Add a0malloc(), a0calloc(), and a0free().

Add a0malloc(), a0calloc(), and a0free(), which are used by FreeBSD's
libc to allocate/deallocate TLS in static binaries.

include/jemalloc/internal/arena.h

@@ -373,7 +373,7 @@ void	arena_stats_merge(arena_t *arena, size_t *nactive, size_t *ndirty,
 void	*arena_ralloc_no_move(void *ptr, size_t oldsize, size_t size,
     size_t extra, bool zero);
 void	*arena_ralloc(void *ptr, size_t oldsize, size_t size, size_t extra,
-    size_t alignment, bool zero);
+    size_t alignment, bool zero, bool try_tcache);
 bool	arena_new(arena_t *arena, unsigned ind);
 void	arena_boot(void);
 void	arena_prefork(arena_t *arena);
@@ -390,9 +390,10 @@ unsigned	arena_run_regind(arena_run_t *run, arena_bin_info_t *bin_info,
     const void *ptr);
 prof_ctx_t	*arena_prof_ctx_get(const void *ptr);
 void	arena_prof_ctx_set(const void *ptr, prof_ctx_t *ctx);
-void	*arena_malloc(size_t size, bool zero);
+void	*arena_malloc(arena_t *arena, size_t size, bool zero, bool try_tcache);
 void	*arena_malloc_prechosen(arena_t *arena, size_t size, bool zero);
-void	arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr);
+void	arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr,
+    bool try_tcache);
 #endif
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_ARENA_C_))
@@ -548,7 +549,7 @@ arena_prof_ctx_set(const void *ptr, prof_ctx_t *ctx)
 }
 
 JEMALLOC_INLINE void *
-arena_malloc(size_t size, bool zero)
+arena_malloc(arena_t *arena, size_t size, bool zero, bool try_tcache)
 {
 	tcache_t *tcache;
 
@@ -556,20 +557,24 @@ arena_malloc(size_t size, bool zero)
 	assert(size <= arena_maxclass);
 
 	if (size <= SMALL_MAXCLASS) {
-		if ((tcache = tcache_get(true)) != NULL)
+		if (try_tcache && (tcache = tcache_get(true)) != NULL)
 			return (tcache_alloc_small(tcache, size, zero));
-		else
+		else {
-			return (arena_malloc_small(choose_arena(), size, zero));
+			return (arena_malloc_small(choose_arena(arena), size,
+			    zero));
+		}
 	} else {
 		/*
 		 * Initialize tcache after checking size in order to avoid
 		 * infinite recursion during tcache initialization.
 		 */
-		if (size <= tcache_maxclass && (tcache = tcache_get(true)) !=
+		if (try_tcache && size <= tcache_maxclass && (tcache =
-		    NULL)
+		    tcache_get(true)) != NULL)
 			return (tcache_alloc_large(tcache, size, zero));
-		else
+		else {
-			return (arena_malloc_large(choose_arena(), size, zero));
+			return (arena_malloc_large(choose_arena(arena), size,
+			    zero));
+		}
 	}
 }
 
@@ -587,11 +592,11 @@ arena_malloc_prechosen(arena_t *arena, size_t size, bool zero)
 }
 
 JEMALLOC_INLINE void
-arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr)
+arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr, bool try_tcache)
 {
 	size_t pageind;
 	arena_chunk_map_t *mapelm;
-	tcache_t *tcache = tcache_get(false);
+	tcache_t *tcache;
 
 	assert(arena != NULL);
 	assert(chunk->arena == arena);
@@ -603,7 +608,7 @@ arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr)
 	assert((mapelm->bits & CHUNK_MAP_ALLOCATED) != 0);
 	if ((mapelm->bits & CHUNK_MAP_LARGE) == 0) {
 		/* Small allocation. */
-		if (tcache != NULL)
+		if (try_tcache && (tcache = tcache_get(false)) != NULL)
 			tcache_dalloc_small(tcache, ptr);
 		else {
 			arena_run_t *run;
@@ -630,7 +635,8 @@ arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr)
 
 		assert(((uintptr_t)ptr & PAGE_MASK) == 0);
 
-		if (size <= tcache_maxclass && tcache != NULL) {
+		if (try_tcache && size <= tcache_maxclass && (tcache =
+		    tcache_get(false)) != NULL) {
 			tcache_dalloc_large(tcache, ptr, size);
 		} else {
 			malloc_mutex_lock(&arena->lock);

include/jemalloc/internal/jemalloc_internal.h.in

@@ -399,7 +399,7 @@ malloc_tsd_protos(JEMALLOC_ATTR(unused), arenas, arena_t *)
 
 size_t	s2u(size_t size);
 size_t	sa2u(size_t size, size_t alignment, size_t *run_size_p);
-arena_t	*choose_arena(void);
+arena_t	*choose_arena(arena_t *arena);
 #endif
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_C_))
@@ -517,10 +517,13 @@ sa2u(size_t size, size_t alignment, size_t *run_size_p)
 
 /* Choose an arena based on a per-thread value. */
 JEMALLOC_INLINE arena_t *
-choose_arena(void)
+choose_arena(arena_t *arena)
 {
 	arena_t *ret;
 
+	if (arena != NULL)
+		return (arena);
+
 	if ((ret = *arenas_tsd_get()) == NULL) {
 		ret = choose_arena_hard();
 		assert(ret != NULL);
@@ -556,7 +559,7 @@ imalloc(size_t size)
 	assert(size != 0);
 
 	if (size <= arena_maxclass)
-		return (arena_malloc(size, false));
+		return (arena_malloc(NULL, size, false, true));
 	else
 		return (huge_malloc(size, false));
 }
@@ -566,7 +569,7 @@ icalloc(size_t size)
 {
 
 	if (size <= arena_maxclass)
-		return (arena_malloc(size, true));
+		return (arena_malloc(NULL, size, true, true));
 	else
 		return (huge_malloc(size, true));
 }
@@ -580,7 +583,7 @@ ipalloc(size_t usize, size_t alignment, bool zero)
 	assert(usize == sa2u(usize, alignment, NULL));
 
 	if (usize <= arena_maxclass && alignment <= PAGE)
-		ret = arena_malloc(usize, zero);
+		ret = arena_malloc(NULL, usize, zero, true);
 	else {
 		size_t run_size JEMALLOC_CC_SILENCE_INIT(0);
 
@@ -594,7 +597,7 @@ ipalloc(size_t usize, size_t alignment, bool zero)
 		 */
 		sa2u(usize, alignment, &run_size);
 		if (run_size <= arena_maxclass) {
-			ret = arena_palloc(choose_arena(), usize, run_size,
+			ret = arena_palloc(choose_arena(NULL), usize, run_size,
 			    alignment, zero);
 		} else if (alignment <= chunksize)
 			ret = huge_malloc(usize, zero);
@@ -647,7 +650,7 @@ idalloc(void *ptr)
 
 	chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
 	if (chunk != ptr)
-		arena_dalloc(chunk->arena, chunk, ptr);
+		arena_dalloc(chunk->arena, chunk, ptr, true);
 	else
 		huge_dalloc(ptr, true);
 }
@@ -711,7 +714,7 @@ iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
 	} else {
 		if (size + extra <= arena_maxclass) {
 			return (arena_ralloc(ptr, oldsize, size, extra,
-			    alignment, zero));
+			    alignment, zero, true));
 		} else {
 			return (huge_ralloc(ptr, oldsize, size, extra,
 			    alignment, zero));

include/jemalloc/internal/tcache.h

@@ -237,7 +237,7 @@ tcache_get(bool create)
 				tcache_enabled_set(false); /* Memoize. */
 				return (NULL);
 			}
-			return (tcache_create(choose_arena()));
+			return (tcache_create(choose_arena(NULL)));
 		}
 		if (tcache == TCACHE_STATE_PURGATORY) {
 			/*

src/arena.c

@@ -1888,7 +1888,7 @@ arena_ralloc_no_move(void *ptr, size_t oldsize, size_t size, size_t extra,
 
 void *
 arena_ralloc(void *ptr, size_t oldsize, size_t size, size_t extra,
-    size_t alignment, bool zero)
+    size_t alignment, bool zero, bool try_tcache)
 {
 	void *ret;
 	size_t copysize;
@@ -1909,7 +1909,7 @@ arena_ralloc(void *ptr, size_t oldsize, size_t size, size_t extra,
 			return (NULL);
 		ret = ipalloc(usize, alignment, zero);
 	} else
-		ret = arena_malloc(size + extra, zero);
+		ret = arena_malloc(NULL, size + extra, zero, try_tcache);
 
 	if (ret == NULL) {
 		if (extra == 0)
@@ -1921,7 +1921,7 @@ arena_ralloc(void *ptr, size_t oldsize, size_t size, size_t extra,
 				return (NULL);
 			ret = ipalloc(usize, alignment, zero);
 		} else
-			ret = arena_malloc(size, zero);
+			ret = arena_malloc(NULL, size, zero, try_tcache);
 
 		if (ret == NULL)
 			return (NULL);

src/ctl.c

@@ -1016,7 +1016,7 @@ thread_arena_ctl(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp,
 	int ret;
 	unsigned newind, oldind;
 
-	newind = oldind = choose_arena()->ind;
+	newind = oldind = choose_arena(NULL)->ind;
 	WRITE(newind, unsigned);
 	READ(oldind, unsigned);
 	if (newind != oldind) {

src/jemalloc.c

@@ -1487,7 +1487,6 @@ je_nallocm(size_t *rsize, size_t size, int flags)
  * End experimental functions.
  */
 /******************************************************************************/
-
 /*
  * The following functions are used by threading libraries for protection of
  * malloc during fork().
@@ -1552,3 +1551,55 @@ jemalloc_postfork_child(void)
 }
 
 /******************************************************************************/
+/*
+ * The following functions are used for TLS allocation/deallocation in static
+ * binaries on FreeBSD.  The primary difference between these and i[mcd]alloc()
+ * is that these avoid accessing TLS variables.
+ */
+
+static void *
+a0alloc(size_t size, bool zero)
+{
+
+	if (malloc_init())
+		return (NULL);
+
+	if (size == 0)
+		size = 1;
+
+	if (size <= arena_maxclass)
+		return (arena_malloc(arenas[0], size, zero, false));
+	else
+		return (huge_malloc(size, zero));
+}
+
+void *
+a0malloc(size_t size)
+{
+
+	return (a0alloc(size, false));
+}
+
+void *
+a0calloc(size_t num, size_t size)
+{
+
+	return (a0alloc(num * size, true));
+}
+
+void
+a0free(void *ptr)
+{
+	arena_chunk_t *chunk;
+
+	if (ptr == NULL)
+		return;
+
+	chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
+	if (chunk != ptr)
+		arena_dalloc(chunk->arena, chunk, ptr, false);
+	else
+		huge_dalloc(ptr, true);
+}
+
+/******************************************************************************/