[parisc-linux-cvs] linux-2.6 jejb

[Grant Grundler]

On Tue, Jan 13, 2004 at 11:24:28AM -0500, James Bottomley wrote:
> On Tue, 2004-01-13 at 11:12, Grant Grundler wrote:
> > The patch looks good to me.
> 
> Could you take a closer look at the ccio-dma.c diff. It corrects what
> appears to be a bug in the ccio resource code (the assumption that all
> resources must fall in the same ccio window that the device has its
> hardware address in).

I took another look but that code isn't familiar to me. When I last
worked on ccio code, IOMMU had nothing to do with resources. Rbrad/willy
wrote the gsc/hppb bus walk code and added the resource tracking
as part of card-mode Dino support. The goal was to enable dynamic
allocation of MMIO space to card-mode Dino devices. Something that
HPUX can't do. :^)

My guess is this code expects the children to be GSC/HPPB devices.
And our parisc bus walk support doesn't seem to know much about secondary
MMIO ranges that GSC graphisc devices or bridges might need.
That leaves dev->hpa as the only obvious MMIO resource a GSC/HPPB
device might need. Migrating this code to support dynamic assignment
to bridges requires (1) the changes you made and (2) corresponding
changes to any other callers who might assume setting dev->hpa
is sufficient. STIFB and Dino code are the only two obvious drivers
I can think of that would want a secondary range.

hth,
grant