Sorry was Re: [parisc-linux] FB on c3k

[Grant Grundler]

"M. Grabert" wrote:
> On Wed, 6 Mar 2002, Grant Grundler wrote:
> Argll, I just took a look at the patch and your previous emails
> and now I understand that you try to fix 64/FB.

yes. But user space is still 32-bit. kernel has to
provide 32-bit address for frame buffer and that
seems to work.

> I thought the Visualize-EG stifb driver is also not working on Cxxx
> machines. Is this due to the same problem ?

Thomas Bogendoerf thinks all PA 2.0 machines don't map the user space
address correctly (40-bit physical address) in the CPU TLB/PDIR.
He also found that U-bit (mark pages as uncacheable) needs to be
set even though the CPU shouldn't require it. Maybe this is
a hint that 32-bit kernel needs more work for PA2.0 architecture.

> Or is this
> specific for true 64bit workstations as the Cyyyy (hence the
> 64iocls problem).

The goal is to run 32-bit kernel on Cxxx/Cyyyy machines.
I tried 64-bit to avoid the TLB/PDIR addressing problems.
32 vs 64-bit behaviours just help narrow down the scope
of the bugs.

grant