Make dss operations lockless.

Rather than protecting dss operations with a mutex, use atomic
operations.  This has negligible impact on synchronization overhead
during typical dss allocation, but is a substantial improvement for
chunk_in_dss() and the newly added chunk_dss_mergeable(), which can be
called multiple times during chunk deallocations.

This change also has the advantage of avoiding tsd in deallocation paths
associated with purging, which resolves potential deadlocks during
thread exit due to attempted tsd resurrection.

This resolves #425.

6 files changed, 118 insertions, 127 deletions

diff --git a/src/arena.c b/src/arena.c
index 90b9d82..7651495 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -3499,7 +3499,7 @@ arena_new(tsdn_t *tsdn, unsigned ind)
 		    (uint64_t)(uintptr_t)arena;
 	}
 
-	arena->dss_prec = chunk_dss_prec_get(tsdn);
+	arena->dss_prec = chunk_dss_prec_get();
 
 	ql_new(&arena->achunks);
 
diff --git a/src/chunk.c b/src/chunk.c
index dff537f..302b98c 100644
--- a/src/chunk.c
+++ b/src/chunk.c
@@ -611,10 +611,10 @@ chunk_dalloc_cache(tsdn_t *tsdn, arena_t *arena, chunk_hooks_t *chunk_hooks,
 }
 
 static bool
-chunk_dalloc_default_impl(tsdn_t *tsdn, void *chunk, size_t size)
+chunk_dalloc_default_impl(void *chunk, size_t size)
 {
 
-	if (!have_dss || !chunk_in_dss(tsdn, chunk))
+	if (!have_dss || !chunk_in_dss(chunk))
 		return (chunk_dalloc_mmap(chunk, size));
 	return (true);
 }
@@ -623,11 +623,8 @@ static bool
 chunk_dalloc_default(void *chunk, size_t size, bool committed,
     unsigned arena_ind)
 {
-	tsdn_t *tsdn;
-
-	tsdn = tsdn_fetch();
 
-	return (chunk_dalloc_default_impl(tsdn, chunk, size));
+	return (chunk_dalloc_default_impl(chunk, size));
 }
 
 void
@@ -645,7 +642,7 @@ chunk_dalloc_wrapper(tsdn_t *tsdn, arena_t *arena, chunk_hooks_t *chunk_hooks,
 	/* Try to deallocate. */
 	if (chunk_hooks->dalloc == chunk_dalloc_default) {
 		/* Call directly to propagate tsdn. */
-		err = chunk_dalloc_default_impl(tsdn, chunk, size);
+		err = chunk_dalloc_default_impl(chunk, size);
 	} else
 		err = chunk_hooks->dalloc(chunk, size, committed, arena->ind);
 
@@ -718,13 +715,12 @@ chunk_split_default(void *chunk, size_t size, size_t size_a, size_t size_b,
 }
 
 static bool
-chunk_merge_default_impl(tsdn_t *tsdn, void *chunk_a, void *chunk_b)
+chunk_merge_default_impl(void *chunk_a, void *chunk_b)
 {
 
 	if (!maps_coalesce)
 		return (true);
-	if (have_dss && chunk_in_dss(tsdn, chunk_a) != chunk_in_dss(tsdn,
-	    chunk_b))
+	if (have_dss && !chunk_dss_mergeable(chunk_a, chunk_b))
 		return (true);
 
 	return (false);
@@ -734,11 +730,8 @@ static bool
 chunk_merge_default(void *chunk_a, size_t size_a, void *chunk_b, size_t size_b,
     bool committed, unsigned arena_ind)
 {
-	tsdn_t *tsdn;
-
-	tsdn = tsdn_fetch();
 
-	return (chunk_merge_default_impl(tsdn, chunk_a, chunk_b));
+	return (chunk_merge_default_impl(chunk_a, chunk_b));
 }
 
 static rtree_node_elm_t *
@@ -782,32 +775,11 @@ chunk_boot(void)
 	chunksize_mask = chunksize - 1;
 	chunk_npages = (chunksize >> LG_PAGE);
 
-	if (have_dss && chunk_dss_boot())
-		return (true);
+	if (have_dss)
+		chunk_dss_boot();
 	if (rtree_new(&chunks_rtree, (unsigned)((ZU(1) << (LG_SIZEOF_PTR+3)) -
 	    opt_lg_chunk), chunks_rtree_node_alloc, NULL))
 		return (true);
 
 	return (false);
 }
-
-void
-chunk_prefork(tsdn_t *tsdn)
-{
-
-	chunk_dss_prefork(tsdn);
-}
-
-void
-chunk_postfork_parent(tsdn_t *tsdn)
-{
-
-	chunk_dss_postfork_parent(tsdn);
-}
-
-void
-chunk_postfork_child(tsdn_t *tsdn)
-{
-
-	chunk_dss_postfork_child(tsdn);
-}
diff --git a/src/chunk_dss.c b/src/chunk_dss.c
index 0b1f82b..85a1354 100644
--- a/src/chunk_dss.c
+++ b/src/chunk_dss.c
@@ -10,20 +10,19 @@ const char	*dss_prec_names[] = {
 	"N/A"
 };
 
-/* Current dss precedence default, used when creating new arenas. */
-static dss_prec_t	dss_prec_default = DSS_PREC_DEFAULT;
-
 /*
- * Protects sbrk() calls.  This avoids malloc races among threads, though it
- * does not protect against races with threads that call sbrk() directly.
+ * Current dss precedence default, used when creating new arenas.  NB: This is
+ * stored as unsigned rather than dss_prec_t because in principle there's no
+ * guarantee that sizeof(dss_prec_t) is the same as sizeof(unsigned), and we use
+ * atomic operations to synchronize the setting.
  */
-static malloc_mutex_t	dss_mtx;
+static unsigned		dss_prec_default = (unsigned)DSS_PREC_DEFAULT;
 
 /* Base address of the DSS. */
 static void		*dss_base;
-/* Current end of the DSS, or ((void *)-1) if the DSS is exhausted. */
-static void		*dss_prev;
-/* Current upper limit on DSS addresses. */
+/* Atomic boolean indicating whether the DSS is exhausted. */
+static unsigned		dss_exhausted;
+/* Atomic current upper limit on DSS addresses. */
 static void		*dss_max;
 
 /******************************************************************************/
@@ -41,30 +40,59 @@ chunk_dss_sbrk(intptr_t increment)
 }
 
 dss_prec_t
-chunk_dss_prec_get(tsdn_t *tsdn)
+chunk_dss_prec_get(void)
 {
 	dss_prec_t ret;
 
 	if (!have_dss)
 		return (dss_prec_disabled);
-	malloc_mutex_lock(tsdn, &dss_mtx);
-	ret = dss_prec_default;
-	malloc_mutex_unlock(tsdn, &dss_mtx);
+	ret = (dss_prec_t)atomic_read_u(&dss_prec_default);
 	return (ret);
 }
 
 bool
-chunk_dss_prec_set(tsdn_t *tsdn, dss_prec_t dss_prec)
+chunk_dss_prec_set(dss_prec_t dss_prec)
 {
 
 	if (!have_dss)
 		return (dss_prec != dss_prec_disabled);
-	malloc_mutex_lock(tsdn, &dss_mtx);
-	dss_prec_default = dss_prec;
-	malloc_mutex_unlock(tsdn, &dss_mtx);
+	atomic_write_u(&dss_prec_default, (unsigned)dss_prec);
 	return (false);
 }
 
+static void *
+chunk_dss_max_update(void *new_addr)
+{
+	void *max_cur;
+	spin_t spinner;
+
+	/*
+	 * Get the current end of the DSS as max_cur and assure that dss_max is
+	 * up to date.
+	 */
+	spin_init(&spinner);
+	while (true) {
+		void *max_prev = atomic_read_p(&dss_max);
+
+		max_cur = chunk_dss_sbrk(0);
+		if ((uintptr_t)max_prev > (uintptr_t)max_cur) {
+			/*
+			 * Another thread optimistically updated dss_max.  Wait
+			 * for it to finish.
+			 */
+			spin_adaptive(&spinner);
+			continue;
+		}
+		if (!atomic_cas_p(&dss_max, max_prev, max_cur))
+			break;
+	}
+	/* Fixed new_addr can only be supported if it is at the edge of DSS. */
+	if (new_addr != NULL && max_cur != new_addr)
+		return (NULL);
+
+	return (max_cur);
+}
+
 void *
 chunk_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
     size_t alignment, bool *zero, bool *commit)
@@ -80,28 +108,20 @@ chunk_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
 	if ((intptr_t)size < 0)
 		return (NULL);
 
-	malloc_mutex_lock(tsdn, &dss_mtx);
-	if (dss_prev != (void *)-1) {
-
+	if (!atomic_read_u(&dss_exhausted)) {
 		/*
 		 * The loop is necessary to recover from races with other
 		 * threads that are using the DSS for something other than
 		 * malloc.
 		 */
-		do {
-			void *ret, *cpad, *dss_next;
+		while (true) {
+			void *ret, *cpad, *max_cur, *dss_next, *dss_prev;
 			size_t gap_size, cpad_size;
 			intptr_t incr;
-			/* Avoid an unnecessary system call. */
-			if (new_addr != NULL && dss_max != new_addr)
-				break;
 
-			/* Get the current end of the DSS. */
-			dss_max = chunk_dss_sbrk(0);
-
-			/* Make sure the earlier condition still holds. */
-			if (new_addr != NULL && dss_max != new_addr)
-				break;
+			max_cur = chunk_dss_max_update(new_addr);
+			if (max_cur == NULL)
+				goto label_oom;
 
 			/*
 			 * Calculate how much padding is necessary to
@@ -120,17 +140,23 @@ chunk_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
 			cpad_size = (uintptr_t)ret - (uintptr_t)cpad;
 			dss_next = (void *)((uintptr_t)ret + size);
 			if ((uintptr_t)ret < (uintptr_t)dss_max ||
-			    (uintptr_t)dss_next < (uintptr_t)dss_max) {
-				/* Wrap-around. */
-				malloc_mutex_unlock(tsdn, &dss_mtx);
-				return (NULL);
-			}
+			    (uintptr_t)dss_next < (uintptr_t)dss_max)
+				goto label_oom; /* Wrap-around. */
 			incr = gap_size + cpad_size + size;
+
+			/*
+			 * Optimistically update dss_max, and roll back below if
+			 * sbrk() fails.  No other thread will try to extend the
+			 * DSS while dss_max is greater than the current DSS
+			 * max reported by sbrk(0).
+			 */
+			if (atomic_cas_p(&dss_max, max_cur, dss_next))
+				continue;
+
+			/* Try to allocate. */
 			dss_prev = chunk_dss_sbrk(incr);
-			if (dss_prev == dss_max) {
+			if (dss_prev == max_cur) {
 				/* Success. */
-				dss_max = dss_next;
-				malloc_mutex_unlock(tsdn, &dss_mtx);
 				if (cpad_size != 0) {
 					chunk_hooks_t chunk_hooks =
 					    CHUNK_HOOKS_INITIALIZER;
@@ -147,68 +173,65 @@ chunk_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
 					*commit = pages_decommit(ret, size);
 				return (ret);
 			}
-		} while (dss_prev != (void *)-1);
-	}
-	malloc_mutex_unlock(tsdn, &dss_mtx);
 
+			/*
+			 * Failure, whether due to OOM or a race with a raw
+			 * sbrk() call from outside the allocator.  Try to roll
+			 * back optimistic dss_max update; if rollback fails,
+			 * it's due to another caller of this function having
+			 * succeeded since this invocation started, in which
+			 * case rollback is not necessary.
+			 */
+			atomic_cas_p(&dss_max, dss_next, max_cur);
+			if (dss_prev == (void *)-1) {
+				/* OOM. */
+				atomic_write_u(&dss_exhausted, (unsigned)true);
+				goto label_oom;
+			}
+		}
+	}
+label_oom:
 	return (NULL);
 }
 
-bool
-chunk_in_dss(tsdn_t *tsdn, void *chunk)
+static bool
+chunk_in_dss_helper(void *chunk, void *max)
 {
-	bool ret;
-
-	cassert(have_dss);
-
-	malloc_mutex_lock(tsdn, &dss_mtx);
-	if ((uintptr_t)chunk >= (uintptr_t)dss_base
-	    && (uintptr_t)chunk < (uintptr_t)dss_max)
-		ret = true;
-	else
-		ret = false;
-	malloc_mutex_unlock(tsdn, &dss_mtx);
 
-	return (ret);
+	return ((uintptr_t)chunk >= (uintptr_t)dss_base && (uintptr_t)chunk <
+	    (uintptr_t)max);
 }
 
 bool
-chunk_dss_boot(void)
+chunk_in_dss(void *chunk)
 {
 
 	cassert(have_dss);
 
-	if (malloc_mutex_init(&dss_mtx, "dss", WITNESS_RANK_DSS))
-		return (true);
-	dss_base = chunk_dss_sbrk(0);
-	dss_prev = dss_base;
-	dss_max = dss_base;
-
-	return (false);
+	return (chunk_in_dss_helper(chunk, atomic_read_p(&dss_max)));
 }
 
-void
-chunk_dss_prefork(tsdn_t *tsdn)
+bool
+chunk_dss_mergeable(void *chunk_a, void *chunk_b)
 {
+	void *max;
 
-	if (have_dss)
-		malloc_mutex_prefork(tsdn, &dss_mtx);
-}
-
-void
-chunk_dss_postfork_parent(tsdn_t *tsdn)
-{
+	cassert(have_dss);
 
-	if (have_dss)
-		malloc_mutex_postfork_parent(tsdn, &dss_mtx);
+	max = atomic_read_p(&dss_max);
+	return (chunk_in_dss_helper(chunk_a, max) ==
+	    chunk_in_dss_helper(chunk_b, max));
 }
 
 void
-chunk_dss_postfork_child(tsdn_t *tsdn)
+chunk_dss_boot(void)
 {
 
-	if (have_dss)
-		malloc_mutex_postfork_child(tsdn, &dss_mtx);
+	cassert(have_dss);
+
+	dss_base = chunk_dss_sbrk(0);
+	dss_exhausted = (unsigned)(dss_base == (void *)-1);
+	dss_max = dss_base;
 }
 
 /******************************************************************************/
diff --git a/src/ctl.c b/src/ctl.c
index dad8008..5d2c8db 100644
--- a/src/ctl.c
+++ b/src/ctl.c
@@ -1685,11 +1685,11 @@ arena_i_dss_ctl(tsd_t *tsd, const size_t *mib, size_t miblen, void *oldp,
 		dss_prec_old = arena_dss_prec_get(tsd_tsdn(tsd), arena);
 	} else {
 		if (dss_prec != dss_prec_limit &&
-		    chunk_dss_prec_set(tsd_tsdn(tsd), dss_prec)) {
+		    chunk_dss_prec_set(dss_prec)) {
 			ret = EFAULT;
 			goto label_return;
 		}
-		dss_prec_old = chunk_dss_prec_get(tsd_tsdn(tsd));
+		dss_prec_old = chunk_dss_prec_get();
 	}
 
 	dss = dss_prec_names[dss_prec_old];
diff --git a/src/huge.c b/src/huge.c
index 3a2877c..19ca3f0 100644
--- a/src/huge.c
+++ b/src/huge.c
@@ -114,7 +114,7 @@ huge_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, size_t alignment,
 #define	huge_dalloc_junk JEMALLOC_N(huge_dalloc_junk_impl)
 #endif
 static void
-huge_dalloc_junk(tsdn_t *tsdn, void *ptr, size_t usize)
+huge_dalloc_junk(void *ptr, size_t usize)
 {
 
 	if (config_fill && have_dss && unlikely(opt_junk_free)) {
@@ -122,7 +122,7 @@ huge_dalloc_junk(tsdn_t *tsdn, void *ptr, size_t usize)
 		 * Only bother junk filling if the chunk isn't about to be
 		 * unmapped.
 		 */
-		if (!config_munmap || (have_dss && chunk_in_dss(tsdn, ptr)))
+		if (!config_munmap || (have_dss && chunk_in_dss(ptr)))
 			memset(ptr, JEMALLOC_FREE_JUNK, usize);
 	}
 }
@@ -221,7 +221,7 @@ huge_ralloc_no_move_shrink(tsdn_t *tsdn, void *ptr, size_t oldsize,
 	if (oldsize > usize) {
 		size_t sdiff = oldsize - usize;
 		if (config_fill && unlikely(opt_junk_free)) {
-			huge_dalloc_junk(tsdn, (void *)((uintptr_t)ptr + usize),
+			huge_dalloc_junk((void *)((uintptr_t)ptr + usize),
 			    sdiff);
 			post_zeroed = false;
 		} else {
@@ -402,7 +402,7 @@ huge_dalloc(tsdn_t *tsdn, void *ptr)
 	ql_remove(&arena->huge, node, ql_link);
 	malloc_mutex_unlock(tsdn, &arena->huge_mtx);
 
-	huge_dalloc_junk(tsdn, extent_node_addr_get(node),
+	huge_dalloc_junk(extent_node_addr_get(node),
 	    extent_node_size_get(node));
 	arena_chunk_dalloc_huge(tsdn, extent_node_arena_get(node),
 	    extent_node_addr_get(node), extent_node_size_get(node));
diff --git a/src/jemalloc.c b/src/jemalloc.c
index aec2a5e..b0ebf81 100644
--- a/src/jemalloc.c
+++ b/src/jemalloc.c
@@ -1115,8 +1115,7 @@ malloc_conf_init(void)
 				for (i = 0; i < dss_prec_limit; i++) {
 					if (strncmp(dss_prec_names[i], v, vlen)
 					    == 0) {
-						if (chunk_dss_prec_set(NULL,
-						   i)) {
+						if (chunk_dss_prec_set(i)) {
 							malloc_conf_error(
 							    "Error setting dss",
 							    k, klen, v, vlen);
@@ -2783,7 +2782,6 @@ _malloc_prefork(void)
 		}
 	}
 	base_prefork(tsd_tsdn(tsd));
-	chunk_prefork(tsd_tsdn(tsd));
 	for (i = 0; i < narenas; i++) {
 		if ((arena = arena_get(tsd_tsdn(tsd), i, false)) != NULL)
 			arena_prefork3(tsd_tsdn(tsd), arena);
@@ -2812,7 +2810,6 @@ _malloc_postfork(void)
 
 	witness_postfork_parent(tsd);
 	/* Release all mutexes, now that fork() has completed. */
-	chunk_postfork_parent(tsd_tsdn(tsd));
 	base_postfork_parent(tsd_tsdn(tsd));
 	for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
 		arena_t *arena;
@@ -2837,7 +2834,6 @@ jemalloc_postfork_child(void)
 
 	witness_postfork_child(tsd);
 	/* Release all mutexes, now that fork() has completed. */
-	chunk_postfork_child(tsd_tsdn(tsd));
 	base_postfork_child(tsd_tsdn(tsd));
 	for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
 		arena_t *arena;