Re: amigappc test kernel for CSPPC and BPPC

[Frank Wille <frank%phoenix.owl.de@localhost>]

Jukka Andberg wrote:

> One problem which I probably mentioned previously was that the machine
> usually gets stuck when using dcbz instructions to clear memory
> (cacheable fast memory pages.) 

Yes, I do remember.


> I found out that this seems to only happen when the accessed memory has
> the memory coherence bit enabled.

Why is coherency enabled? I'm sure all Fast RAM is bat-mapped without BAT_M.
So it must be page-table mapped? I saw the pmap-code using PTE_M quite
often.

One difference for your pmap-system is that it is using Chip RAM (for the
tables?). Maybe the cache-attributes for Chip RAM (cache-inhibited and
guarded) do have a bad influence on the whole pmap module...?
Just guessing - I know next to nothing about pmap. :|


> As far as I can tell, no exception is generated.

But you saw any other exceptions correctly handled on the BPPC? Or is there
a problem with this type of exception.


> Normal dcbz errors such as trying to zero uncacheable memory do
> generate exceptions as expected, so this seems to be something
> different.

I remember I had a similar problem with dcbz many years ago when I started
working on amigappc. Never found out what happened and I gave up. After
some months with updated sources the problem was gone...


> Any ideas about this? Maybe my hardware is just broken.

When any Amiga PPC software is working on it I would doubt that.


-- 
Frank Wille