08:10 <lynxeye> bl4ckb0ne: I don't have any real good ideas anymore. Especially since the address is the zero page, so it doesn't seem to be related to something real set up via kernel or user commandstreams. One shot in the dark might be to stuff all the known VIVM states from rnndb with invalid addresses and see if the fault then hits one of those addresses instead of the zero page. Maybe that gives a hint as to which unit is causing the fault.

12:56 <bl4ckb0ne> good idea, ill try that this morning thanks

12:57 <bl4ckb0ne> we've been comparing streams with ManMower, there's a few things that differs regarding flush cache but we have not been lucky yet

12:58 <bl4ckb0ne> i tried poking around the userland driver and I got the same address in the DMA addr register twice with different command streams so I think that mesa is alright

12:59 <bl4ckb0ne> i have some suspicions regarding the "stallFEprefetch" stuff but nothing concrete

13:42 <bl4ckb0ne> any idea what 0xf is in `TOKEN FROM=FE,TO=0xf,UNK28=0x0` ?

13:42 <bl4ckb0ne> we saw it a lot around `0x68000001, /* CHIP_SELECT (13) `