I have tried to get several versions of grfcl to work in my setup, but
so far I have mainly been able to stare at a black screen. I have
tested precompiled kernels from:

ftp://ftp-uni-regensburg.de/pub/NetBSD/experimental/

and kernels compiled by myself with grfcl from the first, second and
third cirrus package put together by Markus Illenseer and Ezra Story's
ezygrfcl and cirrus-current packages. And now with the grf_cl which is
integrated into -current. The result with X11 has always been black
screen. I have tried the X servers Xamiga24, Xcl, Xamiga24-950305 and
Xcl-950408.

The cirrus-3 package has comments that the Picasso has some kind of
ZorroII to ISA translation which causes black screen problems. But why
would this only be relevant with a 040? It seems that I don't really
understand what the comments are referring to as they claim that the
problem could be solved by adding 0xffff to some register addresses.
In my limited understanding this would only lead to accessing wrong
addresses. The whole register block is 64k, isn't it?

Yesterday I was able to at least produce something else on the screen
with the program gfxtest.c from cirrus-current package. It seems that
it sometimes fails to mmap() the frame buffer. But anyway when I
enabled the following lines:

#if 1
    *(ba + 0x3c6) = 0xff;
    *(ba + 0x3c6);
    *(ba + 0x3c6);
    *(ba + 0x3c6);
    *(ba + 0x3c6);
    *(ba + 0x3c6) = 0xd0;
#endif
    
I was at least able to see something. Without those lines the screen
remained white. So is this piece of code accessing the hidden DAC
register, which is referred to in grf_clreg.h and why would this make
any difference?

Are the sources for the various X servers available from somewhere?