00:54 hakzsam: imirkin_, could you try this patch and tell me if it's fix the crash on your GK208, please? http://hastebin.com/nitolipuma
09:06 BenBE: I have an issue with nouveau (GF106 with NVC3 for Thinkpad W520/Optimus) on Ubuntu 15.04 + linux-image 4.0.1 (kernel-ppa mainline) causing X server to fail to start. Using linux-image 3.19 works.
09:07 BenBE: Issue reported in dmesg is "Invalid ROM contents" followed by DRM failed to create 0x0 (-22).
09:12 imirkin_: BenBE: pastebin the dmesg
09:13 imirkin_: chances are we're poo-pooing the "real" version due to some sort of checksum fail and then reading the totally wrong version from the ROM image.
09:23 BenBE: imirkin_: http://pastebin.com/1w2MLddK
09:23 imirkin_: uhm... wtf
09:23 imirkin_: not what i was expecting to see
09:23 imirkin_: oh, i wonder if this is the ACPI fail a bunch of people were having
09:24 imirkin_: i think it's fixed in 4.1-rc1, or alternatively downgrade to 3.18
09:26 BenBE: Let me check ... (Testing with http://kernel.ubuntu.com/~kernel-ppa/mainline/v4.1-rc2-vivid/ )
09:27 imirkin_: if it's the thing i'm thinking of, should def be fixed in there
09:36 BenBE: imirkin_: Seems fixed in the version I mentioned.
09:36 imirkin_: cool!
09:39 BenBE: Second question: Do you know of issues that might cause the following behaviours? 1) When using an external monitor the image is lagging behind the real content and reveals what WAS there) 2) Trying to move an nVIDIA screen to an offset != 0 gives a failure to set thatgraphics mode)
09:39 BenBE: At least reproduceable til 3.16, not tested since if still persists.
09:40 imirkin_: i'm guessing you're using optimus?
09:40 imirkin_: a lot of people see the lagging issue, but a lot of people don't
09:40 imirkin_: i think it has to do with the specific compositor setup
09:40 imirkin_: no one's narrowed it down afaik
09:41 BenBE: yes, bumblebee-kinda setup.
09:41 imirkin_: note that with optimus, you get a software cursor, so.. that's a bit annoying
09:41 BenBE: Using OpenBox (LXDE)
09:41 imirkin_: i'd strongly advise avoiding bumblebee in the presence of nouveau
09:41 imirkin_: nouveau will auto-suspend the gpu when possible on its own
09:41 BenBE: k.
09:42 imirkin_: as for the offset issue... i think that's a known problem
09:43 imirkin_: i definitely know you (effectively) can't have multiple "reverse prime" screens
09:43 BenBE: Any known workaround yet?
09:43 imirkin_: not to my knowledge
09:43 imirkin_: it's not a nouveau issue though
09:43 BenBE: :(
09:44 imirkin_: it's an xorg one
09:44 imirkin_: i think there's very little appetite in the developer community to deal with optimus
09:45 imirkin_: really only one or two people have really invested time into it, and that time is behind them
09:46 BenBE: k. Thanks anyway for the help.
09:48 hakzsam: imirkin_, does my patch fix your issue?
09:48 imirkin_: dunno
09:48 imirkin_: i'll test it, but haven't had a chance yet
09:48 hakzsam: np
10:51 imirkin_: hakzsam: i'm sure that patch will work, but it seems liek the wrong approach
10:51 imirkin_: why not just test for nvc0->compute up front?
10:51 imirkin_: instead you've now encoded the knowledge of when nvc0->compute is set in a bunch of different places
11:48 hakzsam: imirkin_, I'm not sure it's a better approach. I did this way because nvf0 is really different than nve4, and in case of nvf0 compute support is added one day, MP counters will be exposed because nvc0->compute will be true. But these counters will be wrong since they are quite different.
11:49 imirkin_: hakzsam: so then maybe just == NVE4_CLASS?
11:49 imirkin_: but ok, point taken
11:50 hakzsam: yes, just == NVE4_CLASS will be enough
11:50 imirkin_: however i'd still rather it be surrounded by an if (compute) { if (class >= nve4 and < nvf0) do x else if (class > etc
11:51 imirkin_: you could even have a clause for a hypothetical nvf0 impl with a /* TODO */ style comment
11:51 hakzsam: yeah, to make sure we instantiate a compute thing, I'm okay with that
11:52 imirkin_: and eventually we'll have compute enabled everywhere
11:52 imirkin_: and we'll be able to remove the surrounding if or something
11:52 hakzsam: sounds good, I'm going to update the patch
12:18 hakzsam: imirkin_, http://paste.awesom.eu/qpsg looks good?
12:19 imirkin_: hakzsam: yeah, much better. let me test it out...
12:20 hakzsam: okay
12:20 imirkin_: hrmph. patch apply fail.
12:21 imirkin_: oh. stupid html
12:21 imirkin_: can you make an actual patch version
12:21 hakzsam: oh! sorry
12:21 imirkin_: one that doesn't do
12:21 imirkin_: + if (screen-&gt;base.class_3d == NVE4_3D_CLASS) {
12:21 imirkin_: :)
12:22 imirkin_: hastebin should work fine for that
12:22 hakzsam: http://hastebin.com/raw/ozavahodej
12:22 imirkin_: or even, *gasp*, pastebin
12:22 imirkin_: much better :)
12:22 hakzsam: yes, I installed at work, just need to update my dotfiles repo at home :)
12:24 imirkin_: ok, looks like i get a skip now. is that expected?
12:24 imirkin_: i.e. AMD_performance_monitor is no longer exposed for me
12:25 hakzsam: yes, normal behaviour. AMD_performance_monitor is only enabled if the underlying driver exposes GPU counters
12:25 imirkin_: kk
12:25 imirkin_: just checking :)
12:25 hakzsam: yeah :)
12:25 imirkin_: anyways, seems like it works as expected on GK208. should probably double-check a kepler1 but... yay :)
12:26 hakzsam: I don't have a kepler, but I'm going to check on Fermi before sending the patch to mesa-dev
12:26 hakzsam: just to be sure :)
12:26 imirkin_: sounds good. i have a fermi at home.
12:26 imirkin_: perhaps i'll figure out why you had to disable compute by default
12:29 hakzsam: looks fine on my NVD9
12:29 imirkin_: great
12:29 imirkin_: let me know if you want additional info to RE the counters on my GK208
12:30 hakzsam: okay, thanks
12:32 hakzsam: patch sent to the ML
12:43 hakzsam: imirkin_, could you just run this script on your GK208 for checking bitfield of muxs, please ? http://hastebin.com/raw/uparuyevuc
12:43 imirkin_: hakzsam: ftr, NVASCAN ?= nvascan should do what youw ant
12:43 imirkin_: did you want me to do this with nouveau or blob?
12:44 tobijk: hakzam: imho that patch is fine :)
12:44 hakzsam: imirkin_, up to you :)
12:44 hakzsam: tobijk, thks
12:45 imirkin_: er hm, the ?= is a make thing, not a bash thing
12:46 imirkin_: gr, how do you do this in bash...
12:46 imirkin_: hakzsam: http://hastebin.com/tagocuzata.sm
12:46 hakzsam: thank you imirkin_
12:48 imirkin_: hakzsam: NVASCAN=${NVASCAN:-nvascan}
12:48 imirkin_: er, probably with quotes
12:49 imirkin_: btw, remember that GK208 is one of the weakest cards of the series, while GK110 is one of the most powerful. could well be that they have differences
12:50 hakzsam: I didn't know for both
12:55 imirkin_: and they both use SM35 (vs SM30). the only other card that uses SM35 is GK20A (TK1)
12:56 hakzsam: okay
12:57 imirkin_: (there's also a NV106 which is in the GK208 pci id range, unclear how to report that one... iirc it has the same classes)
12:57 hakzsam: global performance counters seem to be quite similar to GK104, btw
12:57 hakzsam: that's a good news for me :)
13:00 hakzsam: just in case, do you have a GF108 plugged right now?
13:00 imirkin_: yeah, but it's about 1.5km away, so out of reach ;)
13:00 hakzsam: hehe, np :)