Add rtree element witnesses.

5 files changed, 90 insertions, 19 deletions

diff --git a/include/jemalloc/internal/mutex.h b/include/jemalloc/internal/mutex.h
index 52217991..b4e01ff8 100644
--- a/include/jemalloc/internal/mutex.h
+++ b/include/jemalloc/internal/mutex.h
@@ -6,21 +6,24 @@ typedef struct malloc_mutex_s malloc_mutex_t;
 #ifdef _WIN32
 #  define MALLOC_MUTEX_INITIALIZER
 #elif (defined(JEMALLOC_OSSPIN))
-#  define MALLOC_MUTEX_INITIALIZER {0, WITNESS_INITIALIZER(WITNESS_RANK_OMIT)}
+#  define MALLOC_MUTEX_INITIALIZER					\
+     {0, WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
 #elif (defined(JEMALLOC_MUTEX_INIT_CB))
 #  define MALLOC_MUTEX_INITIALIZER					\
-    {PTHREAD_MUTEX_INITIALIZER, NULL, WITNESS_INITIALIZER(WITNESS_RANK_OMIT)}
+    {PTHREAD_MUTEX_INITIALIZER, NULL,					\
+     WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
 #else
 #  if (defined(JEMALLOC_HAVE_PTHREAD_MUTEX_ADAPTIVE_NP) &&		\
        defined(PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP))
 #    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_ADAPTIVE_NP
 #    define MALLOC_MUTEX_INITIALIZER					\
        {PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP,				\
-        WITNESS_INITIALIZER(WITNESS_RANK_OMIT)}
+        WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
 #  else
 #    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_DEFAULT
 #    define MALLOC_MUTEX_INITIALIZER					\
-       {PTHREAD_MUTEX_INITIALIZER, WITNESS_INITIALIZER(WITNESS_RANK_OMIT)}
+       {PTHREAD_MUTEX_INITIALIZER,					\
+        WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
 #  endif
 #endif
 
diff --git a/include/jemalloc/internal/private_symbols.txt b/include/jemalloc/internal/private_symbols.txt
index 42c730c6..102f01c0 100644
--- a/include/jemalloc/internal/private_symbols.txt
+++ b/include/jemalloc/internal/private_symbols.txt
@@ -468,6 +468,10 @@ rtree_elm_lookup
 rtree_elm_read
 rtree_elm_read_acquired
 rtree_elm_release
+rtree_elm_witness_access
+rtree_elm_witness_acquire
+rtree_elm_witness_release
+rtree_elm_witnesses_cleanup
 rtree_elm_write
 rtree_elm_write_acquired
 rtree_read
diff --git a/include/jemalloc/internal/rtree.h b/include/jemalloc/internal/rtree.h
index dbea434c..e62ab6b9 100644
--- a/include/jemalloc/internal/rtree.h
+++ b/include/jemalloc/internal/rtree.h
@@ -7,6 +7,8 @@
 #ifdef JEMALLOC_H_TYPES
 
 typedef struct rtree_elm_s rtree_elm_t;
+typedef struct rtree_elm_witness_s rtree_elm_witness_t;
+typedef struct rtree_elm_witness_tsd_s rtree_elm_witness_tsd_t;
 typedef struct rtree_level_s rtree_level_t;
 typedef struct rtree_s rtree_t;
 
@@ -23,6 +25,29 @@ typedef struct rtree_s rtree_t;
 /* Used for two-stage lock-free node initialization. */
 #define	RTREE_NODE_INITIALIZING	((rtree_elm_t *)0x1)
 
+/*
+ * Maximum number of concurrently acquired elements per thread.  This controls
+ * how many witness_t structures are embedded in tsd.  Ideally rtree_elm_t would
+ * have a witness_t directly embedded, but that would dramatically bloat the
+ * tree.  This must contain enough entries to e.g. coalesce two extents.
+ */
+#define	RTREE_ELM_ACQUIRE_MAX	4
+
+/* Initializers for rtree_elm_witness_tsd_t. */
+#define	RTREE_ELM_WITNESS_INITIALIZER {					\
+	NULL,								\
+	WITNESS_INITIALIZER("rtree_elm", WITNESS_RANK_RTREE_ELM)	\
+}
+
+#define	RTREE_ELM_WITNESS_TSD_INITIALIZER {				\
+	{								\
+		RTREE_ELM_WITNESS_INITIALIZER,				\
+		RTREE_ELM_WITNESS_INITIALIZER,				\
+		RTREE_ELM_WITNESS_INITIALIZER,				\
+		RTREE_ELM_WITNESS_INITIALIZER				\
+	}								\
+}
+
 #endif /* JEMALLOC_H_TYPES */
 /******************************************************************************/
 #ifdef JEMALLOC_H_STRUCTS
@@ -35,6 +60,15 @@ struct rtree_elm_s {
 	};
 };
 
+struct rtree_elm_witness_s {
+	const rtree_elm_t	*elm;
+	witness_t		witness;
+};
+
+struct rtree_elm_witness_tsd_s {
+	rtree_elm_witness_t	witnesses[RTREE_ELM_ACQUIRE_MAX];
+};
+
 struct rtree_level_s {
 	/*
 	 * A non-NULL subtree points to a subtree rooted along the hypothetical
@@ -97,6 +131,13 @@ rtree_elm_t	*rtree_subtree_read_hard(tsdn_t *tsdn, rtree_t *rtree,
     unsigned level);
 rtree_elm_t	*rtree_child_read_hard(tsdn_t *tsdn, rtree_t *rtree,
     rtree_elm_t *elm, unsigned level);
+void	rtree_elm_witness_acquire(tsdn_t *tsdn, const rtree_t *rtree,
+    uintptr_t key, const rtree_elm_t *elm);
+void	rtree_elm_witness_access(tsdn_t *tsdn, const rtree_t *rtree,
+    const rtree_elm_t *elm);
+void	rtree_elm_witness_release(tsdn_t *tsdn, const rtree_t *rtree,
+    const rtree_elm_t *elm);
+void	rtree_elm_witnesses_cleanup(tsd_t *tsd);
 
 #endif /* JEMALLOC_H_EXTERNS */
 /******************************************************************************/
@@ -125,9 +166,11 @@ extent_t	*rtree_read(tsdn_t *tsdn, rtree_t *rtree, uintptr_t key,
     bool dependent);
 rtree_elm_t	*rtree_elm_acquire(tsdn_t *tsdn, rtree_t *rtree, uintptr_t key,
     bool dependent, bool init_missing);
-extent_t	*rtree_elm_read_acquired(rtree_elm_t *elm);
-void	rtree_elm_write_acquired(rtree_elm_t *elm, const extent_t *extent);
-void	rtree_elm_release(rtree_elm_t *elm);
+extent_t	*rtree_elm_read_acquired(tsdn_t *tsdn, const rtree_t *rtree,
+    rtree_elm_t *elm);
+void	rtree_elm_write_acquired(tsdn_t *tsdn, const rtree_t *rtree,
+    rtree_elm_t *elm, const extent_t *extent);
+void	rtree_elm_release(tsdn_t *tsdn, const rtree_t *rtree, rtree_elm_t *elm);
 void	rtree_clear(tsdn_t *tsdn, rtree_t *rtree, uintptr_t key);
 #endif
 
@@ -393,11 +436,14 @@ rtree_elm_acquire(tsdn_t *tsdn, rtree_t *rtree, uintptr_t key, bool dependent,
 		} while (atomic_cas_p(&elm->pun, (void *)extent, s));
 	}
 
+	if (config_debug)
+		rtree_elm_witness_acquire(tsdn, rtree, key, elm);
+
 	return (elm);
 }
 
 JEMALLOC_INLINE extent_t *
-rtree_elm_read_acquired(rtree_elm_t *elm)
+rtree_elm_read_acquired(tsdn_t *tsdn, const rtree_t *rtree, rtree_elm_t *elm)
 {
 	extent_t *extent;
 
@@ -405,24 +451,34 @@ rtree_elm_read_acquired(rtree_elm_t *elm)
 	extent = (extent_t *)((uintptr_t)elm->pun & ~((uintptr_t)0x1));
 	assert(((uintptr_t)extent & (uintptr_t)0x1) == (uintptr_t)0x0);
 
+	if (config_debug)
+		rtree_elm_witness_access(tsdn, rtree, elm);
+
 	return (extent);
 }
 
 JEMALLOC_INLINE void
-rtree_elm_write_acquired(rtree_elm_t *elm, const extent_t *extent)
+rtree_elm_write_acquired(tsdn_t *tsdn, const rtree_t *rtree, rtree_elm_t *elm,
+    const extent_t *extent)
 {
 
 	assert(((uintptr_t)extent & (uintptr_t)0x1) == (uintptr_t)0x0);
 	assert(((uintptr_t)elm->pun & (uintptr_t)0x1) == (uintptr_t)0x1);
+
+	if (config_debug)
+		rtree_elm_witness_access(tsdn, rtree, elm);
+
 	elm->pun = (void *)((uintptr_t)extent | (uintptr_t)0x1);
-	assert(rtree_elm_read_acquired(elm) == extent);
+	assert(rtree_elm_read_acquired(tsdn, rtree, elm) == extent);
 }
 
 JEMALLOC_INLINE void
-rtree_elm_release(rtree_elm_t *elm)
+rtree_elm_release(tsdn_t *tsdn, const rtree_t *rtree, rtree_elm_t *elm)
 {
 
-	rtree_elm_write(elm, rtree_elm_read_acquired(elm));
+	rtree_elm_write(elm, rtree_elm_read_acquired(tsdn, rtree, elm));
+	if (config_debug)
+		rtree_elm_witness_release(tsdn, rtree, elm);
 }
 
 JEMALLOC_INLINE void
@@ -431,8 +487,8 @@ rtree_clear(tsdn_t *tsdn, rtree_t *rtree, uintptr_t key)
 	rtree_elm_t *elm;
 
 	elm = rtree_elm_acquire(tsdn, rtree, key, true, false);
-	rtree_elm_write_acquired(elm, NULL);
-	rtree_elm_release(elm);
+	rtree_elm_write_acquired(tsdn, rtree, elm, NULL);
+	rtree_elm_release(tsdn, rtree, elm);
 }
 #endif
 
diff --git a/include/jemalloc/internal/tsd.h b/include/jemalloc/internal/tsd.h
index f4ff8d76..ca8915ea 100644
--- a/include/jemalloc/internal/tsd.h
+++ b/include/jemalloc/internal/tsd.h
@@ -573,6 +573,7 @@ struct tsd_init_head_s {
     O(arenas_tdata_bypass,	bool)					\
     O(tcache_enabled,		tcache_enabled_t)			\
     O(witnesses,		witness_list_t)				\
+    O(rtree_elm_witnesses,	rtree_elm_witness_tsd_t)		\
     O(witness_fork,		bool)					\
 
 #define	TSD_INITIALIZER {						\
@@ -588,6 +589,7 @@ struct tsd_init_head_s {
     false,								\
     tcache_enabled_default,						\
     ql_head_initializer(witnesses),					\
+    RTREE_ELM_WITNESS_TSD_INITIALIZER,					\
     false								\
 }
 
diff --git a/include/jemalloc/internal/witness.h b/include/jemalloc/internal/witness.h
index c68c9694..f15665bc 100644
--- a/include/jemalloc/internal/witness.h
+++ b/include/jemalloc/internal/witness.h
@@ -4,7 +4,8 @@
 typedef struct witness_s witness_t;
 typedef unsigned witness_rank_t;
 typedef ql_head(witness_t) witness_list_t;
-typedef int witness_comp_t (const witness_t *, const witness_t *);
+typedef int witness_comp_t (const witness_t *, void *, const witness_t *,
+    void *);
 
 /*
  * Lock ranks.  Witnesses with rank WITNESS_RANK_OMIT are completely ignored by
@@ -26,7 +27,8 @@ typedef int witness_comp_t (const witness_t *, const witness_t *);
 #define	WITNESS_RANK_ARENA_CHUNKS	9U
 #define	WITNESS_RANK_ARENA_EXTENT_CACHE	10
 
-#define	WITNESS_RANK_BASE		11U
+#define	WITNESS_RANK_RTREE_ELM		11U
+#define	WITNESS_RANK_BASE		12U
 
 #define	WITNESS_RANK_LEAF		0xffffffffU
 #define	WITNESS_RANK_ARENA_BIN		WITNESS_RANK_LEAF
@@ -38,7 +40,7 @@ typedef int witness_comp_t (const witness_t *, const witness_t *);
 #define	WITNESS_RANK_PROF_NEXT_THR_UID	WITNESS_RANK_LEAF
 #define	WITNESS_RANK_PROF_THREAD_ACTIVE_INIT	WITNESS_RANK_LEAF
 
-#define	WITNESS_INITIALIZER(rank) {"initializer", rank, NULL, {NULL, NULL}}
+#define	WITNESS_INITIALIZER(name, rank) {name, rank, NULL, NULL, {NULL, NULL}}
 
 #endif /* JEMALLOC_H_TYPES */
 /******************************************************************************/
@@ -61,6 +63,9 @@ struct witness_s {
 	 */
 	witness_comp_t		*comp;
 
+	/* Opaque data, passed to comp(). */
+	void			*opaque;
+
 	/* Linkage for thread's currently owned locks. */
 	ql_elm(witness_t)	link;
 };
@@ -70,7 +75,7 @@ struct witness_s {
 #ifdef JEMALLOC_H_EXTERNS
 
 void	witness_init(witness_t *witness, const char *name, witness_rank_t rank,
-    witness_comp_t *comp);
+    witness_comp_t *comp, void *opaque);
 #ifdef JEMALLOC_JET
 typedef void (witness_lock_error_t)(const witness_list_t *, const witness_t *);
 extern witness_lock_error_t *witness_lock_error;
@@ -211,7 +216,8 @@ witness_lock(tsdn_t *tsdn, witness_t *witness)
 		/* Not forking, rank order reversal. */
 		witness_lock_error(witnesses, witness);
 	} else if (w->rank == witness->rank && (w->comp == NULL || w->comp !=
-	    witness->comp || w->comp(w, witness) > 0)) {
+	    witness->comp || w->comp(w, w->opaque, witness, witness->opaque) >
+	    0)) {
 		/*
 		 * Missing/incompatible comparison function, or comparison
 		 * function indicates rank order reversal.