Re: [hatari-devel] Hatari Falcon Crossbar emulation segfault with EmuTOS

[ Thread Index | Date Index | More Archives ]

I really don't think that the problem is in the crossbar, but in the CPU timings.

You can see a screenshot of the last part of Eko sysytem demo I took on hatari's screenshots section with an older version of hatari. When I switched to the new core, it improved many demos, but it generated a regression for this one.

I've done many tests to try to improve it, but nothing good until now.

I've heard this demo from beginning to end without problem some times, but it seems that timings are important. And actually, it seems to crash because of timings between the DSP and the CPU and not because of the crossbar.
(same for another regression demo : MOAI)

But 68030 accurate emulation is quite a hard part to do.


Le 06/06/2012 21:41, Eero Tamminen a écrit :

On keskiviikko 06 kesäkuu 2012, Laurent Sallafranque wrote:
In which case does hatari crash ?
Have you got a demo that generates a crossbar crash ?
(I don't remember having encountered one)
This is enough to get Hatari to crash:
$ hatari --tos etos512k.img  --machine falcon eko_syst.em/system.prg

("--fastfdc yes --fast-forward yes --bios-intercept --trace xbios"
gets you faster to the crash and you see what calls it does before it.)

Use e.g. the latest EmuTOS CVS version to which Vincent just posted
a link.

	- Eero



Le 06/06/2012 20:44, Eero Tamminen a écrit :

On keskiviikko 06 kesäkuu 2012, Nicolas Pomarède wrote:
Le 05/06/2012 22:35, Eero Tamminen a écrit :
Even if Hatari wouldn't crash, it may be emulating things wrong
because at least wrong dmaRecord.frameCounter value can scribble
all over Hatari's own memory, or over the emulated memory.

I guess invalid dmaPlay.frameCounter values just cause invalid
reads i.e. crashes without memory corruption and the evil stuff
that can follow from that.
Well, so far I'm unfortunately very busy, at least not with enough
time to dig more precisely in the falcon sound part.The little spare
time I have at the moment will be used to release next Hatari version
(including any remaining bugs such as this one).

I'm afraid this will have to stay this way for now, unless someone
else wants to give it a look before next release.
Ok, I'll add that as item to todo.txt.

By the way, it could also be an error in emutos where sound addresses
would not be set correctly.
EmuTOS is still lacking most of XBios functions that appeared
with TOS v4; everything for DSP, sound matrix etc.  That's
why things go wrong on the Atari side.

But the point is that regardless of how badly things go at Atari
side, Hatari itself shouldn't crash when Atari registers have
bogus values.

	- Eero

Mail converted by MHonArc 2.6.19+