Re: [MiNT] Crash log

[Odd Skancke <ozk@atari.org>]

ons, 26,.04.2006 kl. 11.25 +0200, skrev Frank Naumann:
> Hello!
> 
> > As we've already talked about, this is most likely a hardware bug in
> > the clones, Milan/Hades. On both my milan/hades applications are allowed
> > to read/write this area. Generating access faults to the first 2Kb of
> > system RAM is a task done by the address-logic in real Atari hardware,
> > and indipendant of the PMMU settings. Even ST's sets supervisor
> > restrictions to this area, it has no PMMU ;-)
> 
> For Milan I can answer the following:
> 
> Milan always use the MMU, Milan use 8kb pages. First versions of 
> Bootblock/TOS setup the the first page (the one from 0) to be supervisor 
> protected.

 Yes, but accesses to the first 2Kb of ram should not be controlled by
the PMMU, but by the address logic. On the CT63, this works as expected.

> 
> But in this area are also some VDI variables (or something like that) that 
> are defined to be non-supervisor accessible. So for MagiC (and TOS/NVDI) 
> compatibility the first page was changed to be non-supervisor accessible.

 Yes, these are details we need to know. We need to find out what has to
be relocated. I guess that we must set membot (lowest memory location
available for systemheap) and do such things at a very early stage in
the bootprocess because of autoprograms, etc. How TOS reacts to this I
dont know. Hopefully, if LineA normally would be a problem, nvdi will
hopefully install correctly.

> 
> I discussed a lot woth Michael Schwingen about this but he didn't changed 
> it back or made it configurable. The only way is that the kernel change 
> back the primary MMU tree so that the first page is supervisor only. Don't 
> know if that have negative side effect for applications.

 By the time the kernel runs, I think it is way too late to do this,
mainly because the systemheap begins below the first 8Kb of system-ram,
and most likely already have code/data in it that should be accessible
from userspace. cookiejars, buffers, autoprograms, etc...

 I take it Michael Schwingen isnt active in here anymore, but if he
reads this; Is there a way to fix the Milan hardware?


 Best Regards,
Odd Skancke