Re: mm, vmscan: commit makes PAE kernel crash nightly (bisected)

[Mel Gorman]

On Fri, Jan 20, 2017 at 12:35:44AM -0600, Trevor Cordes wrote:
> > > Hi!  The git tree above version oom'd after < 24 hours (3:02am) so
> > > it doesn't solve the bug.  If you need a oom messages dump let me
> > > know.  
> > 
> > Yes please.
> 
> The first oom from that night attached.  Note, the oom wasn't as dire
> with your mhocko/4.9.0+ as it usually is with stock 4.8.x: my oom
> detector and reboot script was able to do its thing cleanly before the
> system became unusable.
> 
> I'll await further instructions and test right away.  Maybe I'll try a
> few tuning ideas until then.  Thanks!
> 

Thanks for the OOM report. I was expecting it to be a particular shape and
my expectations were not matched so it took time to consider it further. Can
you try the cumulative patch below? It combines three patches that

1. Allow slab shrinking even if the LRU patches are unreclaimable in
   direct reclaim
2. Shrinks slab based once based on the contents of all memcgs instead
   of shrinking one at a time
3. Tries to shrink slabs if the lowmem usage is too high

Unfortunately it's only boot tested on x86-64 as I didn't get the chance
to setup an i386 test bed.

> > This is why not only Linus hates 32b systems on a large memory
> > systems.
> 
> Completely off-topic: it would be great if rather than pretending PAE
> should work with large RAM (which seems more broken every day), the
> kernel guys put out an officially stated policy of a maximum RAM you
> can use, and try to have the kernel behave for <= that size, and then
> people could use more RAM but clearly "at your own risk, don't bug us
> about problems!".  Other than a few posts about Linus hating it,
> there's nothing official I can find about it in documentation, etc.  It
> gives the (mis)impression that it's perfectly fine to run PAE on a
> zillion GB modern system.  Then we later learn the hard way :-)

The unfortunate reality is that the behaviour is workload dependant so
it's impossible to make a general statement other than "your mileage may
vary considerably".

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 2281ad310d06..76d68a8872c7 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2318,6 +2318,52 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg,
 	}
 }
 
+#ifdef CONFIG_HIGHMEM
+static void balance_slab_lowmem(struct pglist_data *pgdat,
+				struct scan_control *sc)
+{
+	unsigned long lru_pages = 0;
+	unsigned long slab_pages = 0;
+	int zid;
+
+	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
+		struct zone *zone = &pgdat->node_zones[zid];
+
+		if (!populated_zone(zone) || !is_highmem_idx(zid))
+			continue;
+
+		lru_pages += zone_page_state(zone, NR_ZONE_INACTIVE_FILE);
+		lru_pages += zone_page_state(zone, NR_ZONE_ACTIVE_FILE);
+		slab_pages += zone_page_state(zone, NR_SLAB_RECLAIMABLE);
+		slab_pages += zone_page_state(zone, NR_SLAB_UNRECLAIMABLE);
+	}
+
+	/*
+	 * Shrink reclaimable slabs if the number of lowmem slab pages is
+	 * over twice the size of LRU pages. Apply pressure relative to
+	 * the imbalance between LRU and slab pages.
+	 */
+	if (slab_pages > lru_pages << 1) {
+		struct reclaim_state *reclaim_state = current->reclaim_state;
+		unsigned long exceed = (lru_pages << 1) - slab_pages;
+		int nid = pgdat->node_id;
+
+		exceed = min(exceed, slab_pages);
+		shrink_slab(sc->gfp_mask, nid, NULL, exceed, slab_pages);
+		if (reclaim_state) {
+			sc->nr_reclaimed += reclaim_state->reclaimed_slab;
+			reclaim_state->reclaimed_slab = 0;
+		}
+	}
+}
+#else
+static void balance_slab_lowmem(struct pglist_data *pgdat,
+				struct scan_control *sc)
+{
+	return false;
+}
+#endif
+
 /*
  * This is a basic per-node page freer.  Used by both kswapd and direct reclaim.
  */
@@ -2336,6 +2382,27 @@ static void shrink_node_memcg(struct pglist_data *pgdat, struct mem_cgroup *memc
 
 	get_scan_count(lruvec, memcg, sc, nr, lru_pages);
 
+	/*
+	 * If direct reclaiming at elevated priority and the node is
+	 * unreclaimable then skip LRU reclaim and let kswapd poll it.
+	 */
+	if (!current_is_kswapd() &&
+	    sc->priority != DEF_PRIORITY &&
+	    !pgdat_reclaimable(pgdat)) {
+		unsigned long nr_scanned;
+
+		/*
+		 * Fake scanning so that slab shrinking will continue. For
+		 * lowmem restricted allocations, shrink aggressively.
+		 */
+		nr_scanned = SWAP_CLUSTER_MAX << (DEF_PRIORITY - sc->priority);
+		if (!(sc->gfp_mask & __GFP_HIGHMEM))
+			nr_scanned = max(nr_scanned, *lru_pages);
+		sc->nr_scanned += nr_scanned;
+
+		return;
+	}
+
 	/* Record the original scan target for proportional adjustments later */
 	memcpy(targets, nr, sizeof(nr));
 
@@ -2369,6 +2436,7 @@ static void shrink_node_memcg(struct pglist_data *pgdat, struct mem_cgroup *memc
 			}
 		}
 
+		balance_slab_lowmem(pgdat, sc);
 		cond_resched();
 
 		if (nr_reclaimed < nr_to_reclaim || scan_adjusted)
@@ -2533,7 +2601,8 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 			.pgdat = pgdat,
 			.priority = sc->priority,
 		};
-		unsigned long node_lru_pages = 0;
+		unsigned long slab_pressure = 0;
+		unsigned long slab_eligible = 0;
 		struct mem_cgroup *memcg;
 
 		nr_reclaimed = sc->nr_reclaimed;
@@ -2555,12 +2624,8 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 			scanned = sc->nr_scanned;
 
 			shrink_node_memcg(pgdat, memcg, sc, &lru_pages);
-			node_lru_pages += lru_pages;
-
-			if (memcg)
-				shrink_slab(sc->gfp_mask, pgdat->node_id,
-					    memcg, sc->nr_scanned - scanned,
-					    lru_pages);
+			slab_eligible += lru_pages;
+			slab_pressure += sc->nr_reclaimed - reclaimed;
 
 			/* Record the group's reclaim efficiency */
 			vmpressure(sc->gfp_mask, memcg, false,
@@ -2586,12 +2651,12 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
 
 		/*
 		 * Shrink the slab caches in the same proportion that
-		 * the eligible LRU pages were scanned.
+		 * the eligible LRU pages were scanned. For memcg, this
+		 * will apply the cumulative scanning pressure over all
+		 * memcgs.
 		 */
-		if (global_reclaim(sc))
-			shrink_slab(sc->gfp_mask, pgdat->node_id, NULL,
-				    sc->nr_scanned - nr_scanned,
-				    node_lru_pages);
+		shrink_slab(sc->gfp_mask, pgdat->node_id, NULL, slab_pressure,
+							slab_eligible);
 
 		if (reclaim_state) {
 			sc->nr_reclaimed += reclaim_state->reclaimed_slab;
@@ -2683,10 +2748,6 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
 						 GFP_KERNEL | __GFP_HARDWALL))
 				continue;
 
-			if (sc->priority != DEF_PRIORITY &&
-			    !pgdat_reclaimable(zone->zone_pgdat))
-				continue;	/* Let kswapd poll it */
-
 			/*
 			 * If we already have plenty of memory free for
 			 * compaction in this zone, don't free any more.