Add the stats.retained and stats.arenas.<i>.retained statistics.

This resolves #367.

doc/jemalloc.xml.in

@@ -2186,6 +2186,25 @@ typedef struct {
         linkend="stats.resident"><mallctl>stats.resident</mallctl></link>.</para></listitem>
       </varlistentry>
 
+      <varlistentry id="stats.retained">
+        <term>
+          <mallctl>stats.retained</mallctl>
+          (<type>size_t</type>)
+          <literal>r-</literal>
+          [<option>--enable-stats</option>]
+        </term>
+        <listitem><para>Total number of bytes in virtual memory mappings that
+        were retained rather than being returned to the operating system via
+        e.g. <citerefentry><refentrytitle>munmap</refentrytitle>
+        <manvolnum>2</manvolnum></citerefentry>.  Retained virtual memory is
+        typically untouched, decommitted, or purged, so it has no strongly
+        associated physical memory (see <link
+        linkend="arena.i.chunk_hooks">chunk hooks</link> for details).  Retained
+        memory is excluded from mapped memory statistics, e.g. <link
+        linkend="stats.mapped"><mallctl>stats.mapped</mallctl></link>.
+        </para></listitem>
+      </varlistentry>
+
       <varlistentry id="stats.arenas.i.dss">
         <term>
           <mallctl>stats.arenas.&lt;i&gt;.dss</mallctl>
@@ -2266,6 +2285,18 @@ typedef struct {
         <listitem><para>Number of mapped bytes.</para></listitem>
       </varlistentry>
 
+      <varlistentry id="stats.arenas.i.retained">
+        <term>
+          <mallctl>stats.arenas.&lt;i&gt;.retained</mallctl>
+          (<type>size_t</type>)
+          <literal>r-</literal>
+          [<option>--enable-stats</option>]
+        </term>
+        <listitem><para>Number of retained bytes.  See <link
+        linkend="stats.retained"><mallctl>stats.retained</mallctl></link> for
+        details.</para></listitem>
+      </varlistentry>
+
       <varlistentry id="stats.arenas.i.metadata.mapped">
         <term>
           <mallctl>stats.arenas.&lt;i&gt;.metadata.mapped</mallctl>

include/jemalloc/internal/ctl.h

@@ -61,6 +61,7 @@ struct ctl_stats_s {
 	size_t			metadata;
 	size_t			resident;
 	size_t			mapped;
+	size_t			retained;
 	unsigned		narenas;
 	ctl_arena_stats_t	*arenas;	/* (narenas + 1) elements. */
 };

include/jemalloc/internal/stats.h

@@ -102,6 +102,14 @@ struct arena_stats_s {
 	/* Number of bytes currently mapped. */
 	size_t		mapped;
 
+	/*
+	 * Number of bytes currently retained as a side effect of munmap() being
+	 * disabled/bypassed.  Retained bytes are technically mapped (though
+	 * always decommitted or purged), but they are excluded from the mapped
+	 * statistic (above).
+	 */
+	size_t		retained;
+
 	/*
 	 * Total number of purge sweeps, total number of madvise calls made,
 	 * and total pages purged in order to keep dirty unused memory under

src/arena.c

@@ -3414,6 +3414,7 @@ arena_stats_merge(tsd_t *tsd, arena_t *arena, unsigned *nthreads,
 	    decay_time, nactive, ndirty);
 
 	astats->mapped += arena->stats.mapped;
+	astats->retained += arena->stats.retained;
 	astats->npurge += arena->stats.npurge;
 	astats->nmadvise += arena->stats.nmadvise;
 	astats->purged += arena->stats.purged;

src/chunk.c

@@ -444,15 +444,21 @@ static void *
 chunk_alloc_retained(tsd_t *tsd, arena_t *arena, chunk_hooks_t *chunk_hooks,
     void *new_addr, size_t size, size_t alignment, bool *zero, bool *commit)
 {
+	void *ret;
 
 	assert(size != 0);
 	assert((size & chunksize_mask) == 0);
 	assert(alignment != 0);
 	assert((alignment & chunksize_mask) == 0);
 
-	return (chunk_recycle(tsd, arena, chunk_hooks,
+	ret = chunk_recycle(tsd, arena, chunk_hooks,
 	    &arena->chunks_szad_retained, &arena->chunks_ad_retained, false,
-	    new_addr, size, alignment, zero, commit, true));
+	    new_addr, size, alignment, zero, commit, true);
+
+	if (config_stats && ret != NULL)
+		arena->stats.retained -= size;
+
+	return (ret);
 }
 
 void *
@@ -617,6 +623,9 @@ chunk_dalloc_wrapper(tsd_t *tsd, arena_t *arena, chunk_hooks_t *chunk_hooks,
 	    arena->ind);
 	chunk_record(tsd, arena, chunk_hooks, &arena->chunks_szad_retained,
 	    &arena->chunks_ad_retained, false, chunk, size, zeroed, committed);
+
+	if (config_stats)
+		arena->stats.retained += size;
 }
 
 static bool

src/ctl.c

@@ -192,6 +192,7 @@ CTL_PROTO(stats_arenas_i_decay_time)
 CTL_PROTO(stats_arenas_i_pactive)
 CTL_PROTO(stats_arenas_i_pdirty)
 CTL_PROTO(stats_arenas_i_mapped)
+CTL_PROTO(stats_arenas_i_retained)
 CTL_PROTO(stats_arenas_i_npurge)
 CTL_PROTO(stats_arenas_i_nmadvise)
 CTL_PROTO(stats_arenas_i_purged)
@@ -204,6 +205,7 @@ CTL_PROTO(stats_active)
 CTL_PROTO(stats_metadata)
 CTL_PROTO(stats_resident)
 CTL_PROTO(stats_mapped)
+CTL_PROTO(stats_retained)
 
 /******************************************************************************/
 /* mallctl tree. */
@@ -458,6 +460,7 @@ static const ctl_named_node_t stats_arenas_i_node[] = {
 	{NAME("pactive"),	CTL(stats_arenas_i_pactive)},
 	{NAME("pdirty"),	CTL(stats_arenas_i_pdirty)},
 	{NAME("mapped"),	CTL(stats_arenas_i_mapped)},
+	{NAME("retained"),	CTL(stats_arenas_i_retained)},
 	{NAME("npurge"),	CTL(stats_arenas_i_npurge)},
 	{NAME("nmadvise"),	CTL(stats_arenas_i_nmadvise)},
 	{NAME("purged"),	CTL(stats_arenas_i_purged)},
@@ -484,6 +487,7 @@ static const ctl_named_node_t stats_node[] = {
 	{NAME("metadata"),	CTL(stats_metadata)},
 	{NAME("resident"),	CTL(stats_resident)},
 	{NAME("mapped"),	CTL(stats_mapped)},
+	{NAME("retained"),	CTL(stats_retained)},
 	{NAME("arenas"),	CHILD(indexed, stats_arenas)}
 };
 
@@ -591,6 +595,7 @@ ctl_arena_stats_smerge(ctl_arena_stats_t *sstats, ctl_arena_stats_t *astats)
 
 	if (config_stats) {
 		sstats->astats.mapped += astats->astats.mapped;
+		sstats->astats.retained += astats->astats.retained;
 		sstats->astats.npurge += astats->astats.npurge;
 		sstats->astats.nmadvise += astats->astats.nmadvise;
 		sstats->astats.purged += astats->astats.purged;
@@ -745,6 +750,8 @@ ctl_refresh(tsd_t *tsd)
 		    ctl_stats.arenas[ctl_stats.narenas].pdirty) << LG_PAGE);
 		ctl_stats.mapped = base_mapped +
 		    ctl_stats.arenas[ctl_stats.narenas].astats.mapped;
+		ctl_stats.retained =
+		    ctl_stats.arenas[ctl_stats.narenas].astats.retained;
 	}
 
 	ctl_epoch++;
@@ -2108,6 +2115,7 @@ CTL_RO_CGEN(config_stats, stats_active, ctl_stats.active, size_t)
 CTL_RO_CGEN(config_stats, stats_metadata, ctl_stats.metadata, size_t)
 CTL_RO_CGEN(config_stats, stats_resident, ctl_stats.resident, size_t)
 CTL_RO_CGEN(config_stats, stats_mapped, ctl_stats.mapped, size_t)
+CTL_RO_CGEN(config_stats, stats_retained, ctl_stats.retained, size_t)
 
 CTL_RO_GEN(stats_arenas_i_dss, ctl_stats.arenas[mib[2]].dss, const char *)
 CTL_RO_GEN(stats_arenas_i_lg_dirty_mult, ctl_stats.arenas[mib[2]].lg_dirty_mult,
@@ -2119,6 +2127,8 @@ CTL_RO_GEN(stats_arenas_i_pactive, ctl_stats.arenas[mib[2]].pactive, size_t)
 CTL_RO_GEN(stats_arenas_i_pdirty, ctl_stats.arenas[mib[2]].pdirty, size_t)
 CTL_RO_CGEN(config_stats, stats_arenas_i_mapped,
     ctl_stats.arenas[mib[2]].astats.mapped, size_t)
+CTL_RO_CGEN(config_stats, stats_arenas_i_retained,
+    ctl_stats.arenas[mib[2]].astats.retained, size_t)
 CTL_RO_CGEN(config_stats, stats_arenas_i_npurge,
     ctl_stats.arenas[mib[2]].astats.npurge, uint64_t)
 CTL_RO_CGEN(config_stats, stats_arenas_i_nmadvise,

src/stats.c

@@ -259,7 +259,7 @@ stats_arena_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	unsigned nthreads;
 	const char *dss;
 	ssize_t lg_dirty_mult, decay_time;
-	size_t page, pactive, pdirty, mapped;
+	size_t page, pactive, pdirty, mapped, retained;
 	size_t metadata_mapped, metadata_allocated;
 	uint64_t npurge, nmadvise, purged;
 	size_t small_allocated;
@@ -349,6 +349,9 @@ stats_arena_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	CTL_M2_GET("stats.arenas.0.mapped", i, &mapped, size_t);
 	malloc_cprintf(write_cb, cbopaque,
 	    "mapped:                  %12zu\n", mapped);
+	CTL_M2_GET("stats.arenas.0.retained", i, &retained, size_t);
+	malloc_cprintf(write_cb, cbopaque,
+	    "retained:                %12zu\n", retained);
 	CTL_M2_GET("stats.arenas.0.metadata.mapped", i, &metadata_mapped,
 	    size_t);
 	CTL_M2_GET("stats.arenas.0.metadata.allocated", i, &metadata_allocated,
@@ -597,7 +600,7 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 
 	if (config_stats) {
 		size_t *cactive;
-		size_t allocated, active, metadata, resident, mapped;
+		size_t allocated, active, metadata, resident, mapped, retained;
 
 		CTL_GET("stats.cactive", &cactive, size_t *);
 		CTL_GET("stats.allocated", &allocated, size_t);
@@ -605,10 +608,11 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 		CTL_GET("stats.metadata", &metadata, size_t);
 		CTL_GET("stats.resident", &resident, size_t);
 		CTL_GET("stats.mapped", &mapped, size_t);
+		CTL_GET("stats.retained", &retained, size_t);
 		malloc_cprintf(write_cb, cbopaque,
 		    "Allocated: %zu, active: %zu, metadata: %zu,"
-		    " resident: %zu, mapped: %zu\n",
+		    " resident: %zu, mapped: %zu, retained: %zu\n",
-		    allocated, active, metadata, resident, mapped);
+		    allocated, active, metadata, resident, mapped, retained);
 		malloc_cprintf(write_cb, cbopaque,
 		    "Current active ceiling: %zu\n",
 		    atomic_read_z(cactive));