Re: [hatari-devel] VME access

[Thomas Huth <th.huth@xxxxxxxxx>]

On 03/01/2021 12.24, Eero Tamminen wrote:
> Hi,
>
> On 1/3/21 12:33 PM, Thorsten Otto wrote:
> > On Sonntag, 3. Januar 2021 10:02:19 CET Thomas Huth wrote:
> > >   Couldn't you please try to come
> > > up with enough VME emulation instead to fix Linux on the TT?
>
> Thomas, it's about interrupt handling and based on
> comments seems to e.g. redirect FDC interrupts:
> https://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k.git/tree/arch/m68k/atari/config.c#n394 
>
>
> VME interrupts are set here:
> https://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k.git/tree/arch/m68k/atari/ataints.c 
>
>
> And up to 16 VME interrupts are supported:
> https://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k.git/tree/arch/m68k/include/asm/atariints.h#n24 
>
>
> I could add some pre-liminary code to setup
> suitable SCU mask etc variables when corresponding
> regs are written...
>
> But I'm unfamiliar how the whole Hatari interrupt
> handling works, and with need to reroute existing
> interrupts to go through SCU that's something
> I'll leave to you or Nicolas.

At least I also don't have any clue about SCU interrupts either, so it's 
learning by doing - if you never try, you'll never get into it :-)

> > To emulate VME,
> > you would need to specify which kind of VME card should be emulated (AFAIK
> > could be either memory expansion, or graphics-card, or network card). That
> > would be a pain.
>
> [1] "--vme-errors" option would go away when
> there's some real VME implementation, but
> something more similar to --dsp option could
> actually be future-proof:
>
> "--vme <x>", where <x> is:
>    "none" - VME reg access causes errors
>    "dummy" - what's currently done, errors disabled
>
> And in future:
>    "empty" -- VME interrupts working, no VME card
>    "net" -- VME network card emulation
>
> Thomas, does that sound any better?

Ok, yes, still not my favorite solution, but looks certainly a little bit 
nicer, so I guess I could live with it. I'd maybe also omit the config file 
saving/loading as long as there is no entry for this in the GUI yet 
(otherwise people might end up with a wrong setting in their config file at 
one point in time and then wonder why things are not working anymore as 
expected when they later run without the --vme parameter).

 Thomas