X unkillable in R state sometimes on startx , /proc/sysrq-trigger T output attached
Two months ago I started a thread
BUG: fb_imageblit called before fb_check_var and fb_set_par function
in the Linux-fbdev-devel mailing list. I found that the accelerated imageblit function of a framebuffer driver might be called before the graphics engine is initialized ... normally that happens in the fb_set_par function. For cyblafb I solved the problem by extending the fb_sync function to include a call to the graphics engine init function after a short timeout, but the problem is still present in all recent kernels. It might be argued that this is not a kernel bug but a problem of X - have a look at the Linux-fbdevel thread.
Does X start reliably without a linux framebuffer driver? Does X start reliably with vesafb?
If the answer is "yes", then have a look at the radeonfb sync function. After a short timeout, assume that an erroneous X driver disabled mmio, so (re)enable mmio and (re)init the graphics engine.
cu,
knut
Quoted 10 lines
BUG: fb_imageblit called before fb_check_var and fb_set_par functionpresent in all recent kernels. It might be argued that this is not a kernel bug
but a problem of X - have a look at the Linux-fbdevel thread.Does X start reliably without a linux framebuffer driver? Does X start reliably with vesafb?
Hmm I missed the fact that you are using radeonfb, this could point to the X chipset initialisation code as well...
I'll try and track down if the evil patch made it into xorg in FC4..
Dave.