Re: [hatari-devel] Regression in 2.0.0 RC1

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 27/10/2016 à 18:46, Anders Eriksson a écrit :

> Hello.
>
> Ok, have done all the above.
>
> I've tried on Hatari with 1.62 and 2.06.
> I've tried on STe with 1.62 and 2.06.
>
> Register dumps are all zero, both 1.62 and 2.06.
> The test program does not play noise on STe, but does on 2.0.0.
>
> Here are the small tools I made with source and the register dump files
> (if you enjoy looking at 0's in a hexeditor).
>
> http://ae.dhs.nu/tmp/hdma.zip

thanks a lot ; that's strange.

I wonder if the DMA sound doesn't output anything after a reset until 
something is really written to dma start and/or end addresses ?

Could you test this too ?

 - after reset, write 0 to ff8907 (even if it's already 0), then write 
to ff8901 / 8924 / 8922 as before to start dma sound. Do you get sound ?

 - after reset, write $40 (!=0) for example to $ff8907, do you get 
sound then if you write to ff8901 / 8924 / 8922

 - in case you didn't had sound after writing to $ff8907, then write 
$40 to $ff8907 and also write $50 to $ff8913 (end address), then write 
to ff8901 / 8924 / 8922. Do you get sound ? (you should)


Depending on these, one possible fix in Hatari could be to disable STE 
dma sound until a write is really made to start and/or end address, but 
I'm not sure this is how the real STE behaves.


Or maybe it's the microwire volume after reset which is set too high 
under Hatari should be 0 ? Unfortunatelly, it's not possible to read 
microwire regs content.

So, I'm thinking at another test (instead of doing the above ones) : if 
you just "move.b #1,$ffff8901.w" after a reset and read $ff8909/0b/0d a 
few times, does the current dma address changes ?
If so, it would mean that DMA is really started but that's the volume is 0



Nicolas