Re: Improved memory detection for cgsix

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Thu, 3 Mar 2022 17:38:03 +0100
Romain Dolbeau <romain%dolbeau.org@localhost> wrote:

> Le jeu. 3 mars 2022 à 01:59, Michael <macallan%netbsd.org@localhost> a écrit :
> > On my 501-2253 I got an undisturbed red square and next to it a blue
> > one with an uninitialized memory pattern around it, so we didn't wrap
> > and we didn't hit unmapped space either, which means the blitter can
> > indeed see the full 4MB.  
> 
> I can confirm on the 501-2253 in my SS20, a similar set of patches (I just
> used vmsize*1024*1024 rather than fbmapped in the console driver) works
> fine. kernel and console see the 4 MiB, and X11 doesn't seem to have any
> issue having the extra offscreen area (visible in the log). Not sure how to
> confirm it's actually used for anything, though.

Maybe fbmapped is supposed to be the amount mapped by OBP? But if
that's the case, why would they map more than the visible screen?
As for X11, the driver will likely use up off-screen memory pretty
quick since cg6 doesn't support variable strides for source,
destination etc., so all off-screen pixmaps use the full fb width.
Works, but wastes a lot of VRAM.
You'd see things like distorted icons or fonts if you run past the end
of your vram or wrap around, since that's what exa is most likely to
stash away in off-screen memory for repeated use.
As for checking if it's used - just xf86Msg() something in Cg6Copy() if
ySrc + pCg6->srcoff is over 2000. If you see the messages in your log
but no garbage on screen, you're good.

have fun
Michael