Michael Schmitz wrote:
Nothing is perfect, not even ARAnyM. I must admit that there was (yet another) MMU emulation error (pretty nasty to catch)
this is in fact a CPU emulation omission. The original UAE CPU core was not designed with MMU in mind, it seems:
https://lists.bobek.cz/pipermail/cz-bobek-lists-aranym/2006-November/009130.html
Thanks for tracking it down :-) Seriously, ARAnyM has helped me tremendously in sorting out 2.6 porting issues. Now that I have to work on drivers for non-emulated hardware (Yay! Many thanks again for the EtherNEC boards), it still serves to cross check I did not break anything inadevertently.
The current CVS fixes LINK and RTD instructions that were misbehaving in the special case (crossed page boundary to an unmapped region) so the ARAnyM built from todays CVS could work more stable than the 0.9.4beta (as found in Debian). But I must admit that to my knowledge this problem also affects the MOVE instruction ("only" 10 opcodes, not all MOVEs, and only when on page boundary to an unmapped page etc.) so I can't swear that ARAnyM is 100% stable.
Nothing is ever perfect. Console yourself with the thought that even Motorola got that page boundary crossing on exceptions thing horribly wrong in their 040LC series :-)
Really? Oh. That helps :-) The good thing is that ARAnyM's CPU is in software so it can be fixed without wasting silion :-)
Petr