Re: RadeonFB [Re: 2.4.23pre8 - ACPI Kernel Panic on boot]

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

> Yes, I can reproduce without the tainting kernel module (fglrx.o). 
> Debugging a bit, the corruption shows up when using ATI's "fglr" device
> driver in XF86Config.  If I switch to using the unaccelerated "Radeon"
> driver provided by XFree, then the screen remains readable upon return
> from X.  Thus, the kernel module is OK, the ATI proprietary glx and dri
> modules are OK, it's just the driver.
> 
> However, I still think this belongs as a kernel issue, because it
> appears to be a failing of the radeonfb.o module to properly set the
> radeon registers.  Remember that version 0.1.4 of the driver does not
> have this issue; it works just fine with the ATI fglr X driver.

Ok, first thing: XFree "radeon" _is_ accelerated, though it doesn't
do 3D on recent cards

Then, the problem you are having is well known and I don't have a
simple workaround at hand other than re-initing the engine on each
console switch, which I would have very much liked to avoid... maybe
I can add an fbdev hook to the switch from KD_GRAPHICS to KD_TEXT
which is what we really want to cleanup the engine.

Can you verify that running fbset -accel 0 then back 1 cures the
problem for you ?

Ben.