Re: [patch 1/6] sparc: remove dependency on __GFP_NOFAIL

[David Rientjes <rientjes@google.com>]

On Tue, 20 Jul 2010, David Miller wrote:

> > The kmalloc() in mdesc_kmalloc() is failable, so remove __GFP_NOFAIL from
> > its mask.
> > 
> > Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> > Signed-off-by: David Rientjes <rientjes@google.com>
> 
> The __GFP_NOFAIL is there intentionally.
> 
> The code above this, in the cases where the machine description is
> dynamically updated by the hypervisor at run time, long after boot,
> has no failure handling.
> 
> We absolutely must accept the machine descriptor update and fetch it
> from the hypervisor into a new buffer.
> 
> Please don't remove this.
> 

Ok, fair enough.  I was convinced by the error handling in both 
mdesc_update() and mdesc_kmallloc() that this was a failable allocation, 
but I understand how mdesc_update() must succeed given your description.  
We can remove those branches from those two functions, though, since 
__GFP_NOFAIL will always succeed before returning.

I'm planning on replacing __GFP_NOFAIL with a __GFP_KILLABLE flag that 
will use all of the page allocator's capabilities (direct reclaim, memory 
compaction for order > 0, and the oom killer) before failing.  Then, 
existing __GFP_NOFAIL users can use

	do {
		page = alloc_page(GFP_KERNEL | __GFP_KILLABLE);
	} while (!page);

to remove several branches from the page allocator that we'll no longer 
need.  I'll do this in phase two and make sure to convert this instance to 
do that.

Thanks!

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>