On 2017-03-20 22:18:16 [+0100], To Wargreen wrote:

1993 wants exclusive_lock and blocks on it.
2230 has exclusive_lock and blocks on mclk_lock.
2448 has mclk_lock nad reservation_ww_class_mutex. I *think* that one
waits for an IRQ from the GPU which is not coming. My guess is, that it
sits somewhere in ttm_bo_vm_fault_idle() and waits for wakeup.
If this triggers so easy, could you please check if the interrupt
counter for radeon in /proc/interrupts increases after the accident?

The other theory is that the code can't continue because we can't have
multiple reader of exclusive_lock. I will try to verify that but this
will take some time…

Sebastian