07:02karolherbst_work: stupid ISP… didn't even had internet at home today this morning...
07:05karolherbst_work: skeggsb: you know that the problem is propably that the 4th part is splitted in a 224 bit and 32bit interface?
07:05karolherbst_work: ohh, you are
07:15karolherbst_work: Sophira: but I hope that generally OpenGL stuff works without issues? Because then, we could try out if we can increase to 4 parts, but the last being only 512MB big and see if that still works
07:15karolherbst_work: which would be also kind of the best solution to the problem on those cards in my opinion
07:16skeggsb: it will still work, i already figured all this out with gnurou a week or two ago
07:16karolherbst_work: ahh I see
07:16skeggsb: just need more information to implement the detection of it properly
07:16karolherbst_work: I figured
07:16karolherbst_work: I guess we don't plan to support advanced paging just for those card yet?
07:17skeggsb: advanced paging?
07:17karolherbst_work: well the last 512MB are terribly slow
07:17karolherbst_work: and we can't put stuff there which os accessed often
07:17karolherbst_work: otherwise performance would be worse than one of a 710
07:17karolherbst_work: so if we would expose 4GB, we would have to move stuff around on the fly
07:18skeggsb: we already somewhat handle such restrictions in other mixed-memory configurations (ones where paritions have different amounts of memory)
07:18karolherbst_work: yeah well, different sizes aren't the problem though
07:18skeggsb: it's the exact same problem, on those configurations you can't use the "odd" section of memory fully either
07:18karolherbst_work: ahh I see
07:18skeggsb: we just can't detect the missing-LTCs one
07:19karolherbst_work: mhh :/
07:19karolherbst_work: I hope that won't end it stupid blacklisting
07:20skeggsb: i know enough to do a hackish job of implementing it, but i'm uncertain of some specific details that'll no doubt bite us down the track
07:21karolherbst_work: at first I kind of hoped there would be 5 parts with 3x 1GB and 2x 512MB :/
07:21karolherbst_work: but no, nvidia had to do one part with mixed memory interfaces…
08:50karolherbst_work: skeggsb: I see, that you are coming to XDC, too. But I guess you also stay with the red hat guys?
09:01karolherbst_work: pmoreau: I meant, that _we_ stay at the _same_ place, not the same place like in brussels, but more like we at the same ,)
09:02pmoreau: Definitely misunderstood it! Makes way more sense that way!
09:03pmoreau: WIth that in mind, yes, I definitely for that! :-)
09:03pmoreau: s/I/ I am
09:05karolherbst_work: silly me didn't take a day off for monday...
09:12pmoreau: Yeah… it ends up being a really short week
09:12pmoreau: I should have done that as well
09:14skeggsb: karolherbst_work, mupuf: i'll speak to my manager about it, but your place (mupuf) sounds alright!
09:14mupuf: skeggsb: yeah! Sounds great then!
09:14karolherbst_work: not that it gets too full in the end
09:15karolherbst_work: I think hans will stay with the red hat guys though
09:15mupuf: Well, I have space for 3 or 4 double beds. I already have 2, a couch and sleeping mats. I can get another bed and we would be good to go :)
09:16karolherbst_work: so one double bed for each of us?
09:16skeggsb: i slept on a couch last year, so, i'm easy
09:16mupuf: karolherbst_work: you need to check how you count :D
09:17mupuf: we are already 6 ;)
09:17karolherbst_work: well you are the host, so you can't have all the fancy things yourself
09:17karolherbst_work: and ben is on the couch as it seems .p
09:17skeggsb: or sleeping mat, whatever :)
09:17skeggsb: floor isn't unheard of either
09:18mupuf: yeah, we will definitely manage. No one will have to sleep on the flor
09:18karolherbst_work: well I don't care that much anyway, so I am fine with whatever I guess
09:18karolherbst_work: mupuf: by the way, reator seems to be down
09:18mupuf: you'll all finally meet reator in person
09:19mupuf: karolherbst_work: it is not ... it is just that we got a power outage and I went straight to a cou ch surfing summer camp for 3 days
09:19mupuf: I saw this morning that the rasp pi was still down
09:19karolherbst_work: I see
09:19mupuf: the net was down until I came back yesterday
09:20karolherbst_work: bad timing then I guess
09:20mupuf: my friends who are staying at my place were afraid of the beeping from the UPS and turned off everything
09:20karolherbst_work: since 10pm, my ISP got issues until this morning :D
09:21mupuf:needs a way to reboot the machine used to reboot other machines ... how meta
09:23karolherbst_work: aren't there like smart PSUs which can do stuff like that?
09:23karolherbst_work: ohh wait
09:23karolherbst_work: I am silly again
09:25karolherbst_work: mupuf: mind reading my proposal I wrote so far to tell me afterwards it is too short and unspecific? :D
09:25mupuf: karolherbst_work: I read it this morning ... but you were still out of IRC
09:25mupuf: [09:05:52] <mupuf> hey, your talk proposal looks good
09:25karolherbst_work: mhh did you read the newest version?
09:25karolherbst_work: the one with those stupid 1. … 2. … things?
09:26mupuf: I read the one from wednesday
09:26karolherbst_work: k, that's the old one
09:26mupuf: no, it had the 1. 2.
09:26mupuf: yep, very good
09:27karolherbst_work: k :)
09:27mupuf: gnurou will not attend the XDC, so we may need another update talk
09:27mupuf: or we hijack your talk a little bit for this
09:27karolherbst_work: I am all for it
09:27mupuf: wait, no, samuel will give one!
09:27mupuf: he already sent a proposal
09:27karolherbst_work: I was kind of aware he won't come, didn't know for certain
09:28karolherbst_work: did gnurou plan on saying anything?
09:28mupuf: Yeah, it is sad but he has more important things to tend to, unfortunately.
09:28karolherbst_work: on the update talk or so
09:28mupuf: karolherbst_work: sure, a bit on the firmwares and general updates
09:29mupuf: but maybe we can get Andy Ritger to talk about it instead
09:29mupuf: not sure how familiar he is with it though
09:29karolherbst_work: I am sure ben could do it :p
09:29mupuf: Ben could do what?
09:29pmoreau: When is/was the deadlin to submit for a talk?
09:29karolherbst_work: to talk about firmware stuff
09:29mupuf: pmoreau: 14th
09:29mupuf: oh, is it today?
09:29karolherbst_work: mupuf: liar
09:30karolherbst_work: "Proposals due: Wednesday 17 August 2016 17:00 UTC."
09:30mupuf: very good, because I need to submit my talks :D
09:30karolherbst_work: somebody changed it
09:30mupuf: karolherbst_work: /me wrote the quoted text
09:30karolherbst_work: I was sure it was the 15th before
09:30karolherbst_work: or so
09:30pmoreau: Eh, two more days… Hum… I doubt I’ll get anything interesting done by then.
09:30mupuf: https://secure.freedesktop.org/cgit/xorg/log/Other/Press/CFP2016.mdwn <-- no changes
09:30mupuf: pmoreau: a talk can motivate you to do more
09:31karolherbst_work: maybe my brain was smart and thought: better finish it two days earlier
09:31mupuf: karolherbst_work: let's face it, we suck at memorizing dates
09:31karolherbst_work: nah, my brain is just very good with that stuff :p
09:31pmoreau: I should sync with Hans, see if he wants to present something as well.
09:31karolherbst_work: helps me without me needing to tell it what to do :p
09:31mupuf: pmoreau: you may just hijack part of samuel's presentation
09:31karolherbst_work: or mine :O
09:31pmoreau: :-D Yeah, maybe
09:31pmoreau: Or both! O.O
09:32karolherbst_work: mupuf: wanna hijack mine too?
09:32karolherbst_work: like the first part?
09:32mupuf: RSpliet: oh! So sorry, I forgot to add your talk!
09:32karolherbst_work: it is kind of your topic actually :p
09:32mupuf: karolherbst_work: yeah, but I cannot give too many talks. I am OK with reviewing the talk and being present on the stand
09:33pmoreau: What is Samuel going to talk about? Tess on Maxwell? Image support? All of that?
09:33pmoreau: It doesn’t seem to be up on the program page yet
09:33mupuf: "My plan is to be part of the 'Nouveau status update'"
09:33mupuf: that's the extent of his proposal so far
09:33karolherbst_work: mupuf: well 5-10 minutes would be enough, I don't plan to give you more :p
09:34karolherbst_work: well more like 5 minutes
09:34mupuf: hehe, well, sounds like a slide or two
09:34pmoreau: Ah, k :-D
09:34karolherbst_work: but yeah, we can also do it your way :p
09:34karolherbst_work: up to you
09:35mupuf: Well, I could come up with a slide or two detailing the requirements for good PM and what happens if we screw up
09:35karolherbst_work: mupuf: or showing awesome graphs
09:35mupuf: what would be the point?
09:35karolherbst_work: no clue? :D
09:35pmoreau: mupuf: Bring the Titan for the LEDs!
09:35karolherbst_work: light show!
09:36mupuf: pmoreau: lol, yeah, I need to finish this patch and make a video of it :p
09:36pmoreau: Looking forward to it! :-)
09:36karolherbst_work: if you show a video I will rate all your talks with 0/5
09:36mupuf: karolherbst_work: lol, do you want to carry a computer?
09:36mupuf: and it is not FOSDEM, the room is enormous
09:36karolherbst_work: live demo! with turned out lights and the gpu says in morse "welcome to the nouveau talk"!
09:37karolherbst_work: nothing else
09:37mupuf: so... videos are fine
09:37pmoreau: karolherbst_work: German grading style, with 0 being the best grade and 5 the lowest?
09:37mupuf: pmoreau: good one ;)
09:37karolherbst_work: you cunning hoodlums :p
09:37pmoreau: That live demo is going to end up as a modern art representation :-D
09:40karolherbst_work: mupuf: how enourmous? I was thinking that there might be space for like 500+ viewers or is it _much_ bigger?
09:47karolherbst_work: mupuf: k, so do you have any preferences when we should arrive and leave? or is the general tuesday to sunday good enough?
10:01mupuf: karolherbst_work: yeah, 500
10:02mupuf: arrive whenever possible :)
10:03karolherbst_work: K :)
10:03karolherbst_work: so I figure not that much general audience will be there
12:34RSpliet: karolherbst_work: no, assume everyone is a developer of some sorts, but assume that it ranges from 3D-savvy people to display-lords to kernel plumbers
12:34karolherbst_work: RSpliet: sure, but with general audiance I was more thinking about non devs
12:35RSpliet: eg. don't assume people know what you're talking about, but assume they'll pick up quickly from very brief explanations
12:36karolherbst_work: I know, but that wasn't my point really :p
12:42karolherbst_work: I will just explain that much, so that I will be finished with my talk after ~40 minutes and be ready for demos, maybe earlier even
12:42karolherbst_work: mupuf: is there something like a 45min talk? 1 hour might be too long
12:42mupuf: karolherbst_work: just do 45 minutes and let people enjoy a coffee break
12:42RSpliet: karolherbst_work: ~5 for questions, 10 for coffee
12:42RSpliet: /tea/...club mate?
12:43karolherbst_work: I rather choose tea, thanks
12:44pmoreau: +1 for tea!
12:45karolherbst_work: though I really don't plan much time for the demos, because I can't show that much. maybe 5 in total. so 35m talk + 5m demos + 5m questions + 15m break
12:45karolherbst_work: sounds reasonable to me
12:45pmoreau: Or should we do a mega session for status update on Nouveau, rather than multiple smaller talks?
12:46karolherbst_work: should we just plan half a day then?
12:46pmoreau: RSpliet: I assume you are planning to talk about Fermi reclocking? Or some OpenGL stuff maybe?
12:46RSpliet: pmoreau: no
12:46RSpliet: read my abstract, silly ;-)
12:46karolherbst_work: I already mention RSpliet in my talk :p (or not, if he doesn't want to be mentioned)
12:47pmoreau: It wasn’t up yet when I looked at the program page this morning
12:47RSpliet: karolherbst_work: no that's fine. Send me the slides a few days before you've finished them
12:47karolherbst_work: my talk is actually about fermi-maxwell gpus, so
12:47karolherbst_work: RSpliet: well, the part where I mention you is something like that: "fermi memory reclocking being worked on (roy)" :p
12:48pmoreau: RSpliet: Oh nice!
12:48RSpliet: yeah... I expect to be scheduled later
12:48RSpliet: gives me an opportunity to excuse myself and explain why I didn't finish it yet :-P
12:48karolherbst_work: my talk isn't about how shitty we are :p no excuses
12:49pmoreau: RSpliet: Ah ah ah, hopefully you won’t be preempted! :-p
13:52karolherbst_work: by the way, one of the amd devs said, that bioshock creates like 1k+ fences and that the work mareko done doesn't affect nouveau at all, because it just gives the driver the possibility to get more information
13:52karolherbst_work: to handle those fances in a smarter way
13:52karolherbst_work: maybe we could then also make use of that, no clue how hard it would be though
13:53karolherbst_work: I doubt anybody besides me has a copy of bioshock?
13:53karolherbst_work: have fun then
13:53tobijk: but i dont know how helpful i can be
13:53karolherbst_work: too late
13:55tobijk: :D maybe i'll have a look at it later
16:24pmoreau: Since some of you have played way more than I have with GPU assembly and compilers, any idea why the compiler would prefer to emit an fadd + sel rather than a predicated fadd? (initial code: `if (pred) a += b;`
16:25imirkin: does the old value of a get reused somewhere?
16:27imirkin: and which compiler are you talking about?
16:28pmoreau: Using NVCC, and yes a gets reused, both since the if clause is inside a loop, and after the loop, Ifeed it to a sust
16:29imirkin: if you do a predicated add, you lose the old value of a
16:29imirkin: so if you use the old value of a later on
16:30pmoreau: If I do a predicated add, I should only lose the old value of `a`, if the predicate is taken, not if it isn’t, right?
16:32imirkin: sure, but if you have code like
16:32imirkin: a = c
16:32imirkin: if (pred) a += b
16:32imirkin: then a and c get merged
16:32imirkin: and there's an a' which is a + b
16:33pmoreau: Ok, right. If `a` and `c` are merged, I get it.
16:33pmoreau: But my code is more:
16:34pmoreau: a = 0;
16:34pmoreau: for (i = 0; i < 32; ++i)
16:34pmoreau: if (pred) a += b
16:34imirkin: hm ok
16:35pmoreau: I even fed the compiler with the corresponding PTX assembly where I predicate the add, and it still generated the add + sel combo
16:36imirkin: learn to like it.
16:36pmoreau: see you later
17:44karolherbst: mupuf: sent out the stuff :p
18:13pmoreau: Wow, nvbios didn’t really liked that VBIOS! It generated an output containing about ~400k lines of "0x\x\x\x\x: \x\x ???" (with \x being an hexa character)
18:13karolherbst: which one?
18:14pmoreau: Ah, probably due to missing the strap! I forgot about that one
18:14karolherbst: the strap shouldn't be required though
18:14pmoreau: From https://bugs.freedesktop.org/show_bug.cgi?id=93887
18:16karolherbst: this looks borked
18:16karolherbst: yeah, a little
18:20karolherbst: I would assume that something failed getting the vbios
18:20pmoreau: Could be as well
18:21pmoreau:is trying to find the strap value in the mmio trace
19:00karolherbst: mupuf: reator still doesn't work?
19:01karolherbst: really wanted to test maxwell2 reclocking today
19:02pmoreau: karolherbst: I guess you want to add some workload while reclocking I guess? Otherwise, I could try that on my GM206
19:02pmoreau: But I could try that tomorrow evening
19:02karolherbst: you have to be carefull about the heat though
19:02karolherbst: _really_ careful
19:02karolherbst: we have no idea if the hardware emergency shutdowns on high temp
19:02pmoreau: (I do not have a mouse here, so I can’t even log in…)
19:03karolherbst: allthough maybe gm206 are fine, those don't get that hot
19:04karolherbst: huh, they use the gm107 pmu implementation?
19:05karolherbst: ohh but no clk subdev
19:05pmoreau: Do you need some VBIOS/MMIO-trace/other data from it?
19:05karolherbst: pmoreau: so the thing would be to add ".clk = gk104_clk_new," to the gm206 device/base entry
19:05karolherbst: and see what that does
19:06karolherbst: on my branch of course
19:06pmoreau: RIght, was going to check that point ;-)
19:06karolherbst: and then running glxgears vsynced at highest clock should be _enough_ to test
19:06karolherbst: I really don't want you to burn your gpu
19:07pmoreau: I think it should be sufficiently cooled to prevent that, but I’ll still be careful.
19:07karolherbst: but without hardware thermal protection I would be carefull anyway
19:07karolherbst: even my kepler once got to 102°C
19:07karolherbst: with fans at full speed
20:07karolherbst: Sophira: I know a lot of nice things you could try :D
20:08Sophira: Okay. :)
20:08Sophira: What's up? :)
20:08karolherbst: first add the x11 overlay and install envytools
20:09Sophira: Okay. Hang on.
20:11Sophira: (Sorry, was just rebooting; I've been spending more time in 4.4.6 recently because of the font issue I'm having in 4.7.0, but I'm in the new kernel with acceleration now.)
20:14karolherbst: font issues?
20:14karolherbst: mhh might be nouveau related though
20:15Sophira: Sometimes characters will disappear when scrolling
20:16Sophira: (pasting a line of mine from earlier, but imirkin said not to worry about it until Ben comes out with a proper fix)
20:16karolherbst: well maybe in 2 years maxwell will be well supported too :)
20:17Sophira: 22:02 Sophira: I'm noticing some really weird issues on my 4.7.0 config. When I'm in X, text will have random characters disappear when I'm scrolling around in some things. It's happening most often in Leafpad but I just had it happen in Firefox too. I just rebooted back to 4.4.6 (which of course doesn't have the hardware acceleration going) and it seems to be okay there. (It's normally really easy to s
20:17Sophira: ee in Leafpad.)
20:18karolherbst: well after you have envytools installed. start a glxgears with vblank_mode=0
20:18karolherbst: and in another shell start doing "nvaforcetemp 90"
20:19karolherbst: and increase the number until you either reach 110 or glxgears perf goes down
20:19karolherbst: no worries, that doesn't increase the temperature, it just puts in a faked value
20:20Sophira: Getting an error building envytools :/ Hang on, uploading build log.
20:21Sophira: I should point out that I don't have any sensor-monitoring capability yet.
20:21karolherbst: doesn't matter
20:23karolherbst: gcc-4.9.3 right?
20:23Sophira: gcc version 4.9.3 (Gentoo 4.9.3 p1.5, pie-0.6.4)
20:23karolherbst: mhh it defaults to c89 mode afaik
20:23karolherbst: good to know though
20:24Sophira: So I need to add -std=c99 to CFLAGS for this?
20:24karolherbst: we should add it to the cmake files though
20:25Sophira: All you people running global ~amd64, really ;p
20:25karolherbst: kind of
20:25karolherbst: though ~amd64 is a bit for scary people too. but I only have 26 live packages :/
20:26karolherbst: ah crap
20:26karolherbst: with c99 there is no PATH_MAX
20:26karolherbst: stupid gnuc99
20:27karolherbst: Sophira: you need actually std=gnu99
20:27karolherbst: I will fix it for c99 I guess
20:34Sophira: OkayOkay, I put it into a local overlay, added -std=gnu99 to CFLAGS there and it's built now.
20:36Sophira: How do I get the current values for the value I'm forcing?
20:36karolherbst: uhh, sensors? :D
20:36karolherbst: you can cat the hwmon file too
20:37karolherbst: you need to fine the right entry in /sys/class/hwmon/
20:37karolherbst: mine looks like this: ../../devices/pci0000:00/0000:00:01.0/0000:01:00.0/hwmon/hwmon3
20:37Sophira: Well, I meant, I assume this value is a target value, so I was wondering how to get the current target value. But yes.
20:37karolherbst: in ls -l
20:37karolherbst: nah, if you put in 95, the gpu will report 95 °C
20:38karolherbst: temp1_input is the right name within hwmon
20:38Sophira: Ah. So nvaforcetemp basically overrides the temperature that the card reports itself as running as?
20:38karolherbst: and with name you can check which driver that is from
20:38karolherbst: Sophira: yeah
20:38Sophira: That sounds dangerous.
20:39karolherbst: it doesn't make the gpu hotter
20:39Sophira: (I only have one entry in /sys/class/hwmon, hwmon0. But it does seem to be nouveau, so.)
20:40Sophira: Right, but presumably if you force it to a certain temperature than the kernel could react to that and it could mask any actual temperature issues. But fair enough.
20:40karolherbst: well, the gpu may shutdown itself if you put a high enough value in
20:40karolherbst: but before that, the clocks will drop
20:40Sophira: I don't have a temp1_input file. I have in0_input and power1_input.
20:41karolherbst: no temp?
20:41RSpliet: karolherbst: unless suddently the fan drops off and the GPU reaches boiling point yet the driver still thinks it's 90 degrees
20:41karolherbst: RSpliet: maxwell2
20:41karolherbst: no fan controls
20:41RSpliet: Sophira: the odds of anything going wrong are close to zero, and just reboot afterwards to reset your card to default behaviour :-)
20:41karolherbst: silly me
20:42Sophira: RSpliet: Yeah, that's what I figured, I just needed to make sure.
20:42Sophira: karolherbst: Yeah, no temp.
20:42karolherbst: is that right? We can't even _read_ the temperature out on maxwell2?
20:42karolherbst: ... seriously?
20:42karolherbst: wait a second
20:43pmoreau: Eh, you might be doing weird things with it! Better to lock that up :-D
20:43karolherbst: Sophira: name is nouveau?
20:43tobijk: karolherbst: looking at the deferred flushing, i'm not certain it is wise to implement for nouveau where thread safety is not really given :)
20:44karolherbst: tobijk: hush, hush :p
20:44karolherbst: like I ever wanted to hear that
20:44tobijk: well driver development is a crule world on its own ;-)
20:45Sophira: karolherbst: Check the pastebin ;)
20:45Sophira: (Yes, it is.)
20:45karolherbst: Sophira: nvapeek 0x2044c
20:45Sophira: home hwmon0 # nvapeek 0x2044c
20:45Sophira: 0002044c: 0000006f
20:45karolherbst: nvapeek 0x20450
20:46Sophira: Okay.home hwmon0 # nvapeek 0x20450
20:46Sophira: 00020450: 00000040
20:46karolherbst: 47°C then
20:46karolherbst: did you force the temp already?
20:46karolherbst: force it to 90 and the nvapeek 0x2044c should increase
20:47Sophira: Not yet. Didn't want to do it until we had figured out the temp file issues :)
20:47Sophira: Doing that now.
20:47karolherbst: can you do a nvapeek 0x20400 too?
20:47Sophira: home hwmon0 # nvaforcetemp 90
20:47Sophira: home hwmon0 # nvapeek 0x2044c
20:47Sophira: 0002044c: 0000006f
20:48Sophira: home hwmon0 # nvapeek 0x20400
20:48Sophira: 00020400: 0000002f
20:48karolherbst: it seems to work, kind of
20:48karolherbst: could you run glxgears with vblank_mode=0 ?
20:48Sophira: I have been all this time.
20:48karolherbst: and after 10 seconds peek the last value again
20:48karolherbst: then stop it
20:49karolherbst: wait 10 seconds and see if the values drop
20:49karolherbst: the 0x20400 one
20:49Sophira: (I will say that over the last 15 seconds of its life the FPS values *did* drop.)
20:49karolherbst: well with drop I mean more like 50% drop
20:49Sophira: home hwmon0 # nvapeek 0x20400
20:49Sophira: 00020400: 0000002e
20:49karolherbst: we can't fake the temp anymore :O
20:50Sophira: home hwmon0 # nvapeek 0x2044c
20:50Sophira: 0002044c: 0000006e
20:50karolherbst: yeah, it dropped by 1 °C
20:50karolherbst: crappy crap :/
20:50karolherbst: that _is_ troublesome
20:51Sophira: I'll reboot, I think.
20:51karolherbst: wait a second
20:52karolherbst: somewhere around lind 2060 is nv124_chipset
20:52Sophira: (I see it.
20:52karolherbst: add a .therm = gm107_therm_new,
20:53karolherbst: and while we are at it
20:53karolherbst: ohh wait, reclocking with stock nouveau is too unstable though, so we better leave that for now
20:53Sophira: (How are you parsing 0x2044c/0x20400 to get the temperature, btw?)
20:54karolherbst: Sophira: those are hex values
20:54karolherbst: 0x20400 is the temperature since g84 gpus or so
20:54karolherbst: 0x2044c/0x20450 is new since gf119 I think
20:54karolherbst: there you have to calculate a bit
20:54karolherbst: 0x6e - 0x40 == 0x2e ;)
20:55Sophira: Ahhh, so 0x20400 is just the base, got it.
20:55karolherbst: it seems like they thought it is a good idea to support temperatues down to -64°C
20:55karolherbst: no, 0x20450 is the base ,)
20:55Sophira: Oh right, yeah, sorry, mistyped.
20:56Sophira: tempbase = 0x20450; tempaddr = 0x2044c; temp = *tempaddr - *tempbase;
20:58Sophira: One thing I do want to point is that glxgears *is* going slower.
20:58Sophira: Before I was getting things like:
20:58Sophira: 2349 frames in 5.0 seconds = 469.695 FPS
20:58Sophira: 2229 frames in 5.0 seconds = 445.492 FPS
20:58Sophira: 2157 frames in 5.0 seconds = 431.348 FPS
20:58Sophira: 2339 frames in 5.0 seconds = 467.773 FPS
20:58Sophira: Now it's:
20:58Sophira: 1714 frames in 5.0 seconds = 342.704 FPS
20:58Sophira: 1709 frames in 5.0 seconds = 341.718 FPS
20:58Sophira: 1731 frames in 5.0 seconds = 346.128 FPS
20:58Sophira: 1729 frames in 5.0 seconds = 345.696 FPS
20:59karolherbst: glxgears is stupiod
20:59karolherbst: restart glxgears
20:59Sophira: I did, about 10 seconds before those readings.
20:59Sophira: It was off for about 5 minutes before then.
20:59Sophira: Not running, I mean.
21:00Sophira: So I guess you can force temperature but not read back your forced temperature?
21:00karolherbst: you should
21:00karolherbst: try nvaforcetemp 0
21:01Sophira: I'll try 40 rather than 0.
21:01Sophira: A more normal sort of temp.
21:01karolherbst: 0 resets it
21:01Sophira: Ah, okay.
21:02Sophira: (it did take a while before glxgears started slowing down last time, so it might take a while before it speeds up again, perhaps.)
21:02karolherbst: no clue
21:02karolherbst: glxgears is odd
21:02Sophira: It's been holding steady at about 300 FPS for the last 25 readings or so.
21:03Sophira: (298-303 FPS)
21:03karolherbst: anyway, forcing the temperature doesn't seem to workout :/ sad
21:04Sophira: Oh, huh. The FPS increases if I actually have the window in the foreground; I can't just leave it running in the background, I guess.
21:05Sophira: It's about 408 FPS when in the foreground.
21:05Sophira: So yeah.
21:05Sophira: Probably nothing.
21:05karolherbst: .. I blame your window compositor now :p
21:05Sophira: Shall I start X with only twm running and see what it's like there?
21:05karolherbst: wanna compile nouveau yourself out of tree and test some maxwell2 reclcoking stuff? I am kind of sure it will work, but you gotta be carefull about the temperature
21:06karolherbst: we do _not_ know if the gpu will shut down on high temp on maxwell2
21:07tobijk: well just set 105°C and watch?!
21:07Sophira: tobijk: We can't, that's the thing.
21:07tobijk: :/ sorry just came back
21:08Sophira: nvaforcetemp isn't forcing the temperature.
21:08Sophira: No worries.
21:08Sophira: karolherbst: I'm not sure. I've never been one to mess with clock settings unless I know they're absolutely safe.
21:08karolherbst: we will still keep it far below the usualy clocks nvidia would use
21:09karolherbst: well maybe not, cause your vbios is odd :/ the heck
21:09karolherbst: the heck!
21:09Sophira: Oh, is it? Did you take a look at the vbios.rom, then?
21:09karolherbst: I actually thought that happens like never
21:09karolherbst: but it actually does
21:10karolherbst: RSpliet: you remember your comment about the linear csteps?
21:10Sophira: I'm in 4.7.0 now so I can try copying the vbios.rom again from here. (The previous vbios.rom was taken while I was in 4.4.6)
21:10karolherbst: RSpliet: https://gist.github.com/karolherbst/a1748a73f5937c051e2ca23175d7a6b0
21:11karolherbst: I so don't handle that right now I think...
21:11Sophira: home sophie # sha1sum vbios*
21:11Sophira: 2752f24d31c248782ab8b635f8f060cfdaca8b28 vbios2.rom
21:11Sophira: 2752f24d31c248782ab8b635f8f060cfdaca8b28 vbios.rom
21:11Sophira: Okay, never mind :D
21:11tobijk: Sophira: it is always the same
21:11tobijk: or better should be :)
21:11karolherbst: mhh base is 1050 MHz
21:11karolherbst: which is still... high...
21:11Sophira: tobijk: I figured it would be, but I don't know enough about the topic to know if the 4.4.6 kernel had any issues with that, so I wanted to check :)
21:12tobijk: Sophira: you are certainly right, better check :)
21:12karolherbst: Sophira: well if you are up to, we could try out some really safe operations while we have that card reclocked to a rather low clock
21:12Sophira: karolherbst: Sure.
21:13Sophira: So I need to get nouveau from git, right?
21:13karolherbst: I think we won't do more then a simple reclock and then a 2 second vblank_mode=0 glxgears
21:13karolherbst: or rather 1 second
21:14Sophira: Obviously not wanting the readings, then :)
21:14karolherbst: we can do a while true; do cat temp; done thing or so
21:14karolherbst: and as long as it stays below 90°C we are pretty safe
21:15karolherbst: I am only worried about temperature above 105°C
21:15karolherbst: critical domain starts there
21:15Sophira: Okay, a 'make' worked, but I don't seem to have a .ko file.
21:15karolherbst: nvidia keeps max clocks until 97°C for me
21:15karolherbst: well, two things
21:16karolherbst: git checkout maxwell2_playground
21:16karolherbst: cd drm
21:16Sophira: Ahhh, more setup required, okay :)
21:16Sophira: Should I 'make clean' in master_4.7 first?
21:17Sophira: Since I already did a 'make' there, I mean.
21:17karolherbst: wait :D
21:17karolherbst: I forgot something
21:17karolherbst: that parts hack is still missing
21:18karolherbst: "u32 parts = nvkm_rd32(device, 0x022438);" => u32 parts = 3;
21:18Sophira: Ah, of course.
21:19Sophira: Some time we should do whatever the equivalent of setting that to 3.5 is ;p
21:19Sophira: But yes.
21:20Sophira: Okay, all made and I have drm/nouveau/nouveau.ko .
21:20karolherbst: find nouveau.ko inside /lib/modules/4.7
21:20karolherbst: or whatever that dir is
21:20karolherbst: and put the compiled one there
21:20karolherbst: remove the old one first
21:21Sophira: I copied it to .bak.
21:22Sophira: Note to self, be root before trying to do that ;p
21:22Sophira: Okay, done. Ready to reboot.
21:22karolherbst: then reboot
21:23Sophira: Oh, actually, first.
21:23Sophira: Did we want to make that .therm change in these files?
21:23karolherbst: I already did that
21:23Sophira: ".therm = gm107_therm_new,"
21:23Sophira: Fair enough :D
21:25Sophira: Okay, back.
21:25karolherbst: do you have a working temp entry in hwmon now?
21:26Sophira: Yep, I have a bunch of temp1 things now.
21:26Sophira: home hwmon0 # cat temp1_input
21:26karolherbst: in a shell you can always see: watch -n 0.1 -x cat temp1_input
21:27Sophira: Huh, never used -x before.
21:27karolherbst: doesn't spawn a sh or so
21:28karolherbst: now go to /sys/kernel/debug/dri/0
21:28karolherbst: and cat pstate and boost
21:30Sophira: Want me to pastebin it or just paste it here?
21:30Sophira: (10 lines total pastage)
21:32karolherbst: very good
21:32karolherbst: okay.. that might hang your machine immediatly now
21:32karolherbst: echo 0a > pstate
21:32Sophira: Okay, one sec.
21:33Sophira: Computer hasn't hung, but the echo isn't returning.
21:33tobijk: wow 7Ghz
21:33Sophira: temp1_input is holding steady at 44000.
21:33karolherbst: Sophira: mhh crap, damn pmu :/
21:34karolherbst: Sophira: did dmesg print anything?
21:35Sophira: I wasn't watching, sorry. But I see things like:
21:35Sophira: [ 617.957592] nouveau 0000:01:00.0: therm: FAN target request: 27%
21:35Sophira: [ 617.957603] nouveau 0000:01:00.0: therm: FAN update: 31
21:35Sophira: [ 617.957621] nouveau 0000:01:00.0: priv: HUB0: 10eb14 8000014f (19400200)
21:35karolherbst: doesn't matter
21:36Sophira: (the 'echo' process isn't showing up in the output of "ps aux", interestingly.)
21:36karolherbst: because it is bash
21:36Sophira: Oh, of course.
21:36Sophira: Sorry, that was me facepalming right there :D
21:36karolherbst: why is the pmu so silly now :D
21:36Sophira: What's the PMU?
21:37karolherbst: power management unit
21:37karolherbst: it actually reclocks the memory
21:37Sophira: Should I try catting pstate/boost again from another term?
21:37karolherbst: well, you can
21:37karolherbst: but I doubt it will change anything
21:38Sophira: Actually, the output has changed.
21:38Sophira: New pastebin incoming.
21:38karolherbst: the AC line tells you the current clocks
21:39karolherbst: so nothing changed
21:39Sophira: Just that the "AC DC *" wasn't there before.
21:39karolherbst: can you get the pid of the hanging bash process?
21:40Sophira: I can, but not easily. Give me a moment :D
21:40karolherbst: and there should be a D kworker too
21:40karolherbst: also the pid of that one
21:41karolherbst: well I use htop for that, seems simple enough
21:42Sophira: Okay, looks like it's running in pts/4...
21:42Sophira: sophie 2689 0.0 0.0 21964 3740 pts/4 SNs 22:29 0:00 | \_ bash
21:42Sophira: root 2988 0.0 0.0 47780 2856 pts/4 SN 22:29 0:00 | | \_ su -
21:42Sophira: root 3009 0.0 0.0 22064 3820 pts/4 DN+ 22:29 0:00 | | \_ -su
21:42Sophira: So 3009 is probably what you want.
21:42karolherbst: why 3009?
21:43karolherbst: ohh wait
21:43karolherbst: it should be a bash
21:43karolherbst: cat /proc/3009/stack
21:43Sophira: I normally run "su -", if it matters, rather than "su".
21:44Sophira: So looks like the one.
21:44karolherbst: but you also need to find the kworker in D state
21:44Sophira: Want the stack of that too?
21:44karolherbst: then the stack of that one too
21:46karolherbst: as I expected
21:46karolherbst: silly pmu
21:48Sophira: (in case you're wondering, to find that bash process I had to echo to the different /dev/pts/X terms until I found when it printed to the hung terminal :P)
21:48Sophira: Well, it worked.
21:49karolherbst: something is funky with the pmu
21:49karolherbst: well the pmu is funky on maxwell2 anyway
21:50karolherbst: skeggsb: you know anything about why the pmu doesn't behave on maxwell2?
21:53karolherbst: mhh not that anything changed in the ISA again
21:53karolherbst: Sophira: anyway, that is too painfull to debug like this, I guess I need to get my hands on a maxwell2 and see what is broken
21:53karolherbst: pmoreau: you won't need to test that stuff anyway, except you want to mess with the pmu then
21:54karolherbst: Sophira: sorry for the trouble without getting anywhere :/
21:54Sophira: karolherbst: I could give you ssh and we could share a screen session, maybe? Though it would be nice to know what you were doing too, that's the only thing. I don't know enough at the moment.
21:54pmoreau: karolherbst: I’m ok with messing with the PMU, as long as you give the instructions for it, since it is my own card and not a card from work ;-)
21:54karolherbst: pmoreau: nvkm_pmu_send doesn'T get a reply ;)
21:55pmoreau: karolherbst: At some point I will have set up the computer to be accessible like Reator, but that day hasn’t come yet
21:55Sophira: I'm just sorry that I know too little about this so I have to keep asking what's being done. I know that must be painful for you.
21:55karolherbst: Sophira: we have a server for that already, but the guy owning is, isn't here and it seems to be off or so
21:56karolherbst: nah, if you plan to help out, you have to get used to the tools and the process anyway ;)
21:57pmoreau: I’ll grab a mouse tomorrow and be ready to burn everything down
21:57Sophira: I was just worried that you were saying that my cautiousness and asking questions is why it was too painful for you to continue.
21:57karolherbst: pmoreau: you should use my maxwell2_playground branch then
21:57karolherbst: pmoreau: I've enabled temperature readout and the clk subdev there
21:57karolherbst: Sophira: I asked the same question at the start too :D
21:58karolherbst: pmoreau: well, it just works as it used to work :D
21:58karolherbst: seems like we can indeed do a few things
21:58karolherbst: but faking the temperature won't work
21:58karolherbst: ... which is sad
21:58pmoreau: Maybe you should add a do_not_ever_touch_that_branch in the name, just to prevent potential user of thinking: "Oh, maybe there is reclocking support on it!"
21:59Sophira: Should I restore the kernel version of nouveau.ko and reboot, or should I stick to yours?
21:59karolherbst: that branch is totally save as in it doesn't reclock the card anyway .D
21:59pmoreau: K :-D
21:59karolherbst: Sophira: doesn't really matter. My branch has all the 4.8 updates already
21:59karolherbst: maybe things might be better with it, don't know it though
21:59Sophira: (BTW, I think I found why there was glitchiness on the framebuffer; the glitchiness starts when the startup process uses serfont to switch to the ter-u22n font.)
22:00karolherbst: ohh wait
22:00karolherbst: Sophira: check hwmon again
22:00karolherbst: there should be a power_input thing
22:00karolherbst: cat it
22:01karolherbst: power1_input to be precise
22:02Sophira: Okay, one sec.
22:02Sophira: home hwmon0 # cat power1_input
22:02karolherbst: 25W mhhh
22:02karolherbst: we could try out something
22:03karolherbst: nvapeek 0x20200 0x60
22:03Sophira: Okay. (The 'echo' is still hung, btw. Should I kill it that su?)
22:03karolherbst: you can't kill it
22:03karolherbst: I wrote a patch to timeout, but seems like skeggsb isn't interessted :p
22:04karolherbst: well it was more of a workaroun anyway
22:04karolherbst: maybe if I remove those workaround bits, it would be fine
22:06Sophira: What did you want to try?
22:06Sophira: I didn't see the nvapeek.
22:06Sophira: Running now.
22:06karolherbst: power1 still around 24669760?
22:07Sophira: Slightly less. 24572480
22:07karolherbst: Sophira: then nvapoke 22000055 0x60
22:07karolherbst: that doesn't make sense
22:07karolherbst: nvapoke 0x20200 0x60 22000055
22:08karolherbst: and then cat the power1 file again
22:09karolherbst: 3W less power consumption
22:09karolherbst: I really should work on patches for that
22:09Sophira: So I guess you're faking temperature in a different way?
22:09karolherbst: this enabled clock gating
22:09karolherbst: kind of removes the clock signal on idle for all gpu engines
22:10karolherbst: there is a more advanced thing for that where the actual power is also cut off
22:10karolherbst: but never got around to figure this out
22:10karolherbst: but saving more than 10% is good enough for now
22:10Sophira: temp1_input is at 42000 right now.
22:10karolherbst: makes sense :p
22:12pmoreau: karolherbst: Which version is mx2_playgnd based upon?
22:13karolherbst: yay, my talk was added to the proposals list :) just have to wait for end of august then I guess
22:15pmoreau: So to recap: Roy, Samuel and you are will be presenting for Nouveau, right?
22:15karolherbst: I doubt roy is
22:15karolherbst: kind of
22:15karolherbst: does it count?
22:16karolherbst: actually it does
22:16karolherbst: what is wrong with me today
22:16pmoreau: Lack of tea? :-)
22:16karolherbst: I only drank like a fourth glass of wine
22:16karolherbst: maybe that's not enough
22:17karolherbst: I meant like 1/4
22:17karolherbst: not four glasses
22:17pmoreau: I have a bottle over here that I should drink
22:17karolherbst: I got wine from turkey, even if that sounds odd
22:17pmoreau: How is it?
22:18karolherbst: I opened it yesterday
22:18karolherbst: today it is slightless worse
22:18karolherbst: tastes kind of like a normal chianti, just more sour or so
22:19pmoreau: Weird, I wouldn’t expect wine to taste worse on the second day
22:19karolherbst: me neither
22:20karolherbst: has a rather strong taste though
22:20pmoreau: Oh man, `sl` on a 21' screen is epic!
22:20pmoreau: steam locomotive: https://www.archlinux.org/packages/community/x86_64/sl/
22:21karolherbst: I have a 23" display here
22:22karolherbst: so awesome
22:22pmoreau: Eh eh! ;-)
22:22karolherbst: can I make it bigger?
22:23karolherbst: https://i.imgur.com/Lxk4j57.png :3
22:23pmoreau: I didn’t know you could give it flags to alter the rendering!
22:24pmoreau: Well, only 4, but still :-D
22:25karolherbst: -F :O
22:25pmoreau: That combo!
22:27pmoreau: Ok, I have the module compiled.
22:28pmoreau: :-/ HUB0 keeps spamming every sec for some reason
22:30pmoreau: "nouveau 0000:03:00.0: priv: HUB0: 10eb14 8000012c (1b400200)", and it’s cycling between 1b400200, 19400200, 1f400200 and 1d400200
22:31karolherbst: I guess we leave it be for now
22:31karolherbst: maybe skeggsb knows something
22:31pmoreau: Eh, it just switched to 18400200, 1e400200, 1c400200, 1a400200, weird
22:31pmoreau: And now between the two patterns
22:32pmoreau: Let’s see if I can get some X running
22:32karolherbst: sure it doesn't go from 18 to 1f?
22:32karolherbst: and nouveau is just slow
22:33pmoreau: Well, I am only reporting the message being printed on screen. I can’t guess what is not displayed here
22:33pmoreau: But yes, that could make sense
22:34karolherbst: pmoreau: for fun apply that patch: https://github.com/karolherbst/nouveau/commit/d5998e334dc3bbb835990894712234ed029762e4
22:34karolherbst: it lets nvkm_pmu_send timeout
22:37karolherbst: pmoreau: did you see that sl has actually 23 pull reuqests? :D
22:37pmoreau: No, I haven’t! :-D
22:43pmoreau: Ah ah ah!
22:45pmoreau: I have the pach applied, what should I try
22:46karolherbst: memory reclocking
22:46karolherbst: what else? :D
22:48karolherbst: anyway, have to go now, need sleep and stuff
22:48karolherbst: have fun then
22:49pmoreau: Good night
22:49karolherbst: note: having a 23" and a 15" both with fullHD native, is odd
22:57pmoreau: Ah!! Stupid me! I connected twice the same screen to the GPU. And since nothing was changing on one screen, I thought the keyboard was not working with X, whereas things were happening, but on the other input of that screen! --"
23:00pmoreau: Consumption jumped from 28W to 41W (from 07 to 0a), but no memory reclocking.
23:04imirkin_: at least something's working ;)
23:05pmoreau: It didn’t crash, that is true! Nor did my GPU burn :-D
23:06pmoreau: BTW, if you need data from a GM206, I will be able to provide some now that I managed to interact with a GUI.