Re: 2.6.20-mm1 [kernel BUG at mm/swap.c:442]

[Christoph Lameter <clameter@sgi.com>]

On Thu, 15 Feb 2007, Andrew Morton wrote:

> I don't immediately see why that code isn't racy: the page can remain
> in the pagevec for arbitrary amounts of time and someone can come along
> and mlock it again.  But given the ease with which you're hitting this,
> it may not be a race.

As long as the page is on the pagevec it should be off the LRU. 
Marking a page PageMlocked requires the page to be on the LRU. So a page 
cannot be marked PageMlocked as long as it is on the regular pagevecs.

Somehow a page off the LRU was marked PageMlocked. Or a new anonymous page 
was allocated and marked PageMlocked and then some later processing put it 
onto the LRU?

Maybe try_to_set_mlocked does work some havoc here.

Could you see if this patch fixes it? This just disabled an optimization 
to set PageMlocked early.

Index: linux-2.6.20-mm1/mm/memory.c
===================================================================
--- linux-2.6.20-mm1.orig/mm/memory.c	2007-02-15 14:35:41.000000000 -0800
+++ linux-2.6.20-mm1/mm/memory.c	2007-02-15 14:35:54.000000000 -0800
@@ -930,6 +930,8 @@ static void try_to_set_mlocked(struct pa
 	struct zone *zone;
 	unsigned long flags;
 
+	return;
+
 	if (!PageLRU(page) || PageMlocked(page))
 		return;
 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/