Re: [PATCH 08/12] vmscan: Setup pagevec as late as possible inshrink_inactive_list()

[Mel Gorman]

On Wed, Jun 16, 2010 at 04:43:09PM -0700, Andrew Morton wrote:
> On Mon, 14 Jun 2010 12:17:49 +0100
> Mel Gorman <mel@xxxxxxxxx> wrote:
> 
> > shrink_inactive_list() sets up a pagevec to release unfreeable pages. It
> > uses significant amounts of stack doing this. This patch splits
> > shrink_inactive_list() to take the stack usage out of the main path so
> > that callers to writepage() do not contain an unused pagevec on the
> > stack.
> 
> You can get the entire pagevec off the stack - just make it a
> static-to-shrink_inactive_list() pagevec-per-cpu.
> 

That idea has been floated as well. I didn't pursue it because Dave
said that giving page reclaim a stack diet was never going to be the
full solution so I didn't think the complexity was justified.

I kept some of the stack reduction stuff because a) it was there and b)
it would give kswapd extra headroom when calling writepage.

> Locking just requires pinning to a CPU.  We could trivially co-opt
> shrink_inactive_list()'s spin_lock_irq() for that, but
> pagevec_release() can be relatively expensive so it'd be sad to move
> that inside spin_lock_irq().  It'd be better to slap a
> get_cpu()/put_cpu() around the whole thing.
> 

It'd be something interesting to try out when nothing else was happening but
I'm not going to focus on it for the moment unless I think it will really
help this stack overflow problem.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/