Switch to fine-grained reentrancy support.

Previously we had a general detection and support of reentrancy, at the cost of
having branches and inc / dec operations on fast paths.  To avoid taxing fast
paths, we move the reentrancy operations onto tsd slow state, and only modify
reentrancy level around external calls (that might trigger reentrancy).

include/jemalloc/internal/jemalloc_internal_inlines_a.h

@@ -33,6 +33,8 @@ tcache_t *tcache_get(tsd_t *tsd);
 malloc_cpuid_t malloc_getcpu(void);
 unsigned percpu_arena_choose(void);
 unsigned percpu_arena_ind_limit(void);
+void pre_reentrancy(tsd_t *tsd);
+void post_reentrancy(tsd_t *tsd);
 #endif
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_C_))
@@ -445,6 +447,27 @@ tcache_get(tsd_t *tsd) {
 
 	return tsd_tcachep_get(tsd);
 }
+
+JEMALLOC_INLINE void
+pre_reentrancy(tsd_t *tsd) {
+	bool fast = tsd_fast(tsd);
+	++*tsd_reentrancy_levelp_get(tsd);
+	if (fast) {
+		/* Prepare slow path for reentrancy. */
+		tsd_slow_update(tsd);
+		assert(tsd->state == tsd_state_nominal_slow);
+	}
+}
+
+JEMALLOC_INLINE void
+post_reentrancy(tsd_t *tsd) {
+	int8_t *reentrancy_level = tsd_reentrancy_levelp_get(tsd);
+	assert(*reentrancy_level > 0);
+	if (--*reentrancy_level == 0) {
+		tsd_slow_update(tsd);
+	}
+}
+
 #endif
 
 #endif /* JEMALLOC_INTERNAL_INLINES_A_H */

include/jemalloc/internal/jemalloc_internal_inlines_b.h

@@ -16,7 +16,7 @@ arena_choose_impl(tsd_t *tsd, arena_t *arena, bool internal) {
 	}
 
 	/* During reentrancy, arena 0 is the safest bet. */
-	if (*tsd_reentrancy_levelp_get(tsd) > 1) {
+	if (unlikely(tsd_reentrancy_level_get(tsd) > 0)) {
 		return arena_get(tsd_tsdn(tsd), 0, true);
 	}
 

include/jemalloc/internal/jemalloc_internal_inlines_c.h

@@ -117,8 +117,8 @@ idalloctm(tsdn_t *tsdn, void *ptr, tcache_t *tcache, alloc_ctx_t *alloc_ctx,
 	if (config_stats && is_internal) {
 		arena_internal_sub(iaalloc(tsdn, ptr), isalloc(tsdn, ptr));
 	}
-	if (!is_internal && *tsd_reentrancy_levelp_get(tsdn_tsd(tsdn)) != 0) {
+	if (!is_internal && tsd_reentrancy_level_get(tsdn_tsd(tsdn)) != 0) {
-		tcache = NULL;
+		assert(tcache == NULL);
 	}
 	arena_dalloc(tsdn, ptr, tcache, alloc_ctx, slow_path);
 }

include/jemalloc/internal/tsd_inlines.h

@@ -20,7 +20,7 @@ tsd_t *tsdn_tsd(tsdn_t *tsdn);
 rtree_ctx_t *tsd_rtree_ctx(tsd_t *tsd);
 rtree_ctx_t *tsdn_rtree_ctx(tsdn_t *tsdn, rtree_ctx_t *fallback);
 bool tsd_fast(tsd_t *tsd);
-void tsd_assert_fast(tsd_t *tsd);
+bool tsd_assert_fast(tsd_t *tsd);
 #endif
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_TSD_C_))
@@ -52,9 +52,11 @@ MALLOC_TSD
 #undef MALLOC_TSD_getset_no
 #undef O
 
-JEMALLOC_ALWAYS_INLINE void
+JEMALLOC_ALWAYS_INLINE bool
 tsd_assert_fast(tsd_t *tsd) {
-	assert(!malloc_slow && tsd_tcache_enabled_get(tsd));
+	assert(!malloc_slow && tsd_tcache_enabled_get(tsd) &&
+	    tsd_reentrancy_level_get(tsd) == 0);
+	return true;
 }
 
 JEMALLOC_ALWAYS_INLINE bool

include/jemalloc/internal/tsd_structs.h

@@ -55,7 +55,7 @@ struct tsd_init_head_s {
 /*  O(name,			type,		[gs]et,	init,	cleanup) */ \
     O(tcache_enabled,		bool,		yes,	yes,	no)	\
     O(arenas_tdata_bypass,	bool,		no,	no,	no)	\
-    O(reentrancy_level,		int8_t,		no,	no,	no)	\
+    O(reentrancy_level,		int8_t,		yes,	no,	no)	\
     O(narenas_tdata,		uint32_t,	yes,	no,	no)	\
     O(thread_allocated,		uint64_t,	yes,	no,	no)	\
     O(thread_deallocated,	uint64_t,	yes,	no,	no)	\

src/arena.c

@@ -1966,11 +1966,9 @@ arena_new(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks) {
 		 * If we're here, then arena 0 already exists, so bootstrapping
 		 * is done enough that we should have tsd.
 		 */
-		int8_t *reentrancy_level = tsd_reentrancy_levelp_get(tsdn_tsd(
+		pre_reentrancy(tsdn_tsd(tsdn));
-		    tsdn));
-		++*reentrancy_level;
 		hooks_arena_new_hook();
-		--*reentrancy_level;
+		post_reentrancy(tsdn_tsd(tsdn));
 	}
 
 	return arena;

src/jemalloc.c

@@ -1663,13 +1663,8 @@ imalloc_body(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd) {
 	szind_t ind = 0;
 	size_t usize = 0;
 
-	/*
+	/* Reentrancy is only checked on slow path. */
-	 * For reentrancy checking, we get the old reentrancy level from tsd and
+	int8_t reentrancy_level;
-	 * reset it once we're done.  In case of early bailout though, we never
-	 * bother getting the old level, so we shouldn't try to reset it.  This
-	 * is indicated by leaving the pointer as NULL.
-	 */
-	int8_t *reentrancy_level = NULL;
 
 	/* Compute the amount of memory the user wants. */
 	if (unlikely(compute_size_with_overflow(sopts->may_overflow, dopts,
@@ -1716,12 +1711,11 @@ imalloc_body(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd) {
 	 * If we need to handle reentrancy, we can do it out of a
 	 * known-initialized arena (i.e. arena 0).
 	 */
-	reentrancy_level = tsd_reentrancy_levelp_get(tsd);
+	reentrancy_level = tsd_reentrancy_level_get(tsd);
-	++*reentrancy_level;
+	if (reentrancy_level == 0) {
-	if (*reentrancy_level == 1) {
 		witness_assert_lockless(tsd_tsdn(tsd));
 	}
-	if (unlikely(*reentrancy_level > 1)) {
+	if (sopts->slow && unlikely(reentrancy_level > 0)) {
 		/*
 		 * We should never specify particular arenas or tcaches from
 		 * within our internal allocations.
@@ -1795,14 +1789,9 @@ imalloc_body(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd) {
 	}
 
 	/* Success! */
-	if (*reentrancy_level == 1) {
+	if (reentrancy_level == 0) {
 		witness_assert_lockless(tsd_tsdn(tsd));
 	}
-	/*
-	 * If we got here, we never bailed out on a failure path, so
-	 * reentrancy_level is non-null.
-	 */
-	--*reentrancy_level;
 	*dopts->result = allocation;
 	return 0;
 
@@ -1826,10 +1815,6 @@ label_oom:
 		*dopts->result = NULL;
 	}
 
-	if (reentrancy_level != NULL) {
-		--*reentrancy_level;
-	}
-
 	return ENOMEM;
 
 	/*
@@ -1857,10 +1842,6 @@ label_invalid_alignment:
 		*dopts->result = NULL;
 	}
 
-	if (reentrancy_level != NULL) {
-		--*reentrancy_level;
-	}
-
 	return EINVAL;
 }
 
@@ -2053,8 +2034,11 @@ irealloc_prof(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t usize,
 
 JEMALLOC_ALWAYS_INLINE_C void
 ifree(tsd_t *tsd, void *ptr, tcache_t *tcache, bool slow_path) {
-	if (*tsd_reentrancy_levelp_get(tsd) == 0) {
+	assert(slow_path || tsd_assert_fast(tsd));
+	if (tsd_reentrancy_level_get(tsd) == 0) {
 		witness_assert_lockless(tsd_tsdn(tsd));
+	} else {
+		assert(slow_path);
 	}
 
 	assert(ptr != NULL);
@@ -2088,8 +2072,11 @@ ifree(tsd_t *tsd, void *ptr, tcache_t *tcache, bool slow_path) {
 
 JEMALLOC_ALWAYS_INLINE_C void
 isfree(tsd_t *tsd, void *ptr, size_t usize, tcache_t *tcache, bool slow_path) {
-	if (*tsd_reentrancy_levelp_get(tsd) == 0) {
+	assert(slow_path || tsd_assert_fast(tsd));
+	if (tsd_reentrancy_level_get(tsd) == 0) {
 		witness_assert_lockless(tsd_tsdn(tsd));
+	} else {
+		assert(slow_path);
 	}
 
 	assert(ptr != NULL);
@@ -2129,14 +2116,14 @@ je_realloc(void *ptr, size_t size) {
 
 	if (unlikely(size == 0)) {
 		if (ptr != NULL) {
-			tsd_t *tsd;
-
 			/* realloc(ptr, 0) is equivalent to free(ptr). */
 			UTRACE(ptr, 0, 0);
-			tsd = tsd_fetch();
+			tcache_t *tcache;
-			tcache_t *tcache = NULL;
+			tsd_t *tsd = tsd_fetch();
-			if (likely(*tsd_reentrancy_levelp_get(tsd) == 0)) {
+			if (tsd_reentrancy_level_get(tsd) == 0) {
 				tcache = tcache_get(tsd);
+			} else {
+				tcache = NULL;
 			}
 			ifree(tsd, ptr, tcache, true);
 			return NULL;
@@ -2200,29 +2187,25 @@ je_free(void *ptr) {
 	UTRACE(ptr, 0, 0);
 	if (likely(ptr != NULL)) {
 		tsd_t *tsd = tsd_fetch();
-		if (*tsd_reentrancy_levelp_get(tsd) == 0) {
+		if (tsd_reentrancy_level_get(tsd) == 0) {
 			witness_assert_lockless(tsd_tsdn(tsd));
 		}
+
 		tcache_t *tcache;
 		if (likely(tsd_fast(tsd))) {
 			tsd_assert_fast(tsd);
-			if (likely(*tsd_reentrancy_levelp_get(tsd) == 0)) {
+			/* Unconditionally get tcache ptr on fast path. */
-				/* Getting tcache ptr unconditionally. */
 			tcache = tsd_tcachep_get(tsd);
-				assert(tcache == tcache_get(tsd));
-			} else {
-				tcache = NULL;
-			}
 			ifree(tsd, ptr, tcache, false);
 		} else {
-			if (likely(*tsd_reentrancy_levelp_get(tsd) == 0)) {
+			if (likely(tsd_reentrancy_level_get(tsd) == 0)) {
 				tcache = tcache_get(tsd);
 			} else {
 				tcache = NULL;
 			}
 			ifree(tsd, ptr, tcache, true);
 		}
-		if (*tsd_reentrancy_levelp_get(tsd) == 0) {
+		if (tsd_reentrancy_level_get(tsd) == 0) {
 			witness_assert_lockless(tsd_tsdn(tsd));
 		}
 	}
@@ -2707,35 +2690,34 @@ je_sallocx(const void *ptr, int flags) {
 
 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
 je_dallocx(void *ptr, int flags) {
-	tsd_t *tsd;
-	tcache_t *tcache;
-
 	assert(ptr != NULL);
 	assert(malloc_initialized() || IS_INITIALIZER);
 
-	tsd = tsd_fetch();
+	tsd_t *tsd = tsd_fetch();
 	bool fast = tsd_fast(tsd);
 	witness_assert_lockless(tsd_tsdn(tsd));
+
+	tcache_t *tcache;
 	if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) {
 		/* Not allowed to be reentrant and specify a custom tcache. */
-		assert(*tsd_reentrancy_levelp_get(tsd) == 0);
+		assert(tsd_reentrancy_level_get(tsd) == 0);
 		if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE) {
 			tcache = NULL;
 		} else {
 			tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
 		}
 	} else {
-		if (likely(*tsd_reentrancy_levelp_get(tsd) == 0)) {
 		if (likely(fast)) {
 			tcache = tsd_tcachep_get(tsd);
 			assert(tcache == tcache_get(tsd));
 		} else {
+			if (likely(tsd_reentrancy_level_get(tsd) == 0)) {
 				tcache = tcache_get(tsd);
-			}
 			}  else {
 				tcache = NULL;
 			}
 		}
+	}
 
 	UTRACE(ptr, 0, 0);
 	if (likely(fast)) {
@@ -2749,10 +2731,9 @@ je_dallocx(void *ptr, int flags) {
 
 JEMALLOC_ALWAYS_INLINE_C size_t
 inallocx(tsdn_t *tsdn, size_t size, int flags) {
-	size_t usize;
-
 	witness_assert_lockless(tsdn);
 
+	size_t usize;
 	if (likely((flags & MALLOCX_LG_ALIGN_MASK) == 0)) {
 		usize = s2u(size);
 	} else {
@@ -2764,38 +2745,36 @@ inallocx(tsdn_t *tsdn, size_t size, int flags) {
 
 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
 je_sdallocx(void *ptr, size_t size, int flags) {
-	tsd_t *tsd;
-	size_t usize;
-	tcache_t *tcache;
-
 	assert(ptr != NULL);
 	assert(malloc_initialized() || IS_INITIALIZER);
-	tsd = tsd_fetch();
-	bool fast = tsd_fast(tsd);
-	usize = inallocx(tsd_tsdn(tsd), size, flags);
-	assert(usize == isalloc(tsd_tsdn(tsd), ptr));
 
+	tsd_t *tsd = tsd_fetch();
+	bool fast = tsd_fast(tsd);
+	size_t usize = inallocx(tsd_tsdn(tsd), size, flags);
+	assert(usize == isalloc(tsd_tsdn(tsd), ptr));
 	witness_assert_lockless(tsd_tsdn(tsd));
+
+	tcache_t *tcache;
 	if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) {
 		/* Not allowed to be reentrant and specify a custom tcache. */
-		assert(*tsd_reentrancy_levelp_get(tsd) == 0);
+		assert(tsd_reentrancy_level_get(tsd) == 0);
 		if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE) {
 			tcache = NULL;
 		} else {
 			tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
 		}
 	} else {
-		if (likely(*tsd_reentrancy_levelp_get(tsd) == 0)) {
 		if (likely(fast)) {
 			tcache = tsd_tcachep_get(tsd);
 			assert(tcache == tcache_get(tsd));
 		} else {
+			if (likely(tsd_reentrancy_level_get(tsd) == 0)) {
 				tcache = tcache_get(tsd);
-			}
 			} else {
 				tcache = NULL;
 			}
 		}
+	}
 
 	UTRACE(ptr, 0, 0);
 	if (likely(fast)) {

src/tsd.c

@@ -15,7 +15,8 @@ malloc_tsd_data(, , tsd_t, TSD_INITIALIZER)
 void
 tsd_slow_update(tsd_t *tsd) {
 	if (tsd_nominal(tsd)) {
-		if (malloc_slow || !tsd->tcache_enabled) {
+		if (malloc_slow || !tsd->tcache_enabled ||
+		    tsd_reentrancy_level_get(tsd) > 0) {
 			tsd->state = tsd_state_nominal_slow;
 		} else {
 			tsd->state = tsd_state_nominal;
@@ -28,6 +29,7 @@ tsd_fetch_slow(tsd_t *tsd) {
 	if (tsd->state == tsd_state_nominal_slow) {
 		/* On slow path but no work needed. */
 		assert(malloc_slow || !tsd_tcache_enabled_get(tsd) ||
+		    tsd_reentrancy_level_get(tsd) > 0 ||
 		    *tsd_arenas_tdata_bypassp_get(tsd));
 	} else if (tsd->state == tsd_state_uninitialized) {
 		tsd->state = tsd_state_nominal;