Re: [PATCH] mm: compaction: Abort compaction if too many pages are isolated and caller is asynchronous

[Mel Gorman <mgorman@suse.de>]

On Mon, Jun 06, 2011 at 02:38:51PM +0200, Andrea Arcangeli wrote:
> On Mon, Jun 06, 2011 at 11:39:24AM +0100, Mel Gorman wrote:
> > Well spotted.
> > 
> > Acked-by: Mel Gorman <mgorman@suse.de>
> > 
> > Minor nit. swapper_space is rarely referred to outside of the swap
> > code. Might it be more readable to use
> > 
> > 	/*
> > 	 * swapcache is accounted as NR_FILE_PAGES but it is not
> > 	 * accounted as NR_SHMEM
> > 	 *
> > 	if (PageSwapBacked(page) && !PageSwapCache(page))
> 
> I thought the comparison on swapper_space would be faster as it was
> immediate vs register in CPU, instead of forcing a memory
> access. Otherwise I would have used the above. Now the test_bit is
> written in C and lockless so it's not likely to be very different
> considering the cacheline is hot in the CPU but it's still referencing
> memory instead register vs immediate comparison.

Ok, I had not considered that. That is a micro-optimisation but it's
there. I thought my version is more readable and migration is not
really a fast path but yours is still better.

-- 
Mel Gorman
SUSE Labs