diff --git a/Makefile.in b/Makefile.in
index f60823f5..4b0e184f 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -138,6 +138,7 @@ TESTS_UNIT := $(srcroot)test/unit/atomic.c \
 	$(srcroot)test/unit/bitmap.c \
 	$(srcroot)test/unit/ckh.c \
 	$(srcroot)test/unit/decay.c \
+	$(srcroot)test/unit/fork.c \
 	$(srcroot)test/unit/hash.c \
 	$(srcroot)test/unit/junk.c \
 	$(srcroot)test/unit/junk_alloc.c \
diff --git a/include/jemalloc/internal/arena.h b/include/jemalloc/internal/arena.h
index 3519873c..42a78960 100644
--- a/include/jemalloc/internal/arena.h
+++ b/include/jemalloc/internal/arena.h
@@ -584,7 +584,10 @@ void	arena_nthreads_inc(arena_t *arena);
 void	arena_nthreads_dec(arena_t *arena);
 arena_t	*arena_new(unsigned ind);
 bool	arena_boot(void);
-void	arena_prefork(arena_t *arena);
+void	arena_prefork0(arena_t *arena);
+void	arena_prefork1(arena_t *arena);
+void	arena_prefork2(arena_t *arena);
+void	arena_prefork3(arena_t *arena);
 void	arena_postfork_parent(arena_t *arena);
 void	arena_postfork_child(arena_t *arena);
 
diff --git a/include/jemalloc/internal/private_symbols.txt b/include/jemalloc/internal/private_symbols.txt
index 5fcc6692..fafee81f 100644
--- a/include/jemalloc/internal/private_symbols.txt
+++ b/include/jemalloc/internal/private_symbols.txt
@@ -81,7 +81,10 @@ arena_nthreads_inc
 arena_palloc
 arena_postfork_child
 arena_postfork_parent
-arena_prefork
+arena_prefork0
+arena_prefork1
+arena_prefork2
+arena_prefork3
 arena_prof_accum
 arena_prof_accum_impl
 arena_prof_accum_locked
@@ -408,7 +411,8 @@ prof_malloc_sample_object
 prof_mdump
 prof_postfork_child
 prof_postfork_parent
-prof_prefork
+prof_prefork0
+prof_prefork1
 prof_realloc
 prof_reset
 prof_sample_accum_update
diff --git a/include/jemalloc/internal/prof.h b/include/jemalloc/internal/prof.h
index a25502a9..48dd6cc5 100644
--- a/include/jemalloc/internal/prof.h
+++ b/include/jemalloc/internal/prof.h
@@ -316,7 +316,8 @@ bool	prof_gdump_set(bool active);
 void	prof_boot0(void);
 void	prof_boot1(void);
 bool	prof_boot2(void);
-void	prof_prefork(void);
+void	prof_prefork0(void);
+void	prof_prefork1(void);
 void	prof_postfork_parent(void);
 void	prof_postfork_child(void);
 void	prof_sample_threshold_update(prof_tdata_t *tdata);
diff --git a/src/arena.c b/src/arena.c
index b7645d8b..48e9b20b 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -3646,16 +3646,34 @@ arena_boot(void)
 }
 
 void
-arena_prefork(arena_t *arena)
+arena_prefork0(arena_t *arena)
+{
+
+	malloc_mutex_prefork(&arena->lock);
+}
+
+void
+arena_prefork1(arena_t *arena)
+{
+
+	malloc_mutex_prefork(&arena->chunks_mtx);
+}
+
+void
+arena_prefork2(arena_t *arena)
+{
+
+	malloc_mutex_prefork(&arena->node_cache_mtx);
+}
+
+void
+arena_prefork3(arena_t *arena)
 {
 	unsigned i;
 
-	malloc_mutex_prefork(&arena->lock);
-	malloc_mutex_prefork(&arena->huge_mtx);
-	malloc_mutex_prefork(&arena->chunks_mtx);
-	malloc_mutex_prefork(&arena->node_cache_mtx);
 	for (i = 0; i < NBINS; i++)
 		malloc_mutex_prefork(&arena->bins[i].lock);
+	malloc_mutex_prefork(&arena->huge_mtx);
 }
 
 void
@@ -3663,11 +3681,11 @@ arena_postfork_parent(arena_t *arena)
 {
 	unsigned i;
 
+	malloc_mutex_postfork_parent(&arena->huge_mtx);
 	for (i = 0; i < NBINS; i++)
 		malloc_mutex_postfork_parent(&arena->bins[i].lock);
 	malloc_mutex_postfork_parent(&arena->node_cache_mtx);
 	malloc_mutex_postfork_parent(&arena->chunks_mtx);
-	malloc_mutex_postfork_parent(&arena->huge_mtx);
 	malloc_mutex_postfork_parent(&arena->lock);
 }
 
@@ -3676,10 +3694,10 @@ arena_postfork_child(arena_t *arena)
 {
 	unsigned i;
 
+	malloc_mutex_postfork_child(&arena->huge_mtx);
 	for (i = 0; i < NBINS; i++)
 		malloc_mutex_postfork_child(&arena->bins[i].lock);
 	malloc_mutex_postfork_child(&arena->node_cache_mtx);
 	malloc_mutex_postfork_child(&arena->chunks_mtx);
-	malloc_mutex_postfork_child(&arena->huge_mtx);
 	malloc_mutex_postfork_child(&arena->lock);
 }
diff --git a/src/jemalloc.c b/src/jemalloc.c
index 0735376e..71207913 100644
--- a/src/jemalloc.c
+++ b/src/jemalloc.c
@@ -2644,7 +2644,8 @@ JEMALLOC_EXPORT void
 _malloc_prefork(void)
 #endif
 {
-	unsigned i, narenas;
+	unsigned i, j, narenas;
+	arena_t *arena;
 
 #ifdef JEMALLOC_MUTEX_INIT_CB
 	if (!malloc_initialized())
@@ -2652,18 +2653,31 @@ _malloc_prefork(void)
 #endif
 	assert(malloc_initialized());
 
+	narenas = narenas_total_get();
+
 	/* Acquire all mutexes in a safe order. */
 	ctl_prefork();
-	prof_prefork();
 	malloc_mutex_prefork(&arenas_lock);
-	for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
-		arena_t *arena;
-
-		if ((arena = arena_get(i, false)) != NULL)
-			arena_prefork(arena);
+	prof_prefork0();
+	for (i = 0; i < 3; i++) {
+		for (j = 0; j < narenas; j++) {
+			if ((arena = arena_get(j, false)) != NULL) {
+				switch (i) {
+				case 0: arena_prefork0(arena); break;
+				case 1: arena_prefork1(arena); break;
+				case 2: arena_prefork2(arena); break;
+				default: not_reached();
+				}
+			}
+		}
 	}
-	chunk_prefork();
 	base_prefork();
+	chunk_prefork();
+	for (i = 0; i < narenas; i++) {
+		if ((arena = arena_get(i, false)) != NULL)
+			arena_prefork3(arena);
+	}
+	prof_prefork1();
 }
 
 #ifndef JEMALLOC_MUTEX_INIT_CB
@@ -2683,16 +2697,16 @@ _malloc_postfork(void)
 	assert(malloc_initialized());
 
 	/* Release all mutexes, now that fork() has completed. */
-	base_postfork_parent();
 	chunk_postfork_parent();
+	base_postfork_parent();
 	for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
 		arena_t *arena;
 
 		if ((arena = arena_get(i, false)) != NULL)
 			arena_postfork_parent(arena);
 	}
-	malloc_mutex_postfork_parent(&arenas_lock);
 	prof_postfork_parent();
+	malloc_mutex_postfork_parent(&arenas_lock);
 	ctl_postfork_parent();
 }
 
@@ -2704,16 +2718,16 @@ jemalloc_postfork_child(void)
 	assert(malloc_initialized());
 
 	/* Release all mutexes, now that fork() has completed. */
-	base_postfork_child();
 	chunk_postfork_child();
+	base_postfork_child();
 	for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
 		arena_t *arena;
 
 		if ((arena = arena_get(i, false)) != NULL)
 			arena_postfork_child(arena);
 	}
-	malloc_mutex_postfork_child(&arenas_lock);
 	prof_postfork_child();
+	malloc_mutex_postfork_child(&arenas_lock);
 	ctl_postfork_child();
 }
 
diff --git a/src/prof.c b/src/prof.c
index b3872277..a92320d4 100644
--- a/src/prof.c
+++ b/src/prof.c
@@ -2198,20 +2198,32 @@ prof_boot2(void)
 }
 
 void
-prof_prefork(void)
+prof_prefork0(void)
 {
 
 	if (opt_prof) {
 		unsigned i;
 
-		malloc_mutex_prefork(&tdatas_mtx);
+		malloc_mutex_prefork(&prof_dump_mtx);
 		malloc_mutex_prefork(&bt2gctx_mtx);
-		malloc_mutex_prefork(&next_thr_uid_mtx);
-		malloc_mutex_prefork(&prof_dump_seq_mtx);
-		for (i = 0; i < PROF_NCTX_LOCKS; i++)
-			malloc_mutex_prefork(&gctx_locks[i]);
+		malloc_mutex_prefork(&tdatas_mtx);
 		for (i = 0; i < PROF_NTDATA_LOCKS; i++)
 			malloc_mutex_prefork(&tdata_locks[i]);
+		for (i = 0; i < PROF_NCTX_LOCKS; i++)
+			malloc_mutex_prefork(&gctx_locks[i]);
+	}
+}
+
+void
+prof_prefork1(void)
+{
+
+	if (opt_prof) {
+		malloc_mutex_prefork(&prof_active_mtx);
+		malloc_mutex_prefork(&prof_dump_seq_mtx);
+		malloc_mutex_prefork(&prof_gdump_mtx);
+		malloc_mutex_prefork(&next_thr_uid_mtx);
+		malloc_mutex_prefork(&prof_thread_active_init_mtx);
 	}
 }
 
@@ -2222,14 +2234,18 @@ prof_postfork_parent(void)
 	if (opt_prof) {
 		unsigned i;
 
-		for (i = 0; i < PROF_NTDATA_LOCKS; i++)
-			malloc_mutex_postfork_parent(&tdata_locks[i]);
+		malloc_mutex_postfork_parent(&prof_thread_active_init_mtx);
+		malloc_mutex_postfork_parent(&next_thr_uid_mtx);
+		malloc_mutex_postfork_parent(&prof_gdump_mtx);
+		malloc_mutex_postfork_parent(&prof_dump_seq_mtx);
+		malloc_mutex_postfork_parent(&prof_active_mtx);
 		for (i = 0; i < PROF_NCTX_LOCKS; i++)
 			malloc_mutex_postfork_parent(&gctx_locks[i]);
-		malloc_mutex_postfork_parent(&prof_dump_seq_mtx);
-		malloc_mutex_postfork_parent(&next_thr_uid_mtx);
-		malloc_mutex_postfork_parent(&bt2gctx_mtx);
+		for (i = 0; i < PROF_NTDATA_LOCKS; i++)
+			malloc_mutex_postfork_parent(&tdata_locks[i]);
 		malloc_mutex_postfork_parent(&tdatas_mtx);
+		malloc_mutex_postfork_parent(&bt2gctx_mtx);
+		malloc_mutex_postfork_parent(&prof_dump_mtx);
 	}
 }
 
@@ -2240,14 +2256,18 @@ prof_postfork_child(void)
 	if (opt_prof) {
 		unsigned i;
 
-		for (i = 0; i < PROF_NTDATA_LOCKS; i++)
-			malloc_mutex_postfork_child(&tdata_locks[i]);
+		malloc_mutex_postfork_child(&prof_thread_active_init_mtx);
+		malloc_mutex_postfork_child(&next_thr_uid_mtx);
+		malloc_mutex_postfork_child(&prof_gdump_mtx);
+		malloc_mutex_postfork_child(&prof_dump_seq_mtx);
+		malloc_mutex_postfork_child(&prof_active_mtx);
 		for (i = 0; i < PROF_NCTX_LOCKS; i++)
 			malloc_mutex_postfork_child(&gctx_locks[i]);
-		malloc_mutex_postfork_child(&prof_dump_seq_mtx);
-		malloc_mutex_postfork_child(&next_thr_uid_mtx);
-		malloc_mutex_postfork_child(&bt2gctx_mtx);
+		for (i = 0; i < PROF_NTDATA_LOCKS; i++)
+			malloc_mutex_postfork_child(&tdata_locks[i]);
 		malloc_mutex_postfork_child(&tdatas_mtx);
+		malloc_mutex_postfork_child(&bt2gctx_mtx);
+		malloc_mutex_postfork_child(&prof_dump_mtx);
 	}
 }
 
diff --git a/test/unit/fork.c b/test/unit/fork.c
new file mode 100644
index 00000000..890bc869
--- /dev/null
+++ b/test/unit/fork.c
@@ -0,0 +1,39 @@
+#include "test/jemalloc_test.h"
+
+#include <sys/wait.h>
+
+TEST_BEGIN(test_fork)
+{
+	void *p;
+	pid_t pid;
+
+	p = malloc(1);
+	assert_ptr_not_null(p, "Unexpected malloc() failure");
+
+	pid = fork();
+	if (pid == -1) {
+		/* Error. */
+		test_fail("Unexpected fork() failure");
+	} else if (pid == 0) {
+		/* Child. */
+		exit(0);
+	} else {
+		int status;
+
+		/* Parent. */
+		free(p);
+		do {
+			if (waitpid(pid, &status, 0) == -1)
+				test_fail("Unexpected waitpid() failure");
+		} while (!WIFEXITED(status) && !WIFSIGNALED(status));
+	}
+}
+TEST_END
+
+int
+main(void)
+{
+
+	return (test(
+	    test_fork));
+}