Re: [hatari-devel] Hatari and OUTSIDE (virtual memory manager)

[Uwe Seimet <Uwe.Seimet@xxxxxxxxx>]

Hi,

Good to hear that you can reproduce the issue.

Just tried another software using the PMMU: ROMSPEED, also available
including its sources on http://www.seimet.de/atari/en/index.html. It's
not a lot of code, but it crashes with two bombs. It does not really do much
except copying and re-mapping the ROM contents using the PMMU.

Best regards

Uwe

> Le 02/10/2018 à 15:45, Uwe Seimet a écrit :
> > Hi,
> > 
> >> thanks a lot for this image, it saved me a lot of time :)
> >>
> >> I was able to boot it using this line :
> >>
> >> hatari -c ~/hatari_uwe.cfg -d '' --acsi 0=outside.img --tos tos306de.img
> >>
> >> This boots to desktop. If I start outconf.app, then I get a crash with
> >> 11 bombs, but Hatari doesn't crash like in your case.
> > 
> > How long did you wait after those bombs? I also get those bombs, and then
> > it takes several seconds (with Hatari running in full speed) before Hatari
> > itself crashes.
> > 
> 
> Ah ok, I didn't wait that long, I just exited after the bombs.
> I get the crash too if I wait, so I can have a look at this. But the 
> crash is certainly an effect of the bombs/exception that happen before.
> 
> >> I tried with and without "cycle exact mode", and in both cases I get a
> >> crash in emulated atari tt, but Hatari itself doesn't crash, so I can't
> >> really reproduce your issue, this might depend on compiler version and
> >> other installed libraries or stack size (which doesn't mean the bug is
> >> not present)
> >>
> >> In your case, does it work when "cycle exact mode / prefetch mode" are
> >> not used, or does it crash too ? Did you find some settings in Hatari
> >> that allowed outside.app not to crash ?
> > 
> > I did not find any setting where it did not crash. My compiler is gcc
> > 7.3.0 with Linux.
> > 
> >> I will have a closer look anyway to see where those 11 bombs come from
> >> after adding some custom traces for PMMU accesses (it's line-f
> >> exception, but that's not necessarily related to the real error).
> > 
> > I assume that the 11 bombs already indicate the problem. Note that OUTSIDE
> > makes (legal) use of several unused bits in the page descriptors, e.g. to mark
> > pages as dirty, non-swappable etc. And correct handling of 68030 stackframes,
> > function codes, transparent translation registers and master stack pointer
> > is important.
> 
> I will have a look at the source for outside too and compare with the 
> cpu/mmu core code, maybe it's related to those unused bits.
> 
> Nicolas
> 
>