> > > Just in case it's a similar problem with radeonfb :-) > > > > No, radeonfb uses the PCI stuff nowadays, but I'll look into fixing > > atyfb ;) > > It's worse than I originally assumed: the kernel hangs in > aty_power_mgmt(). The info and par pointers are OK.
Update: compiling the kernel without preemption seems to fix it. What's
needed to make critical sections in the framebuffer code preemption safe?
(local_irq_save() didn't help, FWIW)
Michael
