Re: [hatari-devel] Regression (?) in 2.4.1 - Time Slices by Defence Force

[Thomas Huth <th.huth@xxxxxxxxx>]

Am Tue, 27 Dec 2022 16:39:06 +0100
schrieb Nicolas Pomarède <npomarede@xxxxxxxxxxxx>:

> Le 27/12/2022 à 15:20, Nicolas Pomarède a écrit :
> > Le 27/12/2022 à 11:49, Troed Sångberg a écrit :  
> >> Hi all,
> >>
> >> Just had a discussion on mono Hatari compatibility on IRC and ran the 
> >> above mono demo. It's already in the compatibility list as;
> >>
> >> "Monochrome. Vertical STNICCC2015 scroll text is missing and some 
> >> parts of the screen have inverted colors"
> >>
> >> (this is actually one and the same issue btw)
> >>
> >> However, both on Windows 64 (official 2.4.1) and on Linux (my own 
> >> build) the demo now crashes/segfaults (no console output) after the 
> >> "Beat Slappaz" logo. This is where some synced hardware scroll should 
> >> appear.
> >>
> >> 1MB STE mono tested, both EmuTOS and TOS 1.62  
> > 
> > Hi
> > 
> > I confirm the crash :
> > 
> > 
> > Thread 1 "hatari" received signal SIGSEGV, Segmentation fault.
> > 0x000000000062217d in Video_CopyScreenLineMono () at 
> > /home/npomarede/src/hatari.git/src/video.c:3425
> > 3425            memcpy(pSTScreen, pVideoRaster, SCREENBYTES_MONOLINE);
> > 
> > 
> > (gdb) bt
> > #0  0x000000000062217d in Video_CopyScreenLineMono () at 
> > /home/npomarede/src/hatari.git/src/video.c:3425
> > #1  Video_EndHBL () at /home/npomarede/src/hatari.git/src/video.c:3180
> > #2  Video_InterruptHandler_HBL () at 
> > /home/npomarede/src/hatari.git/src/video.c:3000
> > 
> > it seems pVideoRaster points to out of buffer address at this point. In 
> > my case it happens when nHBL=87, might differ for others.
> > 
> > 
> > This is related to switching to "small mem" mode by default when 
> > compiling hatari. If I compile with
> > 
> > ./configure --disable-small-mem
> > 
> > then crash goes away.
> > 
> > Thomas, any idea ? maybe the safe "margin" buffer that was allocated at 
> > the end of existing RAM is not big enough ?  
> 
> The video address was in fact not correctly masked in monochrome mode in 
> ST/STE mode. In these cases, MMU should only sees RAM 0 - 0x3fffff, 
> anything above that will be masked with 0x3fffff to keep only the 22 
> lowest bits of the video address.
> 
> This mask was present in color mode since a long time, but not in mono 
> mode, hence the crash when using small mem.
> 
> I pushed a fix for this, hatari doesn't crash anymore in small mem mode.

Thanks, that looks like the correct fix indeed!

 Thomas