Re: [PATCH v2] slab,slub: ignore __GFP_WAIT if we're booting or suspending

[Pavel Machek <pavel@ucw.cz>]

On Sat 2009-06-20 09:50:09, Benjamin Herrenschmidt wrote:
> On Sat, 2009-06-20 at 01:23 +0200, Pavel Machek wrote:
> > > No. First, code that assumes GFP_KERNEL don't fail is stupid. Any
> > > allocation should always be assumed to potentially fail.
> > 
> > Stupid, yes. Uncommon? Not sure.
> 
> A lot less than it used to be, we've been fixing those by the truckload
> over the past few years. But again, if allocations start failing that
> early at boot, you are likely to be doomed anyway. Still, better to do
> proper error handling, and I think we -mostly- do (ok, not -always-).
> 
> > > Then, if you start failing allocations at boot time, then you aren't
> > > going anywhere are you ?
> > 
> > Exactly. So boot code should have access to all the memory, right?
> > Setting some aside for GFP_ATOMIC does not make sense in that context.
> 
> I'm not certain what you mean here. If you're going to hit the atomic
> reserve that early, you aren't going anywhere neither :-)
> 
> Is there any real problem you are trying to solve here or is it all
> just academic ?

Academic for boot, probably real for suspend/resume. There the atomic
reserves could matter because the memory can be pretty full when you
start suspend.
								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html