Re: [GIT PULL v2] Early SLAB fixes for 2.6.31

[Pekka Enberg <penberg@cs.helsinki.fi>]

Hi Linus,

On Tue, 2009-06-16 at 12:33 -0700, Linus Torvalds wrote:
> I mean that there should never be any _need_ for dynamically querying 
> whether slab is available or not. Nothing should ever do it, because we 
> should strive for slab being available so early that anything that happens 
> before it _knows_ that it is special case code (eg "set up initial page 
> tables") and would never have any reason what-so-ever for even 
> conditionally asking "should I use slab".

So how does the page allocator fit in this new scheme of things? I have
been looking at doing the cleanups you and Christoph suggested and it
seems to me we need a 'system_gfp_mask' that's respected by basically
everyone, including __might_sleep() and other debugging functions.

If we want to keep the masking within the slab allocator, we need to
make sure there are two sets of APIs: one for slab and one for the page
allocator. Otherwise we end up masking in the debugging checks but not
in page_alloc().

			Pekka