Re: [PATCH] add __GFP_ZERP to GFP_LEVEL_MASK

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 23 Jul 2007 15:41:36 -0700 (PDT)
Linus Torvalds <torvalds@linux-foundation.org> wrote:

> 
> 
> On Mon, 23 Jul 2007, Andrew Morton wrote:
> > 
> > OK, well that was weird.  So
> > 
> > 	kmalloc(42, GFP_KERNEL|__GFP_ZERO);
> > 
> > duplicates
> > 
> > 	kzalloc(42, GFP_KERNEL);
> > 
> > Why do it both ways?
> 
> Both ways? The latter *is* the former. That's how kzalloc() is implemented 
> these days.

<looks>

So this:

	/*
	 * Be lazy and only check for valid flags here,  keeping it out of the
	 * critical path in kmem_cache_alloc().
	 */
	BUG_ON(flags & ~(GFP_DMA | __GFP_ZERO | GFP_LEVEL_MASK));

would no longer need the __GFP_ZERO.  Ditto in slob's new_slab().


> Andrew - all these patches came through you. You didn't realize?

Well.  I didn't memorise the past few months' 250-odd slab/slob/slub
patches..

My point is, I don't think we want some code doing
kmalloc(42, GFP_KERNEL|__GFP_ZERO) and some other code doing
kzalloc(42, GFP_KERNEL).  But this patch does nothing to increase the
chances of that happening, so I'm happy.