Re: GFP_DMA use in SCSI midlayer

[Andi Kleen <ak@suse.de>]

On Monday 02 October 2006 16:52, James Bottomley wrote:
> On Sun, 2006-10-01 at 22:55 +0200, Andi Kleen wrote:
> > This means the majority of drivers will never use it? I don't 
> > claim to be a SCSI expert, but a lot of uses looked unchecked
> > to me from a quick look. That is very worrying because it's
> > only 16MB on x86 and 16MB run out quickly.
> 
> Yes, some cases it's used because the code can't be bothered to check to
> see if it's required ... these cases can be audited and cleaned up.

That would be useful.

> > If it's never used I don't care that much, but it certainly
> > doesn't look like that.
> > 
> > Or do you only need it because pci_map_single() on i386 doesn't
> > do bouncing?  If yes we can probably fix that.
> 
> Heh, that would be the worst place to fix it. We usually have use
> context for the allocation, so can sleep in the worst case.  We don't
> have such context where we do pci_map_single().

Ok then you'll likely want a new function that does this.

Do you always have a device or is a mask enough? 



> > > However, what is wanted in every case is 
> > > ordinary memory, not coherent memory.  They can't simply be replaced
> > > with dma_alloc_coherent because
> > > 
> > > a) it will waste memory for platforms that only do it in page size
> > > multiples
> > > b) It will fail on platforms that can't do it at all.
> > 
> > GFP_DMA will also fail on a lot of platforms, so it's the same.
> 
> No, at the moment it succeeds on all of them (although at odd locations,
> like on parisc GFP_DMA goes up to 4GB).

Not after Christoph's patchkit went in.

-Andi