Re: PATCH: Further aacraid work

[Jeff Garzik <jgarzik@pobox.com>]

Alan Cox wrote:
> On Thu, Jun 17, 2004 at 09:10:43PM +0200, Andi Kleen wrote:
> 
>>The AMD64 IOMMU could do it too (and the code to do it exists in
>>2.6). But the problem is that the current IO layer doesn't provide a
>>sufficient fallback path when this fails. You have to promise in
>>advance that you can merge and then later it's too late to change your
>>mind without signalling an IO error.
> 
> 
> I would rather see it below the I/O layer for things like AMD64. The
> reason I say this is that many drivers would suffer from iommu merging not
> gain, and others may have limits.


This reminds me of good ole IDE:  the first-generation SATA controllers 
carry forward the limits of the older PCI IDE controllers:  neither the 
S/G table nor any single S/G entry may span a 64K boundary.

In theory the block layer / SCSI layer DMA boundary stuff takes care of 
this -- but iommu merging _undoes_ the segment splitting that the 
hardware _requires_.  Thus, in ata_fill_sg in libata-core.c, I re-split 
the S/G list after DMA-mapping it.  A bit lame.

James and BenH discussed a solution at the DMA level, but I don't think 
anything ever happened.

	Jeff