Re: Altix I/O code reorganization - 7 of 21

Re: Altix I/O code reorganization - 7 of 21

[Pat Gefre]

> 003-pci-support:
>    The PCI DMA implementation is still ubercomplicated.  See the PCI DMA code
>    I sent you long ago.
>    Of the code in pci_extensions.c only two are actually used in the kernel
>    (snia_pcibr_rrb_alloc and snia_pcidev_endian_set), please remove the unused
>    other ones.
> 

Yes, the PCI DMA mapping code is very robust and tries to export 
the many hardware features that are available on our system.  However, not 
all of these features are "reachable" via the standard Linux PCI mapping 
interfaces.  We have removed all unreacheable features.  However, we can 
still support future requirements with these new interfaces when needed.

Yes, we did look at your patch dated: December 4th 2003.  It was very good, but 
did not go far enough.  We felt that we can trim more of the code.  The current 
result, we believe is much cleaner and easier to support.  The dma path now is 
much leaner.

We have removed unused calls in pci_extensions.c.  We have also removed the
call snia_pcidev_endian_set.  IOC4 endianess is already correctly set by
Prom during configuration and initialization.



For the new code see:
ftp://oss.sgi.com/projects/sn2/sn2-update/003-pci-support

Re: Altix I/O code reorganization - 7 of 21

[Christoph Hellwig]

On Wed, Aug 11, 2004 at 06:29:25PM -0500, Pat Gefre wrote:
> Yes, the PCI DMA mapping code is very robust

no.

> and tries to export 
> the many hardware features that are available on our system.  However, not 
> all of these features are "reachable" via the standard Linux PCI mapping 
> interfaces.  We have removed all unreacheable features.  However, we can 
> still support future requirements with these new interfaces when needed.
> 
> Yes, we did look at your patch dated: December 4th 2003.  It was very good, but 

I mean a more recent patch I sent to Colin.