Relax witness assertions related to prof_gdump().

In some cases the prof machinery allocates (in order to modify the
bt2gctx hash table), and such operations are synchronized via
bt2gctx_mtx.  Rather than asserting that no locks are held on entry
into functions that may call prof_gdump(), make the weaker assertion
that no "core" locks are held.  The prof machinery enqueues dumps
triggered by prof_gdump() calls when bt2gctx_mtx is held, so this
weakened assertion avoids false failures in such cases.

include/jemalloc/internal/witness.h

@@ -25,6 +25,14 @@ typedef int witness_comp_t (const witness_t *, const witness_t *);
 #define	WITNESS_RANK_PROF_TDATA		7U
 #define	WITNESS_RANK_PROF_GCTX		8U
 
+/*
+ * Used as an argument to witness_assert_depth_to_rank() in order to validate
+ * depth excluding non-core locks with lower ranks.  Since the rank argument to
+ * witness_assert_depth_to_rank() is inclusive rather than exclusive, this
+ * definition can have the same value as the minimally ranked core lock.
+ */
+#define WITNESS_RANK_CORE		9U
+
 #define	WITNESS_RANK_ARENA		9U
 #define	WITNESS_RANK_ARENA_CHUNKS	10U
 #define	WITNESS_RANK_ARENA_NODE_CACHE	11U

src/arena.c

@@ -591,7 +591,8 @@ arena_chunk_alloc_internal_hard(tsdn_t *tsdn, arena_t *arena,
 	size_t sn;
 
 	malloc_mutex_unlock(tsdn, &arena->lock);
-	witness_assert_lockless(tsdn); /* prof_gdump() requirement. */
+	/* prof_gdump() requirement. */
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 0);
 
 	chunk = (arena_chunk_t *)chunk_alloc_wrapper(tsdn, arena, chunk_hooks,
 	    NULL, chunksize, chunksize, &sn, zero, commit);
@@ -633,7 +634,7 @@ arena_chunk_alloc_internal(tsdn_t *tsdn, arena_t *arena, bool *zero,
 	size_t sn;
 
 	/* prof_gdump() requirement. */
-	witness_assert_depth(tsdn, 1);
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 1);
 	malloc_mutex_assert_owner(tsdn, &arena->lock);
 
 	chunk = chunk_alloc_cache(tsdn, arena, &chunk_hooks, NULL, chunksize,

src/huge.c

@@ -62,7 +62,8 @@ huge_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, size_t alignment,
 	/* Allocate one or more contiguous chunks for this request. */
 
 	assert(!tsdn_null(tsdn) || arena != NULL);
-	witness_assert_lockless(tsdn); /* prof_gdump() requirement. */
+	/* prof_gdump() requirement. */
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 0);
 
 	ausize = sa2u(usize, alignment);
 	if (unlikely(ausize == 0 || ausize > HUGE_MAXCLASS))
@@ -149,7 +150,8 @@ huge_ralloc_no_move_similar(tsdn_t *tsdn, void *ptr, size_t oldsize,
 	chunk_hooks_t chunk_hooks = CHUNK_HOOKS_INITIALIZER;
 	bool pre_zeroed, post_zeroed, gdump;
 
-	witness_assert_lockless(tsdn); /* prof_gdump() requirement. */
+	/* prof_gdump() requirement. */
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 0);
 
 	/* Increase usize to incorporate extra. */
 	for (usize = usize_min; usize < usize_max && (usize_next = s2u(usize+1))
@@ -223,7 +225,8 @@ huge_ralloc_no_move_shrink(tsdn_t *tsdn, void *ptr, size_t oldsize,
 	chunk_hooks = chunk_hooks_get(tsdn, arena);
 
 	assert(oldsize > usize);
-	witness_assert_lockless(tsdn); /* prof_gdump() requirement. */
+	/* prof_gdump() requirement. */
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 0);
 
 	/* Split excess chunks. */
 	cdiff = CHUNK_CEILING(oldsize) - CHUNK_CEILING(usize);
@@ -278,7 +281,8 @@ huge_ralloc_no_move_expand(tsdn_t *tsdn, void *ptr, size_t oldsize,
 	is_zeroed_subchunk = extent_node_zeroed_get(node);
 	malloc_mutex_unlock(tsdn, &arena->huge_mtx);
 
-	witness_assert_lockless(tsdn); /* prof_gdump() requirement. */
+	/* prof_gdump() requirement. */
+	witness_assert_depth_to_rank(tsdn, WITNESS_RANK_CORE, 0);
 
 	/*
 	 * Use is_zeroed_chunk to detect whether the trailing memory is zeroed,