Implement interval-based heap profile dumping.

Add mallctl interfaces for profiling parameters.

Fix a file descriptor leak in heap profile dumping.

10 files changed, 292 insertions, 76 deletions

diff --git a/jemalloc/doc/jemalloc.3.in b/jemalloc/doc/jemalloc.3.in
index 3865ac9..4266382 100644
--- a/jemalloc/doc/jemalloc.3.in
+++ b/jemalloc/doc/jemalloc.3.in
@@ -388,12 +388,11 @@ will disable dirty page purging.
 @roff_tcache@Note that one cache slot per size class is not a valid
 @roff_tcache@configuration due to implementation details.
 @roff_prof@.It I
-@roff_prof@Double/halve the maximum interval between memory profile dumps, as
+@roff_prof@Double/halve the average interval between memory profile dumps, as
 @roff_prof@measured in bytes of allocation activity.
-@roff_prof@On average, profiles are written four times as often as the maximum
-@roff_prof@interval requires.
-@roff_prof@This is an artifact of the concurrent algorithm that is used to
-@roff_prof@track allocation activity.
+@roff_prof@The actual interval between dumps may be sporadic because
+@roff_prof@decentralized allocation counters are used to avoid synchronization
+@roff_prof@bottlenecks.
 @roff_prof@Profiles are dumped to files named according to the pattern
 @roff_prof@.Pa <prefix>.<pid>.<seq>.i<iseq>.heap ,
 @roff_prof@where
@@ -401,7 +400,7 @@ will disable dirty page purging.
 @roff_prof@is controlled by the
 @roff_prof@JEMALLOC_PROF_PREFIX
 @roff_prof@environment variable.
-@roff_prof@The default maximum interval is 4 GiB.
+@roff_prof@The default average interval is 1 GiB.
 @roff_fill@.It J
 @roff_fill@Each byte of new memory allocated by
 @roff_fill@.Fn @jemalloc_prefix@malloc
@@ -693,6 +692,21 @@ This is useful for detecting whether another thread caused a refresh.
 --enable-lazy-lock was specified during build configuration.
 .Ed
 .\"-----------------------------------------------------------------------------
+.It Sy "config.prof (bool) r-"
+.Bd -ragged -offset indent -compact
+--enable-prof was specified during build configuration.
+.Ed
+.\"-----------------------------------------------------------------------------
+.It Sy "config.prof_libgcc (bool) r-"
+.Bd -ragged -offset indent -compact
+--disable-prof-libgcc was not specified during build configuration.
+.Ed
+.\"-----------------------------------------------------------------------------
+.It Sy "config.prof_libunwind (bool) r-"
+.Bd -ragged -offset indent -compact
+--enable-prof-libunwind was specified during build configuration.
+.Ed
+.\"-----------------------------------------------------------------------------
 .It Sy "config.stats (bool) r-"
 .Bd -ragged -offset indent -compact
 --enable-stats was specified during build configuration.
@@ -782,6 +796,41 @@ See the
 option.
 .Ed
 .\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "opt.prof (bool) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@See the
+@roff_prof@.Dq F
+@roff_prof@option.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "opt.lg_prof_bt_max (size_t) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@See the
+@roff_prof@.Dq B
+@roff_prof@option.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "opt.lg_prof_interval (size_t) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@See the
+@roff_prof@.Dq I
+@roff_prof@option.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "opt.prof_udump (bool) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@See the
+@roff_prof@.Dq U
+@roff_prof@option.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "opt.prof_leak (bool) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@See the
+@roff_prof@.Dq L
+@roff_prof@option.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
 @roff_trace@.It Sy "opt.trace (bool) r-"
 @roff_trace@.Bd -ragged -offset indent -compact
 @roff_trace@See the
@@ -993,6 +1042,15 @@ Maximum size supported by this large size class.
 @roff_prof@environment variable.
 @roff_prof@.Ed
 .\"-----------------------------------------------------------------------------
+@roff_prof@.It Sy "prof.interval (uint64_t) r-"
+@roff_prof@.Bd -ragged -offset indent -compact
+@roff_prof@Average number of bytes allocated between inverval-based profile
+@roff_prof@dumps.
+@roff_prof@See the
+@roff_prof@.Dq I
+@roff_prof@option for additional information.
+@roff_prof@.Ed
+.\"-----------------------------------------------------------------------------
 @roff_stats@.It Sy "stats.allocated (size_t) r-"
 @roff_stats@.Bd -ragged -offset indent -compact
 @roff_stats@Total number of bytes allocated by the application.
diff --git a/jemalloc/src/internal/jemalloc_arena.h b/jemalloc/src/internal/jemalloc_arena.h
index 65054f8..cba954f 100644
--- a/jemalloc/src/internal/jemalloc_arena.h
+++ b/jemalloc/src/internal/jemalloc_arena.h
@@ -290,6 +290,10 @@ struct arena_s {
 	int			trace_fd;
 #endif
 
+#ifdef JEMALLOC_PROF
+	uint64_t		prof_accumbytes;
+#endif
+
 	/* Tree of dirty-page-containing chunks this arena manages. */
 	arena_chunk_tree_t	chunks_dirty;
 
@@ -411,7 +415,14 @@ extern size_t		mspace_mask;
 #define			nlclasses	((chunksize - PAGE_SIZE) >> PAGE_SHIFT)
 
 #ifdef JEMALLOC_TCACHE
-void	arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind);
+void	arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind
+#  ifdef JEMALLOC_PROF
+    , uint64_t prof_accumbytes
+#  endif
+    );
+#endif
+#ifdef JEMALLOC_PROF
+void	arena_prof_accum(arena_t *arena, uint64_t accumbytes);
 #endif
 void	*arena_malloc_small(arena_t *arena, size_t size, bool zero);
 void	*arena_malloc_medium(arena_t *arena, size_t size, bool zero);
diff --git a/jemalloc/src/internal/jemalloc_tcache.h b/jemalloc/src/internal/jemalloc_tcache.h
index 601320b..b499f52 100644
--- a/jemalloc/src/internal/jemalloc_tcache.h
+++ b/jemalloc/src/internal/jemalloc_tcache.h
@@ -36,6 +36,9 @@ struct tcache_s {
 #  ifdef JEMALLOC_STATS
 	ql_elm(tcache_t) link;		/* Used for aggregating stats. */
 #  endif
+#  ifdef JEMALLOC_PROF
+	uint64_t	prof_accumbytes;/* Cleared after arena_prof_accum() */
+#  endif
 	arena_t		*arena;		/* This thread's arena. */
 	unsigned	ev_cnt;		/* Event count since incremental GC. */
 	unsigned	next_gc_bin;	/* Next bin to GC. */
@@ -62,7 +65,11 @@ extern size_t			tcache_nslots;
 /* Number of tcache allocation/deallocation events between incremental GCs. */
 extern unsigned			tcache_gc_incr;
 
-void	tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem);
+void	tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem
+#ifdef JEMALLOC_PROF
+    , tcache_t *tcache
+#endif
+    );
 tcache_t *tcache_create(arena_t *arena);
 void	tcache_bin_destroy(tcache_t *tcache, tcache_bin_t *tbin,
     unsigned binind);
@@ -138,7 +145,11 @@ tcache_event(tcache_t *tcache)
 					 */
 					tcache_bin_flush(tbin, binind,
 					    tbin->ncached - (tbin->low_water >>
-					    1) - (tbin->low_water & 1));
+					    1) - (tbin->low_water & 1)
+#ifdef JEMALLOC_PROF
+					    , tcache
+#endif
+					    );
 				}
 				tbin->low_water = tbin->ncached;
 				tbin->high_water = tbin->ncached;
@@ -206,6 +217,9 @@ tcache_alloc(tcache_t *tcache, size_t size, bool zero)
 #ifdef JEMALLOC_STATS
 	tbin->tstats.nrequests++;
 #endif
+#ifdef JEMALLOC_PROF
+	tcache->prof_accumbytes += tcache->arena->bins[binind].reg_size;
+#endif
 	tcache_event(tcache);
 	return (ret);
 }
@@ -252,7 +266,11 @@ tcache_dalloc(tcache_t *tcache, void *ptr)
 	}
 
 	if (tbin->ncached == tcache_nslots)
-		tcache_bin_flush(tbin, binind, (tcache_nslots >> 1));
+		tcache_bin_flush(tbin, binind, (tcache_nslots >> 1)
+#ifdef JEMALLOC_PROF
+		    , tcache
+#endif
+		    );
 	assert(tbin->ncached < tcache_nslots);
 	tbin->slots[tbin->ncached] = ptr;
 	tbin->ncached++;
diff --git a/jemalloc/src/internal/prof.h b/jemalloc/src/internal/prof.h
index 1d56207..44e11cb 100644
--- a/jemalloc/src/internal/prof.h
+++ b/jemalloc/src/internal/prof.h
@@ -8,7 +8,7 @@ typedef struct prof_thr_cnt_s prof_thr_cnt_t;
 typedef struct prof_ctx_s prof_ctx_t;
 typedef struct prof_s prof_t;
 
-#define	LG_PROF_INTERVAL_DEFAULT	32
+#define	LG_PROF_INTERVAL_DEFAULT	30
 
 /*
  * Hard limit on stack backtrace depth.  Note that the version of
@@ -121,6 +121,15 @@ extern size_t	opt_lg_prof_interval;
 extern bool	opt_prof_udump; /* High-water memory dumping. */
 extern bool	opt_prof_leak; /* Dump leak summary at exit. */
 
+/*
+ * Profile dump interval, measured in bytes allocated.  Each arena triggers a
+ * profile dump when it reaches this threshold.  The effect is that the
+ * interval between profile dumps averages prof_interval, though the actual
+ * interval between dumps will tend to be sporadic, and the interval will be a
+ * maximum of approximately (prof_interval * narenas).
+ */
+extern uint64_t	prof_interval;
+
 bool	prof_init(prof_t *prof, bool master);
 void	prof_destroy(prof_t *prof);
 
@@ -135,7 +144,6 @@ void	prof_mdump(void);
 void	prof_udump(void);
 void	prof_boot0(void);
 bool	prof_boot1(void);
-void	prof_boot2(void);
 
 #endif /* JEMALLOC_H_EXTERNS */
 /******************************************************************************/
diff --git a/jemalloc/src/jemalloc.c b/jemalloc/src/jemalloc.c
index 6a3f53e..813f50b 100644
--- a/jemalloc/src/jemalloc.c
+++ b/jemalloc/src/jemalloc.c
@@ -18,9 +18,9 @@
  *
  * Allocation requests are rounded up to the nearest size class, and no record
  * of the original request size is maintained.  Allocations are broken into
- * categories according to size class.  Assuming runtime defaults, 4 KiB pages
- * and a 16 byte quantum on a 32-bit system, the size classes in each category
- * are as follows:
+ * categories according to size class.  Assuming 1 MiB chunks, 4 KiB pages and
+ * a 16 byte quantum on a 32-bit system, the size classes in each category are
+ * as follows:
  *
  *   |========================================|
  *   | Category | Subcategory      |     Size |
@@ -822,10 +822,6 @@ MALLOC_OUT:
 	next_arena = 0;
 #endif
 
-#ifdef JEMALLOC_PROF
-	prof_boot2();
-#endif
-
 	/* Allocate and initialize arenas. */
 	arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas);
 	if (arenas == NULL) {
diff --git a/jemalloc/src/jemalloc_arena.c b/jemalloc/src/jemalloc_arena.c
index 109b3f6..585be3d 100644
--- a/jemalloc/src/jemalloc_arena.c
+++ b/jemalloc/src/jemalloc_arena.c
@@ -1031,7 +1031,11 @@ arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin)
 
 #ifdef JEMALLOC_TCACHE
 void
-arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind)
+arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind
+#  ifdef JEMALLOC_PROF
+    , uint64_t prof_accumbytes
+#  endif
+    )
 {
 	unsigned i, nfill;
 	arena_bin_t *bin;
@@ -1042,6 +1046,9 @@ arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind)
 
 	bin = &arena->bins[binind];
 	malloc_mutex_lock(&arena->lock);
+#ifdef JEMALLOC_PROF
+	arena_prof_accum(arena, prof_accumbytes);
+#endif
 	for (i = 0, nfill = (tcache_nslots >> 1); i < nfill; i++) {
 		if ((run = bin->runcur) != NULL && run->nfree > 0)
 			ptr = arena_bin_malloc_easy(arena, bin, run);
@@ -1088,6 +1095,19 @@ arena_tcache_fill(arena_t *arena, tcache_bin_t *tbin, size_t binind)
 }
 #endif
 
+#ifdef JEMALLOC_PROF
+void
+arena_prof_accum(arena_t *arena, uint64_t accumbytes)
+{
+
+	arena->prof_accumbytes += accumbytes;
+	if (arena->prof_accumbytes >= prof_interval) {
+		prof_idump();
+		arena->prof_accumbytes -= prof_interval;
+	}
+}
+#endif
+
 /*
  * Calculate bin->run_size such that it meets the following constraints:
  *
@@ -1244,6 +1264,10 @@ arena_malloc_small(arena_t *arena, size_t size, bool zero)
 #  endif
 	arena->stats.allocated_small += size;
 #endif
+#ifdef JEMALLOC_PROF
+	if (isthreaded == false)
+		arena_prof_accum(arena, size);
+#endif
 	malloc_mutex_unlock(&arena->lock);
 
 	if (zero == false) {
@@ -1295,6 +1319,10 @@ arena_malloc_medium(arena_t *arena, size_t size, bool zero)
 #  endif
 	arena->stats.allocated_medium += size;
 #endif
+#ifdef JEMALLOC_PROF
+	if (isthreaded == false)
+		arena_prof_accum(arena, size);
+#endif
 	malloc_mutex_unlock(&arena->lock);
 
 	if (zero == false) {
@@ -1334,6 +1362,9 @@ arena_malloc_large(arena_t *arena, size_t size, bool zero)
 		    arena->stats.lstats[(size >> PAGE_SHIFT) - 1].curruns;
 	}
 #endif
+#ifdef JEMALLOC_PROF
+	arena_prof_accum(arena, size);
+#endif
 	malloc_mutex_unlock(&arena->lock);
 
 	if (zero == false) {
@@ -2035,6 +2066,10 @@ arena_new(arena_t *arena, unsigned ind)
 	}
 #endif
 
+#ifdef JEMALLOC_PROF
+	arena->prof_accumbytes = 0;
+#endif
+
 	/* Initialize chunks. */
 	arena_chunk_tree_dirty_new(&arena->chunks_dirty);
 	arena->spare = NULL;
diff --git a/jemalloc/src/jemalloc_ctl.c b/jemalloc/src/jemalloc_ctl.c
index 75c5c6b..a68e78a 100644
--- a/jemalloc/src/jemalloc_ctl.c
+++ b/jemalloc/src/jemalloc_ctl.c
@@ -34,14 +34,14 @@ CTL_PROTO(epoch)
 #ifdef JEMALLOC_TCACHE
 CTL_PROTO(tcache_flush)
 #endif
-#ifdef JEMALLOC_PROF
-CTL_PROTO(prof_dump)
-#endif
 CTL_PROTO(config_debug)
 CTL_PROTO(config_dss)
 CTL_PROTO(config_dynamic_page_shift)
 CTL_PROTO(config_fill)
 CTL_PROTO(config_lazy_lock)
+CTL_PROTO(config_prof)
+CTL_PROTO(config_prof_libgcc)
+CTL_PROTO(config_prof_libunwind)
 CTL_PROTO(config_stats)
 CTL_PROTO(config_swap)
 CTL_PROTO(config_sysv)
@@ -67,6 +67,13 @@ CTL_PROTO(opt_zero)
 CTL_PROTO(opt_lg_tcache_nslots)
 CTL_PROTO(opt_lg_tcache_gc_sweep)
 #endif
+#ifdef JEMALLOC_PROF
+CTL_PROTO(opt_prof)
+CTL_PROTO(opt_lg_prof_bt_max)
+CTL_PROTO(opt_lg_prof_interval)
+CTL_PROTO(opt_prof_udump)
+CTL_PROTO(opt_prof_leak)
+#endif
 CTL_PROTO(opt_stats_print)
 #ifdef JEMALLOC_TRACE
 CTL_PROTO(opt_trace)
@@ -112,6 +119,10 @@ CTL_PROTO(arenas_nsbins)
 CTL_PROTO(arenas_nmbins)
 CTL_PROTO(arenas_nbins)
 CTL_PROTO(arenas_nlruns)
+#ifdef JEMALLOC_PROF
+CTL_PROTO(prof_dump)
+CTL_PROTO(prof_interval)
+#endif
 #ifdef JEMALLOC_STATS
 CTL_PROTO(stats_chunks_current)
 CTL_PROTO(stats_chunks_total)
@@ -188,18 +199,15 @@ static const ctl_node_t	tcache_node[] = {
 };
 #endif
 
-#ifdef JEMALLOC_PROF
-static const ctl_node_t	prof_node[] = {
-	{NAME("dump"),		CTL(prof_dump)}
-};
-#endif
-
 static const ctl_node_t	config_node[] = {
 	{NAME("debug"),			CTL(config_debug)},
 	{NAME("dss"),			CTL(config_dss)},
 	{NAME("dynamic_page_shift"),	CTL(config_dynamic_page_shift)},
 	{NAME("fill"),			CTL(config_fill)},
 	{NAME("lazy_lock"),		CTL(config_lazy_lock)},
+	{NAME("prof"),			CTL(config_prof)},
+	{NAME("prof_libgcc"),		CTL(config_prof_libgcc)},
+	{NAME("prof_libunwind"),	CTL(config_prof_libunwind)},
 	{NAME("stats"),			CTL(config_stats)},
 	{NAME("swap"),			CTL(config_swap)},
 	{NAME("sysv"),			CTL(config_sysv)},
@@ -228,6 +236,13 @@ static const ctl_node_t opt_node[] = {
 	{NAME("lg_tcache_nslots"),	CTL(opt_lg_tcache_nslots)},
 	{NAME("lg_tcache_gc_sweep"),	CTL(opt_lg_tcache_gc_sweep)},
 #endif
+#ifdef JEMALLOC_PROF
+	{NAME("prof"),			CTL(opt_prof)},
+	{NAME("lg_prof_bt_max"),	CTL(opt_lg_prof_bt_max)},
+	{NAME("lg_prof_interval"),	CTL(opt_lg_prof_interval)},
+	{NAME("prof_udump"),		CTL(opt_prof_udump)},
+	{NAME("prof_leak"),		CTL(opt_prof_leak)},
+#endif
 	{NAME("stats_print"),		CTL(opt_stats_print)},
 #ifdef JEMALLOC_TRACE
 	{NAME("trace"),			CTL(opt_trace)},
@@ -299,6 +314,13 @@ static const ctl_node_t arenas_node[] = {
 	{NAME("lrun"),			CHILD(arenas_lrun)}
 };
 
+#ifdef JEMALLOC_PROF
+static const ctl_node_t	prof_node[] = {
+	{NAME("dump"),		CTL(prof_dump)},
+	{NAME("interval"),	CTL(prof_interval)}
+};
+#endif
+
 #ifdef JEMALLOC_STATS
 static const ctl_node_t stats_chunks_node[] = {
 	{NAME("current"),		CTL(stats_chunks_current)},
@@ -414,12 +436,12 @@ static const ctl_node_t	root_node[] = {
 #ifdef JEMALLOC_TCACHE
 	{NAME("tcache"),	CHILD(tcache)},
 #endif
-#ifdef JEMALLOC_PROF
-	{NAME("prof"),		CHILD(prof)},
-#endif
 	{NAME("config"),	CHILD(config)},
 	{NAME("opt"),		CHILD(opt)},
 	{NAME("arenas"),	CHILD(arenas)},
+#ifdef JEMALLOC_PROF
+	{NAME("prof"),		CHILD(prof)},
+#endif
 	{NAME("stats"),		CHILD(stats)}
 #ifdef JEMALLOC_SWAP
 	,
@@ -938,23 +960,6 @@ RETURN:
 }
 #endif
 
-#ifdef JEMALLOC_PROF
-static int
-prof_dump_ctl(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp,
-    void *newp, size_t newlen)
-{
-	int ret;
-
-	VOID();
-
-	prof_mdump();
-
-	ret = 0;
-RETURN:
-	return (ret);
-}
-#endif
-
 /******************************************************************************/
 
 #ifdef JEMALLOC_DEBUG
@@ -987,6 +992,24 @@ CTL_RO_TRUE_GEN(config_lazy_lock)
 CTL_RO_FALSE_GEN(config_lazy_lock)
 #endif
 
+#ifdef JEMALLOC_PROF
+CTL_RO_TRUE_GEN(config_prof)
+#else
+CTL_RO_FALSE_GEN(config_prof)
+#endif
+
+#ifdef JEMALLOC_PROF_LIBGCC
+CTL_RO_TRUE_GEN(config_prof_libgcc)
+#else
+CTL_RO_FALSE_GEN(config_prof_libgcc)
+#endif
+
+#ifdef JEMALLOC_PROF_LIBUNWIND
+CTL_RO_TRUE_GEN(config_prof_libunwind)
+#else
+CTL_RO_FALSE_GEN(config_prof_libunwind)
+#endif
+
 #ifdef JEMALLOC_STATS
 CTL_RO_TRUE_GEN(config_stats)
 #else
@@ -1054,6 +1077,13 @@ CTL_RO_GEN(opt_zero, opt_zero, bool)
 CTL_RO_GEN(opt_lg_tcache_nslots, opt_lg_tcache_nslots, size_t)
 CTL_RO_GEN(opt_lg_tcache_gc_sweep, opt_lg_tcache_gc_sweep, ssize_t)
 #endif
+#ifdef JEMALLOC_PROF
+CTL_RO_GEN(opt_prof, opt_prof, bool)
+CTL_RO_GEN(opt_lg_prof_bt_max, opt_lg_prof_bt_max, size_t)
+CTL_RO_GEN(opt_lg_prof_interval, opt_lg_prof_interval, size_t)
+CTL_RO_GEN(opt_prof_udump, opt_prof_udump, bool)
+CTL_RO_GEN(opt_prof_leak, opt_prof_leak, bool)
+#endif
 CTL_RO_GEN(opt_stats_print, opt_stats_print, bool)
 #ifdef JEMALLOC_TRACE
 CTL_RO_GEN(opt_trace, opt_trace, bool)
@@ -1145,6 +1175,27 @@ CTL_RO_GEN(arenas_nlruns, nlclasses, size_t)
 
 /******************************************************************************/
 
+#ifdef JEMALLOC_PROF
+static int
+prof_dump_ctl(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp,
+    void *newp, size_t newlen)
+{
+	int ret;
+
+	VOID();
+
+	prof_mdump();
+
+	ret = 0;
+RETURN:
+	return (ret);
+}
+
+CTL_RO_GEN(prof_interval, prof_interval, uint64_t)
+#endif
+
+/******************************************************************************/
+
 #ifdef JEMALLOC_STATS
 CTL_RO_GEN(stats_chunks_current, ctl_stats.chunks.current, size_t)
 CTL_RO_GEN(stats_chunks_total, ctl_stats.chunks.total, uint64_t)
diff --git a/jemalloc/src/jemalloc_stats.c b/jemalloc/src/jemalloc_stats.c
index b0efe74..54aa103 100644
--- a/jemalloc/src/jemalloc_stats.c
+++ b/jemalloc/src/jemalloc_stats.c
@@ -431,13 +431,19 @@ stats_print(void (*write4)(void *, const char *, const char *, const char *,
 		write4(w4opaque, "Boolean JEMALLOC_OPTIONS: ", "", "", "");
 		if ((err = mallctl("opt.abort", &bv, &bsz, NULL, 0)) == 0)
 			write4(w4opaque, bv ? "A" : "a", "", "", "");
+		if ((err = mallctl("opt.prof", &bv, &bsz, NULL, 0)) == 0)
+			write4(w4opaque, bv ? "F" : "f", "", "", "");
 		if ((err = mallctl("opt.junk", &bv, &bsz, NULL, 0)) == 0)
 			write4(w4opaque, bv ? "J" : "j", "", "", "");
+		if ((err = mallctl("opt.prof_leak", &bv, &bsz, NULL, 0)) == 0)
+			write4(w4opaque, bv ? "L" : "l", "", "", "");
 		if ((err = mallctl("opt.overcommit", &bv, &bsz, NULL, 0)) == 0)
 			write4(w4opaque, bv ? "O" : "o", "", "", "");
 		write4(w4opaque, "P", "", "", "");
 		if ((err = mallctl("opt.trace", &bv, &bsz, NULL, 0)) == 0)
 			write4(w4opaque, bv ? "T" : "t", "", "", "");
+		if ((err = mallctl("opt.prof_udump", &bv, &bsz, NULL, 0)) == 0)
+			write4(w4opaque, bv ? "U" : "u", "", "", "");
 		if ((err = mallctl("opt.sysv", &bv, &bsz, NULL, 0)) == 0)
 			write4(w4opaque, bv ? "V" : "v", "", "", "");
 		if ((err = mallctl("opt.xmalloc", &bv, &bsz, NULL, 0)) == 0)
@@ -512,7 +518,6 @@ stats_print(void (*write4)(void *, const char *, const char *, const char *,
 			    "Min active:dirty page ratio per arena: N/A\n", "",
 			    "", "");
 		}
-#ifdef JEMALLOC_TCACHE
 		if ((err = mallctl("opt.lg_tcache_nslots", &sv, &ssz, NULL, 0))
 		    == 0) {
 			size_t tcache_nslots, tcache_gc_sweep;
@@ -528,7 +533,17 @@ stats_print(void (*write4)(void *, const char *, const char *, const char *,
 			    tcache_nslots && ssv >= 0 ? umax2s(tcache_gc_sweep,
 			    10, s) : "N/A", "\n", "");
 		}
-#endif
+		if ((err = mallctl("opt.lg_prof_bt_max", &sv, &ssz, NULL, 0))
+		    == 0) {
+			write4(w4opaque, "Maximum profile backtrace depth: ",
+			    umax2s((1U << sv), 10, s), "\n", "");
+		}
+		if ((err = mallctl("opt.lg_prof_interval", &sv, &ssz, NULL, 0))
+		    == 0) {
+			write4(w4opaque, "Average profile dump interval: ",
+			    umax2s((1U << sv), 10, s), "", "");
+			write4(w4opaque, " (2^", umax2s(sv, 10, s), ")\n", "");
+		}
 		CTL_GET("arenas.chunksize", &sv, size_t);
 		write4(w4opaque, "Chunk size: ", umax2s(sv, 10, s), "", "");
 		CTL_GET("opt.lg_chunk", &sv, size_t);
diff --git a/jemalloc/src/jemalloc_tcache.c b/jemalloc/src/jemalloc_tcache.c
index 4518a14..c54d54e 100644
--- a/jemalloc/src/jemalloc_tcache.c
+++ b/jemalloc/src/jemalloc_tcache.c
@@ -31,14 +31,25 @@ tcache_alloc_hard(tcache_t *tcache, tcache_bin_t *tbin, size_t binind)
 {
 	void *ret;
 
-	arena_tcache_fill(tcache->arena, tbin, binind);
+	arena_tcache_fill(tcache->arena, tbin, binind
+#ifdef JEMALLOC_PROF
+	    , tcache->prof_accumbytes
+#endif
+	    );
+#ifdef JEMALLOC_PROF
+	tcache->prof_accumbytes = 0;
+#endif
 	ret = tcache_bin_alloc(tbin);
 
 	return (ret);
 }
 
 void
-tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem)
+tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem
+#ifdef JEMALLOC_PROF
+    , tcache_t *tcache
+#endif
+    )
 {
 	arena_chunk_t *chunk;
 	arena_t *arena;
@@ -51,6 +62,12 @@ tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem)
 		chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(tbin->slots[0]);
 		arena = chunk->arena;
 		malloc_mutex_lock(&arena->lock);
+#ifdef JEMALLOC_PROF
+		if (arena == tcache->arena) {
+			arena_prof_accum(arena, tcache->prof_accumbytes);
+			tcache->prof_accumbytes = 0;
+		}
+#endif
 		/* Deallocate every object that belongs to the locked arena. */
 		for (i = ndeferred = 0; i < ncached; i++) {
 			ptr = tbin->slots[i];
@@ -216,11 +233,23 @@ tcache_destroy(tcache_t *tcache)
 	for (i = 0; i < nbins; i++) {
 		tcache_bin_t *tbin = tcache->tbins[i];
 		if (tbin != NULL) {
-			tcache_bin_flush(tbin, i, 0);
+			tcache_bin_flush(tbin, i, 0
+#ifdef JEMALLOC_PROF
+			    , tcache
+#endif
+			    );
 			tcache_bin_destroy(tcache, tbin, i);
 		}
 	}
 
+#ifdef JEMALLOC_PROF
+	if (tcache->prof_accumbytes > 0) {
+		malloc_mutex_lock(&tcache->arena->lock);
+		arena_prof_accum(tcache->arena, tcache->prof_accumbytes);
+		malloc_mutex_unlock(&tcache->arena->lock);
+	}
+#endif
+
 	if (arena_salloc(tcache) <= bin_maxclass) {
 		arena_chunk_t *chunk = CHUNK_ADDR2BASE(tcache);
 		arena_t *arena = chunk->arena;
diff --git a/jemalloc/src/prof.c b/jemalloc/src/prof.c
index 8f69c01..7e1d967 100644
--- a/jemalloc/src/prof.c
+++ b/jemalloc/src/prof.c
@@ -21,6 +21,8 @@ size_t		opt_lg_prof_interval = LG_PROF_INTERVAL_DEFAULT;
 bool		opt_prof_udump = false;
 bool		opt_prof_leak = false;
 
+uint64_t	prof_interval;
+
 /*
  * Global hash of (prof_bt_t *)-->(prof_ctx_t *).  This is the master data
  * structure that knows about all backtraces ever captured.
@@ -50,15 +52,6 @@ static pthread_key_t	bt2cnt_tsd;
 /* (1U << opt_lg_prof_bt_max). */
 static unsigned		prof_bt_max;
 
-/*
- * Per arena profile dump interval, measured in bytes allocated.  Each arena
- * triggers a profile dump when it reaches this threshold.  The effect is that
- * the interval between profile dumps is never longer than (prof_interval *
- * narenas), though the actual interval between dumps will tend to be sporadic,
- * and the interval will on average be prof_interval.
- */
-static uint64_t		prof_interval;
-
 static malloc_mutex_t	prof_dump_seq_mtx;
 static uint64_t		prof_dump_seq;
 static uint64_t		prof_dump_iseq;
@@ -79,6 +72,7 @@ static bool		prof_booted = false;
 
 static malloc_mutex_t	enq_mtx;
 static bool		enq;
+static bool		enq_idump;
 static bool		enq_udump;
 
 /******************************************************************************/
@@ -157,16 +151,20 @@ prof_enter(void)
 static inline void
 prof_leave(void)
 {
-	bool udump;
+	bool idump, udump;
 
 	malloc_mutex_unlock(&bt2ctx_mtx);
 
 	malloc_mutex_lock(&enq_mtx);
 	enq = false;
+	idump = enq_idump;
+	enq_idump = false;
 	udump = enq_udump;
 	enq_udump = false;
 	malloc_mutex_unlock(&enq_mtx);
 
+	if (idump)
+		prof_idump();
 	if (udump)
 		prof_udump();
 }
@@ -785,6 +783,7 @@ prof_dump_maps(void)
 			nread = read(mfd, &prof_dump_buf[prof_dump_buf_end],
 			    PROF_DUMP_BUF_SIZE - prof_dump_buf_end);
 		} while (nread > 0);
+		close(mfd);
 	}
 }
 
@@ -955,6 +954,13 @@ prof_idump(void)
 
 	if (prof_booted == false)
 		return;
+	malloc_mutex_lock(&enq_mtx);
+	if (enq) {
+		enq_idump = true;
+		malloc_mutex_unlock(&enq_mtx);
+		return;
+	}
+	malloc_mutex_unlock(&enq_mtx);
 
 	malloc_mutex_lock(&prof_dump_seq_mtx);
 	prof_dump_filename(filename, 'i', prof_dump_iseq);
@@ -1134,6 +1140,7 @@ prof_boot1(void)
 		if (malloc_mutex_init(&enq_mtx))
 			return (true);
 		enq = false;
+		enq_idump = false;
 		enq_udump = false;
 
 		if (atexit(prof_fdump) != 0) {
@@ -1157,17 +1164,5 @@ prof_boot1(void)
 	return (false);
 }
 
-void
-prof_boot2(void)
-{
-
-	if (opt_prof) {
-		/*
-		 * Finish initializing prof_interval, now that narenas is set.
-		 */
-		prof_interval /= narenas;
-	}
-}
-
 /******************************************************************************/
 #endif /* JEMALLOC_PROF */