Re: mmotm hangs on compaction lock_page

[Mel Gorman <mel@csn.ul.ie>]

On Tue, Jan 11, 2011 at 12:45:51PM -0800, Andrew Morton wrote:
> On Tue, 11 Jan 2011 11:45:21 +0000
> Mel Gorman <mel@csn.ul.ie> wrote:
> 
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -1809,12 +1809,15 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
> >  	bool sync_migration)
> >  {
> >  	struct page *page;
> > +	struct task_struct *p = current;
> >  
> >  	if (!order || compaction_deferred(preferred_zone))
> >  		return NULL;
> >  
> > +	p->flags |= PF_MEMALLOC;
> >  	*did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
> >  						nodemask, sync_migration);
> > +	p->flags &= ~PF_MEMALLOC;
> 
> Thus accidentally wiping out PF_MEMALLOC if it was already set.
> 

Subtle but we can't have reached here if PF_MEMALLOC was previous set.
It gets caught by

        /* Avoid recursion of direct reclaim */
        if (p->flags & PF_MEMALLOC)
                goto nopage;

> It's risky, and general bad practice.  The default operation here
> should be to push the old value and to later restore it.
> 

Arguably we would do this in case that check was ever removed but I
can't imagine why we would allow direct reclaim or compaction to recurse
into direct reclaim or compaction.

> If it is safe to micro-optimise that operation then we need to make
> sure that it's really really safe and that there is no risk of
> accidentally breaking things later on as code evolves.
> 

If the code evolves in that direction, it's pretty dangerous.

> One way of doing that would be to add a WARN_ON(p->flags & PF_MEMALLOC)
> on entry.
> 

That would be reasonable.

> Oh, and since when did we use `p' to identify task_structs?
> 

It's pretty stupid all right but it was the name chosen for other parts
of page_alloc.c. A patch that either removed all local caching or
renamed p to tsk throughout page_alloc.c would be reasonable.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>