Re: [hatari-devel] Bad Mood freeze

[ Thread Index | Date Index | More lists.tuxfamily.org/hatari-devel Archives ]


Thanks,


I may know a trigger condition for the freeze: door closing.

Freeze minute ago happened when door closed in front of my face
(it got half closed before freeze).  I think on previous freeze it
happened when (another) door started to close when I went under it,
and on first freeze door may have started to close behind me after
I had gone through it (and were further away).

Ok noted. I'm not so suspicious about the door itself - but the effect it may have on the drawing of floor or ceiling regions as it starts clipping them. It could indicate DSP timing differences (CPU outracing DSP when the region to draw is small enough). This kind of thing did occur in earlier versions of Hatari when some code optimizations were too tight with timing or incorrectly async.

However if the mouse is being moved when this happens, and if the freeze is in that 'background colour loop' dead-stop you posted previously, a failure in the IKBD handler could explain both cases.



I see also wrong keys being triggered, e.g. ESC when mouse is moved
(as I get to menu).

Also interesting.
 

As BM doesn't anymore work with latest Hatari WinUAE CPU core when
cycle accurate mode is enabled, maybe there's some timing issue.
I.e. this could be Hatari issue.

If it doesn't happen on real device, I wouldn't pay too much attention
on this until it can be reproduced with cycle-accurate Hatari WinUAE
CPU core.

TBH I'm not too surprised if timing differences in emulation cause breakage/freezes in drawing code across different versions. It is pretty likely.

The mouse and keyboard stuff though does freak me out - especially anything that correlates on real HW.

Anyway I'll do a bit more investigation and see what turns up.


If you do see another case where it lands in the IKBD dead-stop location, can you collect some state from Hatari on the ACIAs and the IKBD/MIDI data and control ports? Maybe even a short history of traffic would be interesting. I'd be particularly interested if the IKBD is trying to report a DATE packet. :)

Cheers,
D




Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/