Re: [PATCH 3/7] mm: remove GFP_HIGHUSER_PAGECACHE

[Hugh Dickins <hugh@veritas.com>]

On Thu, 20 Nov 2008, Mel Gorman wrote:
> On Thu, Nov 20, 2008 at 01:16:16AM +0000, Hugh Dickins wrote:
> > GFP_HIGHUSER_PAGECACHE is just an alias for GFP_HIGHUSER_MOVABLE,
> > making that harder to track down: remove it, and its out-of-work
> > brothers GFP_NOFS_PAGECACHE and GFP_USER_PAGECACHE.
> 
> The use of GFP_HIGHUSER_PAGECACHE instead of GFP_HIGHUSER_MOVABLE was a
> deliberate decision at the time. I do not have an exact patch to point

I realize it didn't happen by accident!

> you at but the intention behind GFP_HIGHUSER_PAGECACHE was that it be
> self-documenting. i.e. one could easily find what GFP placement decisions
> have been made for page-cache allocations.

I see it as self-obscuring, not self-documenting: of course pagecache
pages will normally be allocated with the GFP mask for pagecache pages,
but what is that?  Ah, it's GFP_HIGHUSER_MOVABLE.

Please let's not go down the road, I mean, let's retrace our steps
up the road, of assigning a unique GFP name for every use of pages.

> So, I'm happy with GFP_NOFS_PAGECACHE and GFP_USER_PAGECACHE going away and
> it makes perfect sense. GFP_HIGHUSER_PAGECACHE I'm not as keen on backing
> out. I like it's self-documenting aspect but aliases sometimes make peoples
> teeth itch.

(No, what made my teeth itch was "is this safe?" in memory.c ;)

> If it's really hated, then could a comment to the affect of
> "Marked movable for a page cache allocation" be placed near the call-sites
> instead?

I'd prefer not.

The only places where GFP_HIGHUSER_PAGECACHE appeared
were the mapping_set_gfp_mask when initializing an inode, and
hotremove_migrate_alloc().  The latter allocating for anonymous
pages also, like most places where GFP_HIGHUSER_MOVABLE is specified.

But I'd better not complain that it's not obvious to me which
should be marked with your comment and which not: you'll point to
that as evidence that we're missing out on the self-documentation!

Perhaps the problem is that nobody else has been following your lead.

Hugh

--
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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>