Re: [hatari-devel] 68030 MMU work

[Thomas Huth <th.huth@xxxxxx>]

 Hi Uwe!

Am Thu, 4 Oct 2012 22:47:20 +0200
schrieb Uwe Seimet <Uwe.Seimet@xxxxxxxxx>:

> > Now that could mean two things ... either the Falcon custom chip
> > that is responsible for dispatching the memory accesses (the Combel
> > chip, I think) still takes care of that supervisor logic ... or the
> > low memory is now protected via the MMU.
> > 
> > I guess we need to decode the MMU tables of TOS to find out...
> 
> Unfortunately I can hardly remember how the PMMU works, and when I
> look at some of my old programs using the PMMU I also don't know
> anymore how they worked ;-).
> 
> Anyway, I don't think the PMMU protects the adresses below $1000. The
> standard page size is 32K and as far as I remember the transparent
> translation registers are not used by TOS 4.0. The memory above $1000
> is not protected, so it cannot be the PMMU.
> And I cannot remember doing anything special in OUTSIDE when a
> smaller page size was configured by the user. All in all I don't
> think there are special rules for the low memory.

Ok, thanks, good to know ... however, in that case, I think there is
still something missing in the MMU emulator - or maybe even my hack on
top of Andreas last patch just might be right.

The problem is, that TOS puts the MMU page tables at address 0x700, so
that's in this special memory area which needs supervisor mode to be
accesses. Now, when the CPU hits a page fault while being in user mode,
it either has to temporarily switch to supervisor mode for accessing
that table or there is some other magic which I did not understand
yet...

 Thomas