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

From: Benjamin Herrenschmidt
Date: Fri Jun 12 2009 - 07:24:34 EST


On Fri, 2009-06-12 at 21:13 +1000, Benjamin Herrenschmidt wrote:
> > I agree with Ingo though that exposing it as a gfp modifier is
> > not so good. I just like the implementation to mask off GFP_WAIT
> > better, and also prefer not to test system state, but have someone
> > just call into slab to tell it not to unconditionally enable
> > interrupts.
>
> But interrupts is just one example. GFP_NOIO is another one vs. suspend
> and resume.
>
> What we have here is the allocator needs to be clamped down based on the
> system state. I think it will not work to try to butcher every caller,
> especially since they don't always know themselves in what state they
> are called.

Let me put it another way....

If you have to teach every call site whether to use one flag or the
other, there is -no- difference with teaching them to call one routine
(alloc_bootmem) vs another (kmalloc).

The way I see thing is that the -whole- point of the exercise is to
remove the need for the callers to have to know in what environment they
are calling kmalloc().

Yes, we do still want that for atomic calls, just because it's a good
way to get people to think twice before allocating things in atomic
context, but that logic pretty much ends there.

If we're going to require any boot time caller of kmalloc() to pass a
different set of flags than any non-boot time caller, then the whole
idea of moving the initialization earlier so a single allocator can be
used is moot.

Cheers,
Ben.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/