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

[Nick Piggin <npiggin@suse.de>]

On Mon, Jun 15, 2009 at 12:57:39PM +0300, Pekka Enberg wrote:
> On Mon, 2009-06-15 at 19:51 +1000, Benjamin Herrenschmidt wrote:
> > I think the boot order is too likely to change to make it a sane thing
> > to have all call sites "know" at what point they are in the boot
> > process. In your example, what does GFP_BOOT would mean ? Before
> > scheduler is initialized ? before interrupts are on ?
> 
> Btw, I think this is a pretty important point. Linus suggested trying to
> make slab initialization even earlier than what we now have. If we do
> require GFP_BOOT annotations, then we'd need to sprinkle those all over
> the place when we do that.

I don't understand. You'd be converting all these from bootmem
anyway, so where's the problem?

 
> So from code shuffling point of view, it's better to support GFP_KERNEL
> (almost) everywhere rather than require special annotations.

Nor this. We require special allocation annotations *everywhere*
according to context. Why is using GFP_KERNEL for unsleeping
allocations a good thing if we have GFP_ATOMIC etc?

We could also mask off __GFP_WAIT from allocations when we take
a spinlock or enter an interrupt, right?

Init code doesn't deserve to be more lazy than anybody else, and
part of the reason why such a core piece of code is so crufty
is exactly because people have been lazy there.