACK (set_tcache_count));
+ TUNABLE_GET (tcache_unsorted_limit, size_t,
+ TUNABLE_CALLBACK (set_tcache_unsorted_limit));
+# endif
+ TUNABLE_GET (mxfast, size_t, TUNABLE_CALLBACK (set_mxfast));
+#else
const char *s = NULL;
- if (__builtin_expect (_environ != NULL, 1))
+ 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]));
-#ifdef PER_THREAD
- else if (memcmp (envline, "ARENA_MAX", 9) == 0)
- __libc_mallopt(M_ARENA_MAX, atoi(&envline[10]));
-#endif
- }
- break;
-#ifdef PER_THREAD
- 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;
-#endif
- 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;
- }
- }
+ 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) = force_reg (__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)
+ if (s && s[0] != '\0' && s[0] != '0')
+ __malloc_check_init ();
#endif
-
-\f
+}
/* Managing heaps and arenas (for concurrent threads) */
/* Print the complete contents of a single heap to stderr. */
static void
-dump_heap(heap_info *heap)
+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, "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) chunksize_nomask(p));
+ if (p == top (heap->ar_ptr))
+ {
+ fprintf (stderr, " (top)\n");
+ break;
+ }
+ else if (chunksize_nomask(p) == (0 | PREV_INUSE))
+ {
+ fprintf (stderr, " (fence)\n");
+ break;
+ }
+ fprintf (stderr, "\n");
+ p = next_chunk (p);
}
- fprintf(stderr, "\n");
- p = next_chunk(p);
- }
}
-
#endif /* MALLOC_DEBUG > 1 */
/* If consecutive mmap (0, HEAP_MAX_SIZE << 1, ...) calls return decreasing
of the page size. */
static heap_info *
-internal_function
-new_heap(size_t size, size_t top_pad)
+new_heap (size_t size, size_t top_pad)
{
- size_t page_mask = GLRO(dl_pagesize) - 1;
+ size_t pagesize = GLRO (dl_pagesize);
char *p1, *p2;
unsigned long ul;
heap_info *h;
- if(size+top_pad < HEAP_MIN_SIZE)
+ if (size + top_pad < HEAP_MIN_SIZE)
size = HEAP_MIN_SIZE;
- else if(size+top_pad <= HEAP_MAX_SIZE)
+ else if (size + top_pad <= HEAP_MAX_SIZE)
size += top_pad;
- else if(size > HEAP_MAX_SIZE)
+ else if (size > HEAP_MAX_SIZE)
return 0;
else
size = HEAP_MAX_SIZE;
- size = (size + page_mask) & ~page_mask;
+ 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 (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);
+ 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
- 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;
- }
+ {
+ /* 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;
+ if (__mprotect (p2, size, mtag_mmap_flags | PROT_READ | PROT_WRITE) != 0)
+ {
+ __munmap (p2, HEAP_MAX_SIZE);
+ return 0;
+ }
+ h = (heap_info *) p2;
h->size = size;
h->mprotect_size = size;
- THREAD_STAT(stat_n_heaps++);
+ LIBC_PROBE (memory_heap_new, 2, h, h->size);
return h;
}
multiple of the page size. */
static int
-grow_heap(heap_info *h, long diff)
+grow_heap (heap_info *h, long diff)
{
- size_t page_mask = GLRO(dl_pagesize) - 1;
+ size_t pagesize = GLRO (dl_pagesize);
long new_size;
- diff = (diff + page_mask) & ~page_mask;
- new_size = (long)h->size + diff;
- if((unsigned long) new_size > (unsigned long) HEAP_MAX_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;
- }
+
+ if ((unsigned long) new_size > h->mprotect_size)
+ {
+ if (__mprotect ((char *) h + h->mprotect_size,
+ (unsigned long) new_size - h->mprotect_size,
+ mtag_mmap_flags | 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)
+shrink_heap (heap_info *h, long diff)
{
long new_size;
- new_size = (long)h->size - diff;
- if(new_size < (long)sizeof(*h))
+ 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 (__builtin_expect (check_may_shrink_heap (), 0))
+ if (__glibc_unlikely (check_may_shrink_heap ()))
{
- if((char *)MMAP((char *)h + new_size, diff, PROT_NONE,
- MAP_FIXED) == (char *) MAP_FAILED)
- return -2;
+ 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);
+ __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)
+ 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)
+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;
+ unsigned long pagesz = GLRO (dl_pagesize);
+ mchunkptr top_chunk = top (ar_ptr), p;
heap_info *prev_heap;
- long new_size, top_size, extra, prev_size, misalign;
+ 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;
- delete_heap(heap);
- heap = prev_heap;
- if(!prev_inuse(p)) { /* consolidate backward */
- p = prev_chunk(p);
- unlink(p, bck, fwd);
+ 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 (chunksize_nomask (p) == (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 += prev_size (p);
+ 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;
+ 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_chunk (ar_ptr, p);
+ }
+ 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);*/
}
- 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);*/
- }
- top_size = chunksize(top_chunk);
- extra = (top_size - pad - MINSIZE - 1) & ~(pagesz - 1);
- if(extra < (long)pagesz)
+
+ /* Uses similar logic for per-thread arenas as the main arena with systrim
+ and _int_free by preserving the top pad and rounding down to the nearest
+ page. */
+ top_size = chunksize (top_chunk);
+ if ((unsigned long)(top_size) <
+ (unsigned long)(mp_.trim_threshold))
+ return 0;
+
+ top_area = top_size - MINSIZE - 1;
+ if (top_area < 0 || (size_t) top_area <= pad)
return 0;
+
+ /* Release in pagesize units and round down to the nearest page. */
+ extra = ALIGN_DOWN(top_area - pad, pagesz);
+ if (extra == 0)
+ return 0;
+
/* Try to shrink. */
- if(shrink_heap(heap, extra) != 0)
+ 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);
+ 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". */
+/* If REPLACED_ARENA is not NULL, detach it from this thread. Must be
+ called while free_list_lock is held. */
+static void
+detach_arena (mstate replaced_arena)
+{
+ if (replaced_arena != NULL)
+ {
+ assert (replaced_arena->attached_threads > 0);
+ /* The current implementation only detaches from main_arena in
+ case of allocation failure. This means that it is likely not
+ beneficial to put the arena on free_list even if the
+ reference count reaches zero. */
+ --replaced_arena->attached_threads;
+ }
+}
+
static mstate
-_int_new_arena(size_t size)
+_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);
+ 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->attached_threads = 1;
/*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;
+ 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);
+ top (a) = (mchunkptr) ptr;
+ set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);
- tsd_setspecific(arena_key, (void *)a);
- mutex_init(&a->mutex);
- (void)mutex_lock(&a->mutex);
+ LIBC_PROBE (memory_arena_new, 2, a, size);
+ mstate replaced_arena = thread_arena;
+ thread_arena = a;
+ __libc_lock_init (a->mutex);
-#ifdef PER_THREAD
- (void)mutex_lock(&list_lock);
-#endif
+ __libc_lock_lock (list_lock);
/* Add the new arena to the global list. */
a->next = main_arena.next;
+ /* FIXME: The barrier is an attempt to synchronize with read access
+ in reused_arena, which does not acquire list_lock while
+ traversing the list. */
atomic_write_barrier ();
main_arena.next = a;
-#ifdef PER_THREAD
- (void)mutex_unlock(&list_lock);
-#endif
+ __libc_lock_unlock (list_lock);
- THREAD_STAT(++(a->stat_lock_loop));
+ __libc_lock_lock (free_list_lock);
+ detach_arena (replaced_arena);
+ __libc_lock_unlock (free_list_lock);
+
+ /* Lock this arena. NB: Another thread may have been attached to
+ this arena because the arena is now accessible from the
+ main_arena.next list and could have been picked by reused_arena.
+ This can only happen for the last arena created (before the arena
+ limit is reached). At this point, some arena has to be attached
+ to two threads. We could acquire the arena lock before list_lock
+ to make it less likely that reused_arena picks this new arena,
+ but this could result in a deadlock with
+ __malloc_fork_lock_parent. */
+
+ __libc_lock_lock (a->mutex);
return a;
}
-#ifdef PER_THREAD
+/* Remove an arena from free_list. */
static mstate
get_free_list (void)
{
+ mstate replaced_arena = thread_arena;
mstate result = free_list;
if (result != NULL)
{
- (void)mutex_lock(&list_lock);
+ __libc_lock_lock (free_list_lock);
result = free_list;
- if (result != NULL)
- free_list = result->next_free;
- (void)mutex_unlock(&list_lock);
-
if (result != NULL)
{
- (void)mutex_lock(&result->mutex);
- tsd_setspecific(arena_key, (void *)result);
- THREAD_STAT(++(result->stat_lock_loop));
+ free_list = result->next_free;
+
+ /* The arena will be attached to this thread. */
+ assert (result->attached_threads == 0);
+ result->attached_threads = 1;
+
+ detach_arena (replaced_arena);
}
+ __libc_lock_unlock (free_list_lock);
+
+ if (result != NULL)
+ {
+ LIBC_PROBE (memory_arena_reuse_free_list, 1, result);
+ __libc_lock_lock (result->mutex);
+ thread_arena = result;
+ }
}
return result;
}
+/* Remove the arena from the free list (if it is present).
+ free_list_lock must have been acquired by the caller. */
+static void
+remove_from_free_list (mstate arena)
+{
+ mstate *previous = &free_list;
+ for (mstate p = free_list; p != NULL; p = p->next_free)
+ {
+ assert (p->attached_threads == 0);
+ if (p == arena)
+ {
+ /* Remove the requested arena from the list. */
+ *previous = p->next_free;
+ break;
+ }
+ else
+ previous = &p->next_free;
+ }
+}
+
/* 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. */
reused_arena (mstate avoid_arena)
{
mstate result;
+ /* FIXME: Access to next_to_use suffers from data races. */
static mstate next_to_use;
if (next_to_use == NULL)
next_to_use = &main_arena;
+ /* Iterate over all arenas (including those linked from
+ free_list). */
result = next_to_use;
do
{
- if (!mutex_trylock(&result->mutex))
- goto out;
+ if (!__libc_lock_trylock (result->mutex))
+ goto out;
+ /* FIXME: This is a data race, see _int_new_arena. */
result = result->next;
}
while (result != next_to_use);
if (result == avoid_arena)
result = result->next;
- /* No arena available. Wait for the next in line. */
- (void)mutex_lock(&result->mutex);
+ /* No arena available without contention. Wait for the next in line. */
+ LIBC_PROBE (memory_arena_reuse_wait, 3, &result->mutex, result, avoid_arena);
+ __libc_lock_lock (result->mutex);
- out:
- tsd_setspecific(arena_key, (void *)result);
- THREAD_STAT(++(result->stat_lock_loop));
+out:
+ /* Attach the arena to the current thread. */
+ {
+ /* Update the arena thread attachment counters. */
+ mstate replaced_arena = thread_arena;
+ __libc_lock_lock (free_list_lock);
+ detach_arena (replaced_arena);
+
+ /* We may have picked up an arena on the free list. We need to
+ preserve the invariant that no arena on the free list has a
+ positive attached_threads counter (otherwise,
+ arena_thread_freeres cannot use the counter to determine if the
+ arena needs to be put on the free list). We unconditionally
+ remove the selected arena from the free list. The caller of
+ reused_arena checked the free list and observed it to be empty,
+ so the list is very short. */
+ remove_from_free_list (result);
+
+ ++result->attached_threads;
+
+ __libc_lock_unlock (free_list_lock);
+ }
+
+ LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
+ thread_arena = result;
next_to_use = result->next;
return result;
}
-#endif
static mstate
-internal_function
-arena_get2(mstate a_tsd, size_t size, mstate avoid_arena)
+arena_get2 (size_t size, mstate avoid_arena)
{
mstate a;
-#ifdef PER_THREAD
static size_t narenas_limit;
a = get_free_list ();
{
/* 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);
- }
- }
+ {
+ 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 (__builtin_expect (n <= narenas_limit - 1, 0))
- {
- if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n))
- goto repeat;
- a = _int_new_arena (size);
- if (__builtin_expect (a == NULL, 0))
- catomic_decrement (&narenas);
- }
+ 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);
+ a = reused_arena (avoid_arena);
}
-#else
- if(!a_tsd)
- a = a_tsd = &main_arena;
- else {
- a = a_tsd->next;
- if(!a) {
- /* This can only happen while initializing the new arena. */
- (void)mutex_lock(&main_arena.mutex);
- THREAD_STAT(++(main_arena.stat_lock_wait));
- return &main_arena;
- }
- }
-
- /* Check the global, circularly linked list for available arenas. */
- bool retried = false;
- repeat:
- do {
- if(!mutex_trylock(&a->mutex)) {
- if (retried)
- (void)mutex_unlock(&list_lock);
- THREAD_STAT(++(a->stat_lock_loop));
- tsd_setspecific(arena_key, (void *)a);
- return a;
- }
- a = a->next;
- } while(a != a_tsd);
-
- /* If not even the list_lock can be obtained, try again. This can
- happen during `atfork', or for example on systems where thread
- creation makes it temporarily impossible to obtain _any_
- locks. */
- if(!retried && mutex_trylock(&list_lock)) {
- /* We will block to not run in a busy loop. */
- (void)mutex_lock(&list_lock);
-
- /* Since we blocked there might be an arena available now. */
- retried = true;
- a = a_tsd;
- goto repeat;
- }
-
- /* Nothing immediately available, so generate a new arena. */
- a = _int_new_arena(size);
- (void)mutex_unlock(&list_lock);
-#endif
-
return a;
}
static mstate
arena_get_retry (mstate ar_ptr, size_t bytes)
{
- 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);
- }
+ LIBC_PROBE (memory_arena_retry, 2, bytes, ar_ptr);
+ if (ar_ptr != &main_arena)
+ {
+ __libc_lock_unlock (ar_ptr->mutex);
+ ar_ptr = &main_arena;
+ __libc_lock_lock (ar_ptr->mutex);
+ }
+ else
+ {
+ __libc_lock_unlock (ar_ptr->mutex);
+ ar_ptr = arena_get2 (bytes, ar_ptr);
+ }
return ar_ptr;
}
-#ifdef PER_THREAD
-static void __attribute__ ((section ("__libc_thread_freeres_fn")))
-arena_thread_freeres (void)
+void
+__malloc_arena_thread_freeres (void)
{
- void *vptr = NULL;
- mstate a = tsd_getspecific(arena_key, vptr);
- tsd_setspecific(arena_key, NULL);
+ /* Shut down the thread cache first. This could deallocate data for
+ the thread arena, so do this before we put the arena on the free
+ list. */
+ tcache_thread_shutdown ();
+
+ mstate a = thread_arena;
+ thread_arena = NULL;
if (a != NULL)
{
- (void)mutex_lock(&list_lock);
- a->next_free = free_list;
- free_list = a;
- (void)mutex_unlock(&list_lock);
+ __libc_lock_lock (free_list_lock);
+ /* If this was the last attached thread for this arena, put the
+ arena on the free list. */
+ assert (a->attached_threads > 0);
+ if (--a->attached_threads == 0)
+ {
+ a->next_free = free_list;
+ free_list = a;
+ }
+ __libc_lock_unlock (free_list_lock);
}
}
-text_set_element (__libc_thread_subfreeres, arena_thread_freeres);
-#endif
/*
* Local variables:
projects / thirdparty / glibc.git / blobdiff