00:02RSpliet: thanks for reporting though
00:35vpelletier: hi. I'm trying to understand how MSI work in order to understand why it doesn't with nouveau on a powermac G5. I ave working MSI on the same machine under the same pci express root controller, although it is under a PCIe-PCIX bridge (on-board network card)
00:35vpelletier: does the bridge do anything active about MSI ?
00:35vpelletier: like, card would tell the bridge to propagate the MSI
00:36vpelletier: or is it just low-level protocol&speed translation, so it would merely see the device emit a write to an address which would happen to be the pci magic address
00:36vpelletier: s/pci magic/MSI magic/
00:42vpelletier: mmh... actually parent bus is not pci-express, but hypertransport
13:34karolherbst: yeah okay, I guess our best bet is to have some form of GPU crash dump whenever we detect a hang inside gr, aka when we timeout inside gk104_fifo_gpfifo_kick
13:41karolherbst1: nice, and nvidia brings down the entire machine if there is an infinite loop
13:43karolherbst: but it seems like that nvidia dumps something in dmesg: https://gist.githubusercontent.com/karolherbst/2d4946a1f244a9ed001f61938951d6e1/raw/76ce3fc268f69fc9e2602cacc33595a212e20283/gistfile1.txt
13:55karolherbst: seems like it just dumps most of PGRAPH
14:20karolherbst: mwk: did you at any time looked into dumping the states of the gpu cores?
14:58karolherbst: that linux debugger doesn't work for me :/ It is able to connect and launch processes, but it won't attach to those :(
15:05karolherbst: hakzsam: how did you manage to get it working on your local machine?
15:34karolherbst: orbea: I think the biggest issue regarding ubershaders is that we don't do optimized recompilations if we have proper knowledge about inputs/outputs
15:34karolherbst: RSpliet: scheduling won't help much here
15:37karolherbst: it is expected to get perf hits with ubershaders anyway
15:37karolherbst: the main concern is to reduce recompilation and reduce CPU overhead
16:05karolherbst: skeggsb_: I've resend my clock updates patches without those two commits extracting reclocking things into chipset specific files, so that indeed only issues on prime based systems are resolved and nothing more
22:18martinus: Hi everyone, is this the correct channel to ask for troubleshooting help?