diff options
Diffstat (limited to 'contrib/jemalloc/src/jemalloc.c')
| -rw-r--r-- | contrib/jemalloc/src/jemalloc.c | 212 |
1 files changed, 155 insertions, 57 deletions
diff --git a/contrib/jemalloc/src/jemalloc.c b/contrib/jemalloc/src/jemalloc.c index fa9fcf0..aaf5012 100644 --- a/contrib/jemalloc/src/jemalloc.c +++ b/contrib/jemalloc/src/jemalloc.c @@ -37,7 +37,8 @@ unsigned ncpus; malloc_mutex_t arenas_lock; arena_t **arenas; -unsigned narenas; +unsigned narenas_total; +unsigned narenas_auto; /* Set to true once the allocator has been initialized. */ static bool malloc_initialized = false; @@ -148,14 +149,14 @@ choose_arena_hard(void) { arena_t *ret; - if (narenas > 1) { + if (narenas_auto > 1) { unsigned i, choose, first_null; choose = 0; - first_null = narenas; + first_null = narenas_auto; malloc_mutex_lock(&arenas_lock); assert(arenas[0] != NULL); - for (i = 1; i < narenas; i++) { + for (i = 1; i < narenas_auto; i++) { if (arenas[i] != NULL) { /* * Choose the first arena that has the lowest @@ -164,7 +165,7 @@ choose_arena_hard(void) if (arenas[i]->nthreads < arenas[choose]->nthreads) choose = i; - } else if (first_null == narenas) { + } else if (first_null == narenas_auto) { /* * Record the index of the first uninitialized * arena, in case all extant arenas are in use. @@ -178,7 +179,8 @@ choose_arena_hard(void) } } - if (arenas[choose]->nthreads == 0 || first_null == narenas) { + if (arenas[choose]->nthreads == 0 + || first_null == narenas_auto) { /* * Use an unloaded arena, or the least loaded arena if * all arenas are already initialized. @@ -207,7 +209,7 @@ stats_print_atexit(void) { if (config_tcache && config_stats) { - unsigned i; + unsigned narenas, i; /* * Merge stats from extant threads. This is racy, since @@ -216,7 +218,7 @@ stats_print_atexit(void) * out of date by the time they are reported, if other threads * continue to allocate. */ - for (i = 0; i < narenas; i++) { + for (i = 0, narenas = narenas_total_get(); i < narenas; i++) { arena_t *arena = arenas[i]; if (arena != NULL) { tcache_t *tcache; @@ -258,12 +260,13 @@ malloc_ncpus(void) result = si.dwNumberOfProcessors; #else result = sysconf(_SC_NPROCESSORS_ONLN); +#endif if (result == -1) { /* Error. */ ret = 1; - } -#endif - ret = (unsigned)result; + } else { + ret = (unsigned)result; + } return (ret); } @@ -381,6 +384,22 @@ malloc_conf_init(void) const char *opts, *k, *v; size_t klen, vlen; + /* + * Automatically configure valgrind before processing options. The + * valgrind option remains in jemalloc 3.x for compatibility reasons. + */ + if (config_valgrind) { + opt_valgrind = (RUNNING_ON_VALGRIND != 0) ? true : false; + if (config_fill && opt_valgrind) { + opt_junk = false; + assert(opt_zero == false); + opt_quarantine = JEMALLOC_VALGRIND_QUARANTINE_DEFAULT; + opt_redzone = true; + } + if (config_tcache && opt_valgrind) + opt_tcache = false; + } + for (i = 0; i < 3; i++) { /* Get runtime configuration. */ switch (i) { @@ -542,6 +561,30 @@ malloc_conf_init(void) */ CONF_HANDLE_SIZE_T(opt_lg_chunk, "lg_chunk", LG_PAGE + (config_fill ? 2 : 1), (sizeof(size_t) << 3) - 1) + if (strncmp("dss", k, klen) == 0) { + int i; + bool match = false; + for (i = 0; i < dss_prec_limit; i++) { + if (strncmp(dss_prec_names[i], v, vlen) + == 0) { + if (chunk_dss_prec_set(i)) { + malloc_conf_error( + "Error setting dss", + k, klen, v, vlen); + } else { + opt_dss = + dss_prec_names[i]; + match = true; + break; + } + } + } + if (match == false) { + malloc_conf_error("Invalid conf value", + k, klen, v, vlen); + } + continue; + } CONF_HANDLE_SIZE_T(opt_narenas, "narenas", 1, SIZE_T_MAX) CONF_HANDLE_SSIZE_T(opt_lg_dirty_mult, "lg_dirty_mult", @@ -558,20 +601,7 @@ malloc_conf_init(void) CONF_HANDLE_BOOL(opt_utrace, "utrace") } if (config_valgrind) { - bool hit; - CONF_HANDLE_BOOL_HIT(opt_valgrind, - "valgrind", hit) - if (config_fill && opt_valgrind && hit) { - opt_junk = false; - opt_zero = false; - if (opt_quarantine == 0) { - opt_quarantine = - JEMALLOC_VALGRIND_QUARANTINE_DEFAULT; - } - opt_redzone = true; - } - if (hit) - continue; + CONF_HANDLE_BOOL(opt_valgrind, "valgrind") } if (config_xmalloc) { CONF_HANDLE_BOOL(opt_xmalloc, "xmalloc") @@ -700,9 +730,9 @@ malloc_init_hard(void) * Create enough scaffolding to allow recursive allocation in * malloc_ncpus(). */ - narenas = 1; + narenas_total = narenas_auto = 1; arenas = init_arenas; - memset(arenas, 0, sizeof(arena_t *) * narenas); + memset(arenas, 0, sizeof(arena_t *) * narenas_auto); /* * Initialize one arena here. The rest are lazily created in @@ -760,20 +790,21 @@ malloc_init_hard(void) else opt_narenas = 1; } - narenas = opt_narenas; + narenas_auto = opt_narenas; /* * Make sure that the arenas array can be allocated. In practice, this * limit is enough to allow the allocator to function, but the ctl * machinery will fail to allocate memory at far lower limits. */ - if (narenas > chunksize / sizeof(arena_t *)) { - narenas = chunksize / sizeof(arena_t *); + if (narenas_auto > chunksize / sizeof(arena_t *)) { + narenas_auto = chunksize / sizeof(arena_t *); malloc_printf("<jemalloc>: Reducing narenas to limit (%d)\n", - narenas); + narenas_auto); } + narenas_total = narenas_auto; /* Allocate and initialize arenas. */ - arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas); + arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas_total); if (arenas == NULL) { malloc_mutex_unlock(&init_lock); return (true); @@ -782,7 +813,7 @@ malloc_init_hard(void) * Zero the array. In practice, this should always be pre-zeroed, * since it was just mmap()ed, but let's be sure. */ - memset(arenas, 0, sizeof(arena_t *) * narenas); + memset(arenas, 0, sizeof(arena_t *) * narenas_total); /* Copy the pointer to the one arena that was already initialized. */ arenas[0] = init_arenas[0]; @@ -1267,11 +1298,10 @@ je_valloc(size_t size) * passed an extra argument for the caller return address, which will be * ignored. */ -JEMALLOC_EXPORT void (* const __free_hook)(void *ptr) = je_free; -JEMALLOC_EXPORT void *(* const __malloc_hook)(size_t size) = je_malloc; -JEMALLOC_EXPORT void *(* const __realloc_hook)(void *ptr, size_t size) = - je_realloc; -JEMALLOC_EXPORT void *(* const __memalign_hook)(size_t alignment, size_t size) = +JEMALLOC_EXPORT void (* __free_hook)(void *ptr) = je_free; +JEMALLOC_EXPORT void *(* __malloc_hook)(size_t size) = je_malloc; +JEMALLOC_EXPORT void *(* __realloc_hook)(void *ptr, size_t size) = je_realloc; +JEMALLOC_EXPORT void *(* __memalign_hook)(size_t alignment, size_t size) = je_memalign; #endif @@ -1284,7 +1314,7 @@ JEMALLOC_EXPORT void *(* const __memalign_hook)(size_t alignment, size_t size) = */ size_t -je_malloc_usable_size(const void *ptr) +je_malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *ptr) { size_t ret; @@ -1348,18 +1378,19 @@ je_mallctlbymib(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp, #ifdef JEMALLOC_EXPERIMENTAL JEMALLOC_INLINE void * -iallocm(size_t usize, size_t alignment, bool zero) +iallocm(size_t usize, size_t alignment, bool zero, bool try_tcache, + arena_t *arena) { assert(usize == ((alignment == 0) ? s2u(usize) : sa2u(usize, alignment))); if (alignment != 0) - return (ipalloc(usize, alignment, zero)); + return (ipallocx(usize, alignment, zero, try_tcache, arena)); else if (zero) - return (icalloc(usize)); + return (icallocx(usize, try_tcache, arena)); else - return (imalloc(usize)); + return (imallocx(usize, try_tcache, arena)); } int @@ -1370,6 +1401,9 @@ je_allocm(void **ptr, size_t *rsize, size_t size, int flags) size_t alignment = (ZU(1) << (flags & ALLOCM_LG_ALIGN_MASK) & (SIZE_T_MAX-1)); bool zero = flags & ALLOCM_ZERO; + unsigned arena_ind = ((unsigned)(flags >> 8)) - 1; + arena_t *arena; + bool try_tcache; assert(ptr != NULL); assert(size != 0); @@ -1377,6 +1411,14 @@ je_allocm(void **ptr, size_t *rsize, size_t size, int flags) if (malloc_init()) goto label_oom; + if (arena_ind != UINT_MAX) { + arena = arenas[arena_ind]; + try_tcache = false; + } else { + arena = NULL; + try_tcache = true; + } + usize = (alignment == 0) ? s2u(size) : sa2u(size, alignment); if (usize == 0) goto label_oom; @@ -1393,18 +1435,19 @@ je_allocm(void **ptr, size_t *rsize, size_t size, int flags) s2u(SMALL_MAXCLASS+1) : sa2u(SMALL_MAXCLASS+1, alignment); assert(usize_promoted != 0); - p = iallocm(usize_promoted, alignment, zero); + p = iallocm(usize_promoted, alignment, zero, + try_tcache, arena); if (p == NULL) goto label_oom; arena_prof_promoted(p, usize); } else { - p = iallocm(usize, alignment, zero); + p = iallocm(usize, alignment, zero, try_tcache, arena); if (p == NULL) goto label_oom; } prof_malloc(p, usize, cnt); } else { - p = iallocm(usize, alignment, zero); + p = iallocm(usize, alignment, zero, try_tcache, arena); if (p == NULL) goto label_oom; } @@ -1441,6 +1484,9 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags) & (SIZE_T_MAX-1)); bool zero = flags & ALLOCM_ZERO; bool no_move = flags & ALLOCM_NO_MOVE; + unsigned arena_ind = ((unsigned)(flags >> 8)) - 1; + bool try_tcache_alloc, try_tcache_dalloc; + arena_t *arena; assert(ptr != NULL); assert(*ptr != NULL); @@ -1448,6 +1494,19 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags) assert(SIZE_T_MAX - size >= extra); assert(malloc_initialized || IS_INITIALIZER); + if (arena_ind != UINT_MAX) { + arena_chunk_t *chunk; + try_tcache_alloc = true; + chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(*ptr); + try_tcache_dalloc = (chunk == *ptr || chunk->arena != + arenas[arena_ind]); + arena = arenas[arena_ind]; + } else { + try_tcache_alloc = true; + try_tcache_dalloc = true; + arena = NULL; + } + p = *ptr; if (config_prof && opt_prof) { prof_thr_cnt_t *cnt; @@ -1474,9 +1533,10 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags) if (prof_promote && (uintptr_t)cnt != (uintptr_t)1U && ((alignment == 0) ? s2u(size) : sa2u(size, alignment)) <= SMALL_MAXCLASS) { - q = iralloc(p, SMALL_MAXCLASS+1, (SMALL_MAXCLASS+1 >= + q = irallocx(p, SMALL_MAXCLASS+1, (SMALL_MAXCLASS+1 >= size+extra) ? 0 : size+extra - (SMALL_MAXCLASS+1), - alignment, zero, no_move); + alignment, zero, no_move, try_tcache_alloc, + try_tcache_dalloc, arena); if (q == NULL) goto label_err; if (max_usize < PAGE) { @@ -1485,7 +1545,8 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags) } else usize = isalloc(q, config_prof); } else { - q = iralloc(p, size, extra, alignment, zero, no_move); + q = irallocx(p, size, extra, alignment, zero, no_move, + try_tcache_alloc, try_tcache_dalloc, arena); if (q == NULL) goto label_err; usize = isalloc(q, config_prof); @@ -1502,7 +1563,8 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags) old_size = isalloc(p, false); old_rzsize = u2rz(old_size); } - q = iralloc(p, size, extra, alignment, zero, no_move); + q = irallocx(p, size, extra, alignment, zero, no_move, + try_tcache_alloc, try_tcache_dalloc, arena); if (q == NULL) goto label_err; if (config_stats) @@ -1563,10 +1625,19 @@ je_dallocm(void *ptr, int flags) { size_t usize; size_t rzsize JEMALLOC_CC_SILENCE_INIT(0); + unsigned arena_ind = ((unsigned)(flags >> 8)) - 1; + bool try_tcache; assert(ptr != NULL); assert(malloc_initialized || IS_INITIALIZER); + if (arena_ind != UINT_MAX) { + arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); + try_tcache = (chunk == ptr || chunk->arena != + arenas[arena_ind]); + } else + try_tcache = true; + UTRACE(ptr, 0, 0); if (config_stats || config_valgrind) usize = isalloc(ptr, config_prof); @@ -1579,7 +1650,7 @@ je_dallocm(void *ptr, int flags) thread_allocated_tsd_get()->deallocated += usize; if (config_valgrind && opt_valgrind) rzsize = p2rz(ptr); - iqalloc(ptr); + iqallocx(ptr, try_tcache); JEMALLOC_VALGRIND_FREE(ptr, rzsize); return (ALLOCM_SUCCESS); @@ -1616,6 +1687,27 @@ je_nallocm(size_t *rsize, size_t size, int flags) * malloc during fork(). */ +/* + * If an application creates a thread before doing any allocation in the main + * thread, then calls fork(2) in the main thread followed by memory allocation + * in the child process, a race can occur that results in deadlock within the + * child: the main thread may have forked while the created thread had + * partially initialized the allocator. Ordinarily jemalloc prevents + * fork/malloc races via the following functions it registers during + * initialization using pthread_atfork(), but of course that does no good if + * the allocator isn't fully initialized at fork time. The following library + * constructor is a partial solution to this problem. It may still possible to + * trigger the deadlock described above, but doing so would involve forking via + * a library constructor that runs before jemalloc's runs. + */ +JEMALLOC_ATTR(constructor) +static void +jemalloc_constructor(void) +{ + + malloc_init(); +} + #ifndef JEMALLOC_MUTEX_INIT_CB void jemalloc_prefork(void) @@ -1633,14 +1725,16 @@ _malloc_prefork(void) assert(malloc_initialized); /* Acquire all mutexes in a safe order. */ + ctl_prefork(); malloc_mutex_prefork(&arenas_lock); - for (i = 0; i < narenas; i++) { + for (i = 0; i < narenas_total; i++) { if (arenas[i] != NULL) arena_prefork(arenas[i]); } + prof_prefork(); + chunk_prefork(); base_prefork(); huge_prefork(); - chunk_dss_prefork(); } #ifndef JEMALLOC_MUTEX_INIT_CB @@ -1660,14 +1754,16 @@ _malloc_postfork(void) assert(malloc_initialized); /* Release all mutexes, now that fork() has completed. */ - chunk_dss_postfork_parent(); huge_postfork_parent(); base_postfork_parent(); - for (i = 0; i < narenas; i++) { + chunk_postfork_parent(); + prof_postfork_parent(); + for (i = 0; i < narenas_total; i++) { if (arenas[i] != NULL) arena_postfork_parent(arenas[i]); } malloc_mutex_postfork_parent(&arenas_lock); + ctl_postfork_parent(); } void @@ -1678,14 +1774,16 @@ jemalloc_postfork_child(void) assert(malloc_initialized); /* Release all mutexes, now that fork() has completed. */ - chunk_dss_postfork_child(); huge_postfork_child(); base_postfork_child(); - for (i = 0; i < narenas; i++) { + chunk_postfork_child(); + prof_postfork_child(); + for (i = 0; i < narenas_total; i++) { if (arenas[i] != NULL) arena_postfork_child(arenas[i]); } malloc_mutex_postfork_child(&arenas_lock); + ctl_postfork_child(); } /******************************************************************************/ |
