00:34imirkin: mwk: trying to get things perfect before the last known card dies? :)
00:35hakzsam: mupuf, done, gm206 now :)
00:35hakzsam: it took almost 2h on gf100
01:06imirkin:is getting life is strange, will see if it works on nouveau
02:48Keringar: Does anyone have any ideas on why I can't boot with nouveau unless I set nofbaccel to 1 in my kernel parameters?
02:49Keringar: Actually it's probably something to do with maxwell
02:51imirkin: Keringar: which gpu do you have, and which kernel?
02:52Keringar: I have a EVGA 970 SSC and im on 4.6.4
02:54imirkin: do you have the nvidia-supplied firmware installed for acceleration?
02:54imirkin: i.e. /lib/firmware/nvidia/gm204/...
02:55Keringar: probably not
02:55Keringar: Would I get those from the nvidia drivers?
02:56Keringar: Ill go check, brb
02:56imirkin: no, it's in linux-firmware
02:57Keringar: I get messages like invalid pointer to flat table and unknown chipset so I'm probably missing something. But hold on a moment I gotta go check
02:59imirkin: Keringar: your full dmesg will probably contain interesting info
03:00imirkin: invalid pointer to flat panel table is harmless
03:00imirkin: unknown chipset is fatal
03:06Keringar: Alright ignore how it's in a git repository but here's my dmesg log
03:11imirkin: that all seems perfectly happy
03:12imirkin: what happens if you don't have the nofbaccel=1 in there?
03:13imirkin: (it probably tries to bring up gr, and something fails in secboot?)
03:13Keringar: I get a black screen partway through booting
03:13Keringar: I'll try taking it off and I'll post it again hopefully
03:15imirkin: what's in dmesg?
03:18Keringar: I'm trying to get it still it's hard because I can't see anything. Is there a way to get my previous boot logs?
03:20Keringar: My screen goes black right after starting kernel device manager
03:20imirkin: i dunno... mine end up in /var/log/messages
03:28s0be: Lekensteyn, ran the acpi patches fro a full day with no issues, +1 ship it.
03:28Keringar: I could get precise enough so this is for all of today https://raw.githubusercontent.com/Keringar/OpenGL/master/oldBoot
03:30imirkin: Keringar: looks like you're having the same issue a few other GTX 970 owners are seeing
03:30imirkin: Keringar: https://bugs.freedesktop.org/show_bug.cgi?id=94990
03:30Keringar: Do you guys need anything from me to help?
03:31imirkin: unfortunately these are super-locked-down gpu's
03:31imirkin: so they're basically about as useful as bricks =/
03:31imirkin: anyways... perhaps gnurou (who is an nvidia employee and contributed the secboot logic to nouveau) can take a look
03:32imirkin: gnurou: looks like a lot of people with GTX 970's are having issues with secboot.
03:32imirkin: gnurou: dunno if GTX 970's are more popular and that's why more people are showing up with issues, or if there's something GTX 970-specific
03:33Keringar: Thanks for your time. Ill just hope that something changes
03:33imirkin: i think that there were some patches that went into 4.7-rc or perhaps are on their way to 4.8
03:33imirkin: which improve the secboot situation
03:33imirkin: however it's unclear whether that's helped others in your predicament
03:35Keringar: Well I will hope for the best
04:55gnurou: imirkin: I have tried booting with a GTX970, and found no issue so far
04:56skeggsb: it might be useful to get a mmiotrace (of the binary driver, perhaps nouveau too) on both gnurou's working board, and someone's that isn't
04:56gnurou: the boot issue when the firmware is not present should be fixed in 4.7
05:10Keringar: How do I do that mmiotrace cause I have a non working one
05:16skeggsb: Keringar: http://dpaste.com/2FD3K38
05:16skeggsb: can you apply that, and retry? i want to see which buffer that fault corresponds to, which might help gnurou track down what's wrong
05:25Keringar: Could you give me a few pointers on applying this. Do I need to compile nouveau from the form rep
05:25Keringar: *drm repository?
05:36Keringar: Never mind I think I figured it out
06:53Keringar: skeggsb: I'm currently compiling the kernel patch I'll get back to you tomorrow
08:56karolherbst_work: update to the reclocking patches: ben wants to merge those after I fixed some smaller issues. I am sure I will be able to finish this today
08:57karolherbst_work: mhh, up until a specific point, somehow I missed that part .D
09:07hakzsam: mupuf, https://people.freedesktop.org/~hakzsam/piglit/maxwell/
09:14hakzsam: mupuf, the gf100 is quite fine btw :)
09:34mupuf: hakzsam: can't you push the results yourself? :D
09:36mupuf: doing it now
09:39mupuf: pushed the 3 runs
09:39mupuf: will take time to re-generate the report
09:40hakzsam: I did only 2 :)
09:40hakzsam: gf100 and gm206
09:40mupuf: there was a new one for gm207
09:40hakzsam: ah right
09:56mupuf: http://piglit.mupuf.org/results/fermi/html/problems.html and http://piglit.mupuf.org/results/maxwell/html/problems.html
09:57hakzsam: seems fine
09:57mupuf: well, one gotta love nvidia for not breaking things inside families
09:58mupuf: because AFAIK, the GM2xx has barely been tested
09:58mupuf: it may even be one of the first piglit runs on it
09:58hakzsam: probably yeah :)
09:59mupuf: did oyu fix ext_framebuffer_multisample?
09:59mupuf: http://piglit.mupuf.org/results/maxwell/html/regressions.html <-- this view is quite interesting
10:00mupuf: well, let's wait a little more and see what you did fix on the GM107
10:01hakzsam: or what I did break ;)
10:15hakzsam: we fixed a few number of tests since mesa 12
10:15hakzsam: the other ones are known :)
10:16mupuf: 35 minutes to generate all the reports
12:26Misanthropos: karolherbst_work, i have applied the 4.6_recklocking branch to kernel 4.6.4 and get theses messages: [12537.255832] nouveau 0000:04:00.0: clk: failed to set cstate 160
12:26Misanthropos: [12537.255838] nouveau 0000:04:00.0: clk: error setting cstate 160: -22
12:26Misanthropos: [12538.255845] nouveau 0000:04:00.0: clk: failed to set cstate 160
12:26Misanthropos: [12538.255850] nouveau 0000:04:00.0: clk: error setting cstate 160: -22
12:26Misanthropos: i have the gk106 card with strange bios.
12:28karolherbst_work: yeah, seems that way
12:28karolherbst_work: did you give it to me at one point? I can't check from here if I have it
12:31drathir: Lekensteyn: thanks a lot, only wanna confirm if blacklist driver will prevent autoswitching to dedicated gpu, bc gettn some issues with intel...
12:32Misanthropos: karolherbst_work, yes. you have the bios and you fixed it for 4.5
12:32Misanthropos: or rather 4.4
12:35karolherbst_work: ahh I fxied it already, so I broke it again
12:36karolherbst_work: will have a look after I come homw
12:48Tomin: Hi, I don't know if this is about Mesa or kernel or both, but how do I check what clocks my graphics are running on? I've got GTX 460
12:49Tomin: hmm /sys/kernel/debug/dri/0/pstate gives me a list of frequencies and one of them is listed twice
12:50Tomin: prefixed with AC, does that stand for active?
12:50karolherbst_work: Tomin: the last one is the active one
12:51karolherbst_work: with information about if you run on an AC adapter or battery
12:51karolherbst_work: Tomin: 460 is fermi right? You can't reclock fermi with nouveau, yet
12:52Tomin: karolherbst_work: yes, I know I can't reclock it, I just wanted to know what clocks it runs on
12:52karolherbst_work: I see, well the last line tells you
12:52Tomin: core 50 MHz memory 135 MHz, no wonder the performance is abysmal
12:53Tomin: yeah, thanks
12:53karolherbst_work: fermi cards are odd anyway
12:53Tomin: what do you mean?
12:53karolherbst_work: I have a kepler, and the default clocks are 405MHz / 810 MHz :p
16:07imirkin_: mupuf: i played around with that life is whatever newly-free game on steam yesterday. seems that display hangs randomly with the game going into an infinite loop somewhere :(
16:07imirkin_: it happens semi-randomly, after a few minutes of gameplay
16:07mupuf: hmm, then it is nouveau-specific
16:07mupuf: any other bug like this one on any other UE4 demo?
16:08imirkin_: only time i've seen stuff like that is extremely-long loops in nouveau waiting for fences to time out that never do due to parallelism. but that was not the case here - every thread was somewhere in the lib, waiting on a lock or otherwise.
16:09imirkin_: s/in the lib/in the game/
16:09mupuf: worked fine on i965
16:09mupuf: Fuck! I am two hours late! Gotta go!
16:24hakzsam: well, F1 2015 hangs the GPU as well :)
16:24hakzsam: and we fail at compiling one shader in grid autosport
16:25imirkin_: right, but in this case, no gpu-side issue
16:25imirkin_: the game just hangs somewhere in its own code
16:25imirkin_: at least ... iirc
16:25imirkin_: i should triple-check
16:25hakzsam: which one?
16:25imirkin_: the one that came out yesterday and is free
16:25imirkin_: life is ____
16:25imirkin_: interesting maybe? i forget
16:25hakzsam: strange :)
16:26hakzsam: at some point, those games need to be fixed anyways
16:26imirkin_: i'm not arguing :)
16:27imirkin_: but i'm also not fixing
16:27imirkin_: i forgot about that shader issue with grid
16:27imirkin_: i filed a bug, i hope
16:27hakzsam: I will have a look once maxwell has GL 4.3
16:27hakzsam: yeah, I think too
16:27imirkin_: it was pretty straightforward to repro
16:27imirkin_: but iirc i had no clue what was going wrong
16:28hakzsam: yeah, easy to reproduce but hard to fix
16:28imirkin_: my genius solution was "don't enable that" :)
16:32hakzsam: imirkin_, I have just forced gl43 on top of my tess branch, piglit is running and it appears to not break the universe :)
16:34hakzsam: I have disabled arb_shader_image_load_store-atomicity though because that one totally hangs the gpu...
16:34imirkin_: oh right. the scheduler thing.
16:34hakzsam: yeah, the race condition with suatom
16:34imirkin_: just stick the max wait cycles on all atom ops
16:34imirkin_: should be easy to do
16:34imirkin_: but it's on all ops AFTER an an atom op =/
16:34imirkin_: ok, that requires a teeny bit of work. not a lot.
16:35hakzsam: wll do after tess ;)
16:36hakzsam: and I will try to set up x11vnc on reator for testing heaven+tess on maxwell
16:37imirkin_: should be easy
16:37imirkin_: just run x11vnc -server :1
16:37imirkin_: or whatever
16:37imirkin_: and then you can connect using any vnc program
16:38imirkin_: [you have to be a little careful with port forwarding and whatnot]
16:41Lekensteyn: drathir: blacklisting nouveau will indeed Xorg from using the dedicated GPU but will also prevent autosuspend from working so your laptop runs hotter and the battery life gets worse
16:49hakzsam: imirkin_, well, it seems like some fails are unexpectedwith gl43 on gm206
16:49imirkin_: care to share?
16:49hakzsam: the run is not done yet
16:52hakzsam: [ 5083.949878] nouveau 0000:01:00.0: ce2: LAUNCHERR 00000003 [MEM2MEM_RECT_OUT_OF_BOUNDS]
16:53hakzsam: or it's very sloooow
16:53hakzsam: I will wait before killing the process
16:55imirkin_: which test?
16:57hakzsam: recovered :)
17:01hakzsam: oh nice! actually no regressions, except arb_shader_image_load_store-max-size which somehow hangs the gpu :
17:02hakzsam: I will investigate
17:02hakzsam: so GL 4.3 for maxwell is almost done
17:04hakzsam: imirkin_, btw, arb_shader_image_load_store-max-size works fine on gm107...
17:06hakzsam: I will make a mmt
17:07imirkin_: all precipitated by the RECT_OUT_OF_BOUNDS
17:07hakzsam: maybe something has changed between gm107 and gm206
17:07imirkin_: maybe m2mf has diff limits?
17:07imirkin_: or maybe they're better enforced on gm20x?
17:07hakzsam: no clue
17:08hakzsam: imirkin_, you might be interested with http://piglit.mupuf.org/results/maxwell/html/problems.html
17:09imirkin_: i'd actually be a lot more interested in a single combined one
17:09imirkin_: wtf is all this MS bullshit
17:09imirkin_: kill it with fire.
17:10dcomp: would a piglet run on my special gm108 be any use?
17:10imirkin_: the gm107 run must have been at a newer commit
17:10imirkin_: oh - it was. ok.
17:10hakzsam: imirkin_, yes, more recent
17:11imirkin_: hakzsam: but without your shader_draw_parameters fixes to the MME?
17:11hakzsam: dcomp, I would say that gm108 is same as gm107
17:11hakzsam: imirkin_, yes, but how this can be related?
17:11imirkin_: hakzsam: it's related to drawid-indirect failing :)
17:12hakzsam: imirkin_, ah yeah, those ones :)
17:12hakzsam: same for gettextimage
17:12imirkin_: well, those i fixed recently
17:12hakzsam: so except this ms bs, no differences
17:13hakzsam: which is fine by me
17:13imirkin_: still would be interesting to compare it to kepler
17:13hakzsam: I did
17:13hakzsam: but I don't remember
17:13hakzsam: because I generated like 15 summaries today :)
17:13hakzsam: can't remember all differences between all cards :p
17:14imirkin_: should generate one with 1 nvc0, 1 nve0, and 1 nv110
17:14hakzsam: mupuf, ^
17:14imirkin_: (*not* all of them)
17:14imirkin_: just the latest in each category. but fermi+. not including tesla.
17:15hakzsam: but iirc I don't remember any regressions between gm107 and gk107
17:15hakzsam: with mesa 12
17:15imirkin_: this is probably bad and a sign of sadness: http://piglit.mupuf.org/results/maxwell/html/gm206-mesa-12.0.0/spec@arb_framebuffer_no_attachments@arb_framebuffer_no_attachments-atomic.html
17:16imirkin_: i'm guessing this has to do with conservative rasterization
17:16imirkin_: maybe it's enabled by default
17:16hakzsam: no idea
17:16imirkin_: iirc someone sent me a trace from which i documented which bit it was
17:17imirkin_: hakzsam: https://github.com/envytools/envytools/commit/e5e4cea7ada54e954baa336559c28b40f4b8d8a4
17:17imirkin_: hakzsam: try sticking a 0 in there and see if that fixes anything
17:17imirkin_: ideally we'd hook up the NV_conservative_raster ext
17:17imirkin_: but we're not perfect people
17:18imirkin_: Leftmost: actually that'd be a fairly easy thing to do if you're interested.
17:18imirkin_: Leftmost: it's GM20x+ only, but i think that's what you have?
17:18hakzsam: feel free to do it if you want
17:18Leftmost: I'll take a closer look in a bit. In a meeting at the moment.
17:19imirkin_: Leftmost: i'm guessing when it all shakes out, that ext will be no more than +100 lines of code in various parts of mesa
17:19imirkin_: Leftmost: but it'll be a lot of little things
17:21imirkin_: Leftmost: you'll have to trace where SUBPIXEL_PRECISION_BIAS_X_BITS_NV get set - i don't think the trace i have had that
17:22imirkin_: Leftmost: you can find the traces here: https://people.freedesktop.org/~imirkin/traces/gm206/
17:23imirkin_: (no clue how useful that ext is or who would use it, esp on nouveau, but just pointing it out.)
17:25Leftmost: Even so, it's an opportunity to become a bit more familiar with the code and process.
17:25imirkin_: if you want a set of patches to "copy", have a look at my patchset which implemented EXT_polygon_offset_clamp
17:25imirkin_: it's a very similar extension in many regards
17:32Keringar: skeggsb: Here's the boot log with that patch applied https://raw.githubusercontent.com/Keringar/OpenGL/master/nouveauPatchBootLog
17:34imirkin_: Keringar: how sure are you that you loaded the right module?
17:34imirkin_: none of ben's prints appear to be there
17:35imirkin_: oh wait, nm
17:35imirkin_: they're there
17:35hakzsam: imirkin_, sticking 0 in CONSERVATIVE_RASTER doesn't fx anything
17:35imirkin_: hakzsam: o well.
17:35imirkin_: hakzsam: how about 1? :)
17:35imirkin_: maybe it's DISABLE_CONSERVATIVE_RASTER and i messed up
17:36imirkin_: hakzsam: and i'm specifically talking about that no-fb-attachments thing
17:36imirkin_: now to see how that test works
17:37imirkin_: oh interesting
17:37imirkin_: so the issue there is that the seam *doesn't* get double-drawn
17:39imirkin_: whereas the test assumes it does. (iirc i wrote that bit of it... hm...)
17:41imirkin_: gnurou: btw, see Keringar's log above. it has ben's hackpatch to print some addresses - http://dpaste.com/2FD3K38
17:42imirkin_: gnurou: looks like the PMU is having trouble accessing the WPR, whatever that is
17:43imirkin_: and then things go south from there
17:46karolherbst: Misanthropos: are you sure you use my lates tree
17:48karolherbst: ohh I think I know what the issue is
17:50imirkin_: hakzsam: based on the ARB_texture_multisample fails, i'd guess that MS is just fubar'd on GM20x
17:52imirkin_: hakzsam: and that's why the seam doesn't get double-counted -- because MS isn't actually turned on, in which case there is no double-counting of the seam.
17:52imirkin_: (because each pixel of the seam goes to either one or the other triangle of the quad)
17:52imirkin_: (whereas with MS, each triangle hits *some* samples of each pixel in the seam, and so the seam is double-counted)
18:00rhn: does valgrind-mmt follow pointers in CREATE ioctls just as it does in CALL ioctls?
18:01imirkin_: mslusarz: --^
18:02rhn: judging by demmt code, it does, but I can't see the code on valgrind side
18:03imirkin_: perhaps it should but doesn't? :)
18:05rhn: no, it definitely does
18:06imirkin_: i mean perhaps mmt should but doesn't
18:06rhn: it works with other calls, just not mine, so it's only my fault
18:06rhn: as soon as that's sorted out, I'll have a useful patch for you guys
19:07rhn: some guesses about undocumented objects: valgrind-mmt: http://wklej.org/id/2770998/ envytools: http://wklej.org/id/2771000/
19:22thomas_sch: I got a GeForce GTX 560 SE and neither the nvidia nor the nouveau driver can drive all three connectors. Is this a hardware limitation?
19:23imirkin_: thomas_sch: yes - only 2 CRTC's for pre-kepler hardware
19:23imirkin_: in theory, the hw might support driving all 3 connectors if 2 of them were mirrored, but i doubt nouveau or the blob would properly support that
19:25thomas_sch: so it's a limitation, because of the analog signal I use from my dvi-connector?
19:25thomas_sch: (with CRTC you mean the analog signal or do you mean any video output?)
19:25thomas_sch: oh ok sorry
19:25imirkin_: CRTC = scanout engine
19:25imirkin_: aka thing that reads pixels out of video memory
19:26thomas_sch: ahh ok
19:26urjaman: named in the 80s originally i guess :P
19:26thomas_sch: that's sad, but I can't do anything about that hmm thank you for the information
19:26imirkin_: you could buy a new amd gpu
19:26thomas_sch: didin't even know the CRTC term
19:26imirkin_: some of those have up to 6 CRTC's :)
19:29thomas_sch: if I wanted multi monitor and nothing else I would buy a matrox card :P
19:29imirkin_: amd gpu's have the best open-source support...
19:29thomas_sch: weird to see an amd recommdation in the nouveau channel :P
19:30hakzsam: imirkin_, okay, maybe we can add a card on trello? because I don't plan to look at this now
19:30imirkin_: thomas_sch: that should give you an indication of the situation with nvidia gpu's on linux :)
19:30imirkin_: hakzsam: ok
19:31hakzsam: imirkin_, can you add it please?
19:31thomas_sch: I prefer nvidia nonetheless, since if I need I can get at least a decent power usage/performance if I use the blob (had really bad experiences with amd)
19:31imirkin_: ok, well if you don't care about open-source support, then do whatever
19:31hakzsam: imirkin_, I will investigate on arb_shader_image_load_store-max-size
19:35thomas_sch: imirkin_: I care about open source support ^^ that's why I use nouveau, but neither nouveau nor radeon can get "good enough" (tm) performance for me so I want to have at least the option of a good blob (even thought I normally don't use it)
19:35imirkin_: thomas_sch: ok - obv make your own decisions :)
19:36imirkin_: amd has a full-time team with access to docs and hw whose job it is to make sure stuff works. nvidia does not.
19:36thomas_sch: imirkin_: when I purchase my next gpu I will come again and get your opinion and if it's the same I will probably get an amd card ;) let's wait and see
19:37imirkin_: i highly doubt i can effect change by convincing people to buy amd, but ... worth a shot :)
19:38thomas_sch: last time I had to buy 2 gpus I bought two nearly equal gpus (performance wise) from amd and nvidia - blob and source sucked for amd so I'm biased now :P
19:38thomas_sch: But I at least try to remain open :P
19:38thomas_sch: anyway thanks again
19:40karolherbst: Misanthropos: I have no idea, but it works for me :/ Maybe something else is messed up, but for that I would need more information actually
19:40imirkin_: Misanthropos: which branch did you use?
19:41karolherbst: imirkin_: he said, he used my recent one
19:41karolherbst: his vbios is also a bit oddish compared to the usual kepler one
19:41karolherbst: but it shouldnt matter actually
19:41imirkin_: he was imprecise, and you have a ton of branches.
19:42imirkin_: perhaps he used your master_4.6 branch
19:42imirkin_: which, mind you, is the default branch
19:42karolherbst: yeah, might be this
19:42karolherbst: I know he had an issue with my v1 version :D
19:42imirkin_: some people might be fooled into thinking that's the one they should use
19:42karolherbst: but I am sure I didnt broke it
19:43karolherbst: yeah, thats why I also say it
19:43karolherbst: the thing is, his issue wont happen with stock nouveau
19:46karolherbst: Misanthropos: do you have a "boost" file? Because on my current one, you wont get one
19:47karolherbst: wait.. you will get one, but you will only get errors reading it
20:03Misanthropos: karolherbst, there is a boost file bios directory
20:03karolherbst: Misanthropos: try to read it
20:04Misanthropos: not sure what you mean. its a boost.o
20:05karolherbst: Misanthropos: I meant where the pstate file is
20:06Misanthropos: cat boost
20:06Misanthropos: 2: 1032 MHz
20:06karolherbst: mhh, then you use a rather old version of my branch
20:07Misanthropos: git branch
20:07Misanthropos: * 4.6_reclocking
20:08Misanthropos: do i have to reset or force update?
20:08karolherbst: shouldnt matter
20:11karolherbst: ohh wait
20:11karolherbst: that 4.6_reclocking branch doesnt exist anymore
20:11karolherbst: Misanthropos: you have to use stable_reclocking_kepler_v5
20:11Misanthropos: oh well ok! :D
20:12Misanthropos: checking out and trying .. will let you know about the result
20:17imirkin_: karolherbst: good thing i asked :p
20:18karolherbst: it was my mistake as well
20:18karolherbst: "karolherbst_work, i have applied the 4.6_recklocking branch" ...
20:18imirkin_: of having too many branches?
20:18karolherbst: should have notice that that branch doesnt exist :D
20:18karolherbst: nope, my lacking of proper reading skills :p
20:19karolherbst: and I dont have that many branches actually
20:19karolherbst: there are much worse repositories
20:20thomas_sch: ahh karolherbst you were the guy who was pushing for including fermi reclocking support right?
20:21karolherbst: more or less
20:21imirkin_: thomas_sch: RSpliet was actually working on the mem reclock logic for fermi, fwiw
20:21imirkin_: [afaik he doesn't have a ton of time to spend on it these days]
20:22karolherbst: it is rather sad that I dont have contact to many students somehow :/ I really have some nice tasks some newbies could do :p maybe I should finish by graduation and just search for new devs :D
20:22thomas_sch: just read about it cause 2/3 of my nvidia gpus are fermi right now so I had quite an interest in it a while ago. What's the status if I'm allowed to ask? sounds stagnating
20:23karolherbst: sounds about right :p
20:23imirkin_: thomas_sch: https://github.com/RSpliet/kernel-nouveau-nv50-pm/commits/master
20:23imirkin_: not usable yet.
20:24karolherbst: but the problem about not getting new devs is kind of critical, right. I mean on the kernel side besides supporting new gpus, there isnt much going on for older ones :/
20:25imirkin_: actually the issue is lack of suitable newbie tasks and mentors to ease them into it
20:25karolherbst: I have some nice ones :O
20:26imirkin_: put them up in the trello
20:26karolherbst: I think they are actually, let me check
20:27karolherbst: well or not
20:28Keringar_: Well how new is too new, I'm interested in this sort of thing but it seems hard to get in to
20:28imirkin_: Keringar_: it is hard to get into =/
20:28imirkin_: most of the people involved today started out by solving a problem that they had
20:28karolherbst: before working on nouveau I didnt wrote any C code :p
20:28imirkin_: and then sticking around and solving other problems
20:29Keringar_: I would like to do that but I'm not even sure where to begin looking to solve my problem
20:30Keringar_: I'm gonna go read that introductory course
20:30karolherbst: imirkin_: https://trello.com/c/qpRpOfd6/145-software-overheating-protection :)
20:31imirkin_: karolherbst: you seriously think a newbie can tackle that one on their own?
20:31karolherbst: it isnt much
20:32imirkin_: Keringar_: another problem is that, as with many projects, the docs are wrong and/or wildly out of date
20:32imirkin_: like ... that introductory course :)
20:32karolherbst: I looked into the table and it is just a bit of changing it and see what nvidia does
20:32Keringar_: oh, do you have any suggestions?
20:33imirkin_: Keringar_: sadly, nothing constructive. your problem is in the nvidia-supplied firmware or the kernel's interaction with that firmware. you could RE the firmware (the ISA is well-known), but it would be a big task and annoying.
20:34karolherbst: Keringar_: well, it would make sense if you would see what issues kind of annoy you
20:34imirkin_: i'm guessing the most annoying issue right now is "nouveau doesn't load secboot properly" :)
20:34karolherbst: imirkin_: though I think, some tasks look hard at first, but are trivially solveable in the end
20:34imirkin_: any other issues are probably secondary
20:34karolherbst: imirkin_: the thing I was working on at the start, wasnt really considered beginner task as well :p
20:35imirkin_: karolherbst: indeed. but you wanted it. really bad.
20:35karolherbst: ohh, he has a gm20x gpu?
20:35karolherbst: imirkin_: right :D
20:35imirkin_: much like the task i started with, VP2, wasn't exactly trivial either.
20:37karolherbst: the gddr5 think was trivialy in the end, I was just lucky guessing where the issue was
20:38karolherbst: or you meant the entire reclocking thing
20:38karolherbst: and then yeah, that was kind of annoying in some parts
20:41karolherbst: Keringar_: if you want to mess a bit with the vbios, I also would +have a nice task related to the fans
20:43Keringar_: karolherbst: I'm up for that but I'm not at my other computer right now. I would also need to learn about what a vbios is
20:44karolherbst: that part in the vbios: https://gist.github.com/karolherbst/a31afe30b7023bece65adc3a95a07e8c
20:45karolherbst: I am like sure, that this table contains three points and a curve, and maybe some other things
20:46hakzsam: imirkin_, I'm going to push this https://cgit.freedesktop.org/~hakzsam/mesa/commit/?h=op_sub&id=49d411909dc2a7ea7bc2dde53029a7cc579979a1
20:46Keringar_: karlosherbst: I'm pretty lost right now
20:46imirkin_: hakzsam: sgtm
20:46hakzsam: didn't check on nv50 though
20:46imirkin_: hakzsam: i can check tonight
20:46hakzsam: but should work
20:46imirkin_: i think so too
20:47karolherbst: hakzsam: something related kind of: https://github.com/karolherbst/mesa/commit/afaf87a5439507d4bf1297d0f177488e22505793
20:47hakzsam: imirkin_, okay, I will tomorrow :)
20:47imirkin_: hakzsam: you can push now
20:47hakzsam: karolherbst, yep
20:47hakzsam: imirkin_, okay
20:50Keringar_: karloherbst: So am I verifying this by dumping my vbios? I'm pretty lost on what verifying this means https://nouveau.freedesktop.org/wiki/DumpingVideoBios/
20:51karolherbst: Keringar_: well, the idea is: upload a faked vbios to the nvidia gpu. Load nvidia kernel module, see what the changes in the faked vbios are changing in the behaviour of nvidia :p
20:53Misanthropos: karolherbst, kernel works. boots exists.. but i cant read it
20:53Keringar_: karolherbst: I'll try to work on this but it might take a while for me to grasp the basics so don't expect to much
20:54karolherbst: Misanthropos: thats because your vbios dont have any boost information
20:58hakzsam: imirkin_, I think I will wait, at least GL 4.1, for the ADD3 series because maybe it will improve the thing a bit more
20:58imirkin_: hakzsam: wait for/on what?
20:59imirkin_: hakzsam: you mean delaying ADD3 until GL 4.x is sorted?
20:59hakzsam: no rush for that anyways
21:09Keringar_: karolherbst: can I use envytools to flash the vbios?
21:10karolherbst: you dont flash it
21:11Keringar_: Yeah I have no idea how to start. Do you have any links to docs or something?
21:13imirkin_: docs about what?
21:13imirkin_: there's a decent reference at http://envytools.readthedocs.io/en/latest/ -- however it's quite hard to read, esp for a beginner
21:13imirkin_: nvidia gpu's are complex beasts :(
21:13karolherbst: Keringar_: nvafakebios is the tool for faking a vbios
21:14Keringar_: karolherbst: thanks
21:14karolherbst: Keringar_: basically you have to do something like that: get your vbios with nvagetbios -s prom
21:14imirkin_: that's why it helps to have a purpose in the learning - you focus in on the things that matter quickly.
21:14imirkin_: and ignore the millions that don't
21:14karolherbst: Keringar_: modify it as you like (you also can change bytes through nvagakebios)
21:14karolherbst: Keringar_: load that vbios with nvafakebios, then load nvidia
21:15karolherbst: Keringar_: the nice thing about that table is, that you somehow hear if you changed something :p
21:15karolherbst: allthough maybe you should get familiar with the vbios first
21:15karolherbst: Keringar_: run your vbios through nvbios
21:16karolherbst: nvbios -v also shows the bytes and offsets quite nice
21:16Keringar_: Thanks now I can probably get started once I get back to my home computer in a couple of hours
21:17karolherbst: having a second machines helps :p
21:17Keringar_: Yeah last time to get those dmesg logs I had to use my phone as a reference
21:25Lekensteyn: would it be possible to get some acks for the four nouveau_acpi.c patches? (current -next is broken, and was reportedly fixed by these patches)
21:26karolherbst: Lekensteyn: ping skeggsb :p
21:37Lekensteyn: skeggsb: ping, would it be possible to get acks for https://lists.freedesktop.org/archives/nouveau/2016-July/025607.html? The mentioned audio issue turns out not to exist on standard installations
21:40karolherbst: Lekensteyn: did you check if the bus rescan might work?
21:50Lekensteyn: karolherbst: in the code (nope) or manual (it requires remove first, otherwise it has no effect)?
21:51karolherbst: mhh, I am wondering why the device has to be removed first. Sounds like a bug inside the kernel
21:52Lekensteyn: I'll get back to audio later once I figure out why the video cards fails to resume on some laptops with AMI BIOSes
22:47imirkin_: hakzsam: when you get a chance, grab a 'glxinfo -l -s' against blob on gm20x
23:31Keringar: That secure boot bug I'm having only seems to affect people with Asus-z97 motherboards
23:34Keringar: and nvidia 970's