Re: [hatari-devel] Hatari regression with TOS 2.05 (--fast-boot off, <2MB RAM)

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 08/10/2019 à 00:20, Eero Tamminen a écrit :
> Hi,
>
> On 10/7/19 9:01 AM, Thomas Huth wrote:
> > Am Sun, 6 Oct 2019 19:46:19 +0300
> > schrieb Eero Tamminen <oak@xxxxxxxxxxxxxx>:
> > > While running TOS bootup tester, I noticed that Hatari has regression
> > > with TOS v2.05 (regardless of TOS language version).
> > >
> > > If there's <2MB of ST-RAM and one uses "--fast-boot off" option,
> > > TOS never boots with ST, it just bus errors:
> > [...]
> > > It works also with MegaSTE, and some STE configs, failure is mainly
> > > with ST:
> > > hatari --fast-boot off -s 1 --machine st --tos tos205de.img
> > >
> > > This works fine with Hatari v2.0.0, but doesn't anymore in
> > > Hatari v2.2.1, i.e. it's a regression, a bit old one.
> >
> > I assume you're using the correct version of TOS 2.05, the one that
> > works with plain STs, too? Original TOS 2.05 only worked on STEs, but
>
> Thanks, I didn't know that.  I think I'll just disable
> testing anything else than (Mega)STE with TOS 2.x < 2.06.
>
>
> > there are patched versions available in the internet which work on
> > both. Could you provide the md5sum of your TOS 2.05 image?
>
> $ md5sum tos205*
> 1abfc1d02017cf0f2e877a87973168b1  tos205de.img
> 30052433dc31e90beaff482ca1efdf17  tos205uk.img
> 7cdd45b6aac66a21bfb357d9334e46db  tos205us.img
>
>
> > Anyway, if you've got some spare time, could you bisect the issue to
> > see where it broke?
>
> Here:
> -------------------------------------------------
> commit 8409e8b05b027df80170acb42fc3b7ecce810866
> Author: Nicolas Pomarède <npomarede@xxxxxxxxxxxx>
> Date:   Wed Jan 4 23:24:08 2017 +0100
>
>      Add full support for STF/STE MMU at $FF8001 and address translation
>      This emulates the behaviour of the RAS/CAS signals when addressing 
> RAM,
>      depending on the size of the 2 physical RAM banks and on the size of
>      the logical banks defined in bit 0-1 and 2-3 at $FF8001.
>      This allows TOS/Emutos to detect RAM size as a real ST would do.
>      Thanks to Christian Zietz for his documentation of the CAS/RAS 
> signal mapping
> -------------------------------------------------

Hi

this migh explain the problem : if the TOS 2.05 you tested is only able 
to test RAM in an "STE way" and your machine type was set to "STF", then 
RAM testing might indeed fail due to unsupported HW by this TOS version.

RAM testing and RAS/CAS signals are really different between STF and STE 
(similarily, running an old TOS 1.02 (STF only) for example would fail 
on STE machine type)

Nicolas