Add arena-specific and selective dss allocation.

Add the "arenas.extend" mallctl, so that it is possible to create new
arenas that are outside the set that jemalloc automatically multiplexes
threads onto.

Add the ALLOCM_ARENA() flag for {,r,d}allocm(), so that it is possible
to explicitly allocate from a particular arena.

Add the "opt.dss" mallctl, which controls the default precedence of dss
allocation relative to mmap allocation.

Add the "arena.<i>.dss" mallctl, which makes it possible to set the
default dss precedence on a per arena or global basis.

Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge".

Add the "stats.arenas.<i>.dss" mallctl.

include/jemalloc/internal/jemalloc_internal.h.in

@@ -514,13 +514,19 @@ extern size_t	opt_narenas;
 /* Number of CPUs. */
 extern unsigned		ncpus;
 
-extern malloc_mutex_t	arenas_lock; /* Protects arenas initialization. */
+/* Protects arenas initialization (arenas, arenas_total). */
+extern malloc_mutex_t	arenas_lock;
 /*
  * Arenas that are used to service external requests.  Not all elements of the
  * arenas array are necessarily used; arenas are created lazily as needed.
+ *
+ * arenas[0..narenas_auto) are used for automatic multiplexing of threads and
+ * arenas.  arenas[narenas_auto..narenas_total) are only used if the application
+ * takes some action to create them and allocate from them.
  */
 extern arena_t		**arenas;
-extern unsigned		narenas;
+extern unsigned		narenas_total;
+extern unsigned		narenas_auto; /* Read-only after initialization. */
 
 arena_t	*arenas_extend(unsigned ind);
 void	arenas_cleanup(void *arg);
@@ -575,6 +581,7 @@ malloc_tsd_protos(JEMALLOC_ATTR(unused), arenas, arena_t *)
 
 size_t	s2u(size_t size);
 size_t	sa2u(size_t size, size_t alignment);
+unsigned	narenas_total_get(void);
 arena_t	*choose_arena(arena_t *arena);
 #endif
 
@@ -679,6 +686,18 @@ sa2u(size_t size, size_t alignment)
 	}
 }
 
+JEMALLOC_INLINE unsigned
+narenas_total_get(void)
+{
+	unsigned narenas;
+
+	malloc_mutex_lock(&arenas_lock);
+	narenas = narenas_total;
+	malloc_mutex_unlock(&arenas_lock);
+
+	return (narenas);
+}
+
 /* Choose an arena based on a per-thread value. */
 JEMALLOC_INLINE arena_t *
 choose_arena(arena_t *arena)
@@ -714,15 +733,24 @@ choose_arena(arena_t *arena)
 #include "jemalloc/internal/quarantine.h"
 
 #ifndef JEMALLOC_ENABLE_INLINE
+void	*imallocx(size_t size, bool try_tcache, arena_t *arena);
 void	*imalloc(size_t size);
+void	*icallocx(size_t size, bool try_tcache, arena_t *arena);
 void	*icalloc(size_t size);
+void	*ipallocx(size_t usize, size_t alignment, bool zero, bool try_tcache,
+    arena_t *arena);
 void	*ipalloc(size_t usize, size_t alignment, bool zero);
 size_t	isalloc(const void *ptr, bool demote);
 size_t	ivsalloc(const void *ptr, bool demote);
 size_t	u2rz(size_t usize);
 size_t	p2rz(const void *ptr);
+void	idallocx(void *ptr, bool try_tcache);
 void	idalloc(void *ptr);
+void	iqallocx(void *ptr, bool try_tcache);
 void	iqalloc(void *ptr);
+void	*irallocx(void *ptr, size_t size, size_t extra, size_t alignment,
+    bool zero, bool no_move, bool try_tcache_alloc, bool try_tcache_dalloc,
+    arena_t *arena);
 void	*iralloc(void *ptr, size_t size, size_t extra, size_t alignment,
     bool zero, bool no_move);
 malloc_tsd_protos(JEMALLOC_ATTR(unused), thread_allocated, thread_allocated_t)
@@ -730,29 +758,44 @@ malloc_tsd_protos(JEMALLOC_ATTR(unused), thread_allocated, thread_allocated_t)
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_C_))
 JEMALLOC_INLINE void *
-imalloc(size_t size)
+imallocx(size_t size, bool try_tcache, arena_t *arena)
 {
 
 	assert(size != 0);
 
 	if (size <= arena_maxclass)
-		return (arena_malloc(NULL, size, false, true));
+		return (arena_malloc(arena, size, false, try_tcache));
 	else
 		return (huge_malloc(size, false));
 }
 
+JEMALLOC_INLINE void *
+imalloc(size_t size)
+{
+
+	return (imallocx(size, true, NULL));
+}
+
+JEMALLOC_INLINE void *
+icallocx(size_t size, bool try_tcache, arena_t *arena)
+{
+
+	if (size <= arena_maxclass)
+		return (arena_malloc(arena, size, true, try_tcache));
+	else
+		return (huge_malloc(size, true));
+}
+
 JEMALLOC_INLINE void *
 icalloc(size_t size)
 {
 
-	if (size <= arena_maxclass)
-		return (arena_malloc(NULL, size, true, true));
-	else
-		return (huge_malloc(size, true));
+	return (icallocx(size, true, NULL));
 }
 
 JEMALLOC_INLINE void *
-ipalloc(size_t usize, size_t alignment, bool zero)
+ipallocx(size_t usize, size_t alignment, bool zero, bool try_tcache,
+    arena_t *arena)
 {
 	void *ret;
 
@@ -760,11 +803,11 @@ ipalloc(size_t usize, size_t alignment, bool zero)
 	assert(usize == sa2u(usize, alignment));
 
 	if (usize <= arena_maxclass && alignment <= PAGE)
-		ret = arena_malloc(NULL, usize, zero, true);
+		ret = arena_malloc(arena, usize, zero, try_tcache);
 	else {
 		if (usize <= arena_maxclass) {
-			ret = arena_palloc(choose_arena(NULL), usize, alignment,
-			    zero);
+			ret = arena_palloc(choose_arena(arena), usize,
+			    alignment, zero);
 		} else if (alignment <= chunksize)
 			ret = huge_malloc(usize, zero);
 		else
@@ -775,6 +818,13 @@ ipalloc(size_t usize, size_t alignment, bool zero)
 	return (ret);
 }
 
+JEMALLOC_INLINE void *
+ipalloc(size_t usize, size_t alignment, bool zero)
+{
+
+	return (ipallocx(usize, alignment, zero, true, NULL));
+}
+
 /*
  * Typical usage:
  *   void *ptr = [...]
@@ -833,7 +883,7 @@ p2rz(const void *ptr)
 }
 
 JEMALLOC_INLINE void
-idalloc(void *ptr)
+idallocx(void *ptr, bool try_tcache)
 {
 	arena_chunk_t *chunk;
 
@@ -841,24 +891,38 @@ idalloc(void *ptr)
 
 	chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
 	if (chunk != ptr)
-		arena_dalloc(chunk->arena, chunk, ptr, true);
+		arena_dalloc(chunk->arena, chunk, ptr, try_tcache);
 	else
 		huge_dalloc(ptr, true);
 }
 
+JEMALLOC_INLINE void
+idalloc(void *ptr)
+{
+
+	idallocx(ptr, true);
+}
+
+JEMALLOC_INLINE void
+iqallocx(void *ptr, bool try_tcache)
+{
+
+	if (config_fill && opt_quarantine)
+		quarantine(ptr);
+	else
+		idallocx(ptr, try_tcache);
+}
+
 JEMALLOC_INLINE void
 iqalloc(void *ptr)
 {
 
-	if (config_fill && opt_quarantine)
-		quarantine(ptr);
-	else
-		idalloc(ptr);
+	iqallocx(ptr, true);
 }
 
 JEMALLOC_INLINE void *
-iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
-    bool no_move)
+irallocx(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
+    bool no_move, bool try_tcache_alloc, bool try_tcache_dalloc, arena_t *arena)
 {
 	void *ret;
 	size_t oldsize;
@@ -881,7 +945,7 @@ iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
 		usize = sa2u(size + extra, alignment);
 		if (usize == 0)
 			return (NULL);
-		ret = ipalloc(usize, alignment, zero);
+		ret = ipallocx(usize, alignment, zero, try_tcache_alloc, arena);
 		if (ret == NULL) {
 			if (extra == 0)
 				return (NULL);
@@ -889,7 +953,8 @@ iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
 			usize = sa2u(size, alignment);
 			if (usize == 0)
 				return (NULL);
-			ret = ipalloc(usize, alignment, zero);
+			ret = ipallocx(usize, alignment, zero, try_tcache_alloc,
+			    arena);
 			if (ret == NULL)
 				return (NULL);
 		}
@@ -900,7 +965,7 @@ iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
 		 */
 		copysize = (size < oldsize) ? size : oldsize;
 		memcpy(ret, ptr, copysize);
-		iqalloc(ptr);
+		iqallocx(ptr, try_tcache_dalloc);
 		return (ret);
 	}
 
@@ -914,15 +979,25 @@ 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, true));
+			return (arena_ralloc(arena, ptr, oldsize, size, extra,
+			    alignment, zero, try_tcache_alloc,
+			    try_tcache_dalloc));
 		} else {
 			return (huge_ralloc(ptr, oldsize, size, extra,
-			    alignment, zero));
+			    alignment, zero, try_tcache_dalloc));
 		}
 	}
 }
 
+JEMALLOC_INLINE void *
+iralloc(void *ptr, size_t size, size_t extra, size_t alignment, bool zero,
+    bool no_move)
+{
+
+	return (irallocx(ptr, size, extra, alignment, zero, no_move, true, true,
+	    NULL));
+}
+
 malloc_tsd_externs(thread_allocated, thread_allocated_t)
 malloc_tsd_funcs(JEMALLOC_INLINE, thread_allocated, thread_allocated_t,
     THREAD_ALLOCATED_INITIALIZER, malloc_tsd_no_cleanup)