> This may be all moot, but...
> 
> Wouldn't it be possible to lazily do all operations in some sort of
> bank cache.  When a fault occurs in a bank not cached, save the
> previously mapped bank, if modified, in the cache, schedule the bank
> for update later and map the copy instead of the bank being lost.  Then
> map the new bank.  Maintain a modified bit for the banks.  Scheduled
> banks for update could be handled at the next vbl or maybe even a
> vbl later than that.

I don't think -even if this approach would work- it's even worth the effort
thinking about it. There are PC servers (XFree comes to mind) that are
specifically built and optimized to deal with VGA chips that use banked
addressing. I even think the retina server from our french friends was
built from such a PC server. It would make much more sense to get such
a server running on RetinaZ2, instead of going all the way to make kernel-level
bank-switching sort of efficient...

> However, I have seen a real bug I think.  The Z2 Retina seems only to
> allow 16K of chars sized screens.  Shouldn't it allow 32K?  If I change
> to a 158x126 mode I've done the lower part of the screen (approx after
> 16K) contains the test pattern put out in rt_load_mon.  My guess is that
> a signed short is used where an unsigned should be, but I haven't had
> the time to look deep enough to find this.

I think to remember that the chips text mode only works in certain display
modes (due to increased bus load in this mode compared to the corresponding
graphics mode). I think it's for example not possible to run text mode
at anything larger than 1024x768.

> Another problem I've had is that for some time the Xamiga+retina haven't

That's the french server, right?

-Markus
-- 
EUnet Switzerland   Tel:     +41 1 291 45 80	Markus Wild
Zweierstrasse 35    Hotline: +41 1 291 45 60	mw@eunet.ch
CH-8004 Zuerich	    Fax:     +41 1 291 46 42	S=mw;P=EUnet;A=EUnet;C=CH
>Solaris 2 is not an upgrade from Solaris 1. They just want you to THINK it is.