On Tue, 26 Apr 2011 15:26:24 +0100 Mel Gorman [off-list ref] wrote:

On Tue, Apr 26, 2011 at 10:30:59PM +1000, NeilBrown wrote:

quoted

On Tue, 26 Apr 2011 08:36:53 +0100 Mel Gorman [off-list ref] wrote:

quoted

+/*
+ * Throttle direct reclaimers if backing storage is backed by the network
+ * and the PFMEMALLOC reserve for the preferred node is getting dangerously
+ * depleted. kswapd will continue to make progress and wake the processes
+ * when the low watermark is reached
+ */
+static void throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
+					nodemask_t *nodemask)
+{
+	struct zone *zone;
+	int high_zoneidx = gfp_zone(gfp_mask);
+	DEFINE_WAIT(wait);
+
+	/* Check if the pfmemalloc reserves are ok */
+	first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone);
+	prepare_to_wait(&zone->zone_pgdat->pfmemalloc_wait, &wait,
+							TASK_INTERRUPTIBLE);
+	if (pfmemalloc_watermark_ok(zone->zone_pgdat, high_zoneidx))
+		goto out;
+
+	/* Throttle */
+	do {
+		schedule();
+		finish_wait(&zone->zone_pgdat->pfmemalloc_wait, &wait);
+		prepare_to_wait(&zone->zone_pgdat->pfmemalloc_wait, &wait,
+							TASK_INTERRUPTIBLE);
+	} while (!pfmemalloc_watermark_ok(zone->zone_pgdat, high_zoneidx) &&
+			!fatal_signal_pending(current));
+
+out:
+	finish_wait(&zone->zone_pgdat->pfmemalloc_wait, &wait);
+}

You are doing an interruptible wait, but only checking for fatal signals.
So if a non-fatal signal arrives, you will busy-wait.

So I suspect you want TASK_KILLABLE, so just use:

    wait_event_killable(zone->zone_pgdat->pfmemalloc_wait,
                        pgmemalloc_watermark_ok(zone->zone_pgdata,
                                                high_zoneidx));

Well, if a normal signal arrives, we do not necessarily want the
process to enter reclaim. For fatal signals, I allow it to continue
because it's not likely to be putting the system under more pressure
if it's exiting.

Yep, I understand that and it doesn't seem unreasonable.

However I don't think the code implements that correctly.

If you get a non-fatal signal, schedule will exit immediately (because of the
TASK_INTERRUPTIBLE setting) and the 'while' clause will succeed because the
signal is not fatal, so it will loop around and try to schedule again, which
will again exit immediately - busy loop.

quoted

(You also have an extraneous call to finish_wait)

Which one? I'm not seeing a flow where finish_wait gets called twice
without a prepare_to_wait in between. 

You don't need to call finish_wait immediately before prepare_to_wait.

It really is best to just use the appropriate 'wait_event*' macro....

NeilBrown

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>