15:05AndrewR: Hello again. Sorry for asking for such trivial matter - but can be info about (non)reclockability (and probably link to project website) added directly to dmesg? because right now users apparently fail into this trap regulary ...
15:06imirkin_: AndrewR: like what?
15:06imirkin_: "Thank you for choosing NVIDIA. We appreciate that you have a choice of vendors for your GPU, and you appear to have made the wrong one"
15:07HdkR: might as well as print out to stdout and stderr every time a GL context is created to ensure they know
15:09imirkin_: nah - just do an OSD.
15:10HdkR: Oh, even better
15:10HdkR: Permanent watermark
15:11imirkin_: someone redo the nvidia logo into a sadface somehow
15:11imirkin_: too bad i suck at visual things
15:12AndrewR: imirkin_, not everyone was buying hw because they knew they will run linux at some point in their lives :/ I did, thanks to my monitoring of this channel, but .... sometimes you forced to live with what you have, or what was given :/
15:13imirkin_: i know
15:13imirkin_: did i say anything to the contrary?
15:15AndrewR: imirkin_, probably bad wording or me badly understanding english: choise in some cases was done years ago ... so, because distros pick nouveau as free driver users suddently become penalized without any simple indicationw where root of problem lies - and even may not discover manual reclock method for earlier hardware!
15:15imirkin_: the root of the problem lies in their choice of hardware
15:16AndrewR: imirkin_ but what if it was done not by them, or when they were noobs ?
15:16imirkin_: doesn't change th root of the problem
15:16JayFoxRox: AndrewR: then it's a problem by the one that made the choice
15:17JayFoxRox: whoever orders systems is responsible to make a good choice - and nvidia has often been criticized publicly for their shitty driver / documentation politics
15:17JayFoxRox: so if anybody is placing an order for a >= $100 device and intends to use it with Linux, then they should know
15:17imirkin_: AndrewR: lobby distros to not use nouveau by default
15:18imirkin_: or suggest an "error" message that is better than mine.
15:18JayFoxRox: lol.. imo the issue isn't nouveau - but nvidia
15:18AndrewR: JayFoxRox, mmm... well, it usually (in windows world) refers to vendor (binary) drivers ..Linux is different, but old habits/mental inertion dies hard ....
15:18imirkin_: i'm not at all against such an error message, btw
15:18JayFoxRox: we should lobby for nvidia to be more supportive / release their stuff FOSS etc.
15:18imirkin_: JayFoxRox: easier to complain about nouveau and blacklist it everywhere.
15:18imirkin_: seems to be the path being taken.
15:19JayFoxRox: it's a stupid path because it just makes the problem worse, as nouveau will get less attention
15:19JayFoxRox: less users = less developers = more problems
15:19HdkR:currently has nvidia and nouveau blacklisted
15:19imirkin_: HdkR: passthrough? :)
15:19imirkin_: JayFoxRox: has the benefit of being easy though.
15:19HdkR: Nah, I don't want linux to use the iGPU or the one in the thunderbolt3 enclosure
15:20JayFoxRox: imirkin_: if people want it easy, they either shouldn't use linux, or they shouldn't be buying nvidia :P
15:20HdkR: er, mobile one inside the laptop?
15:20JayFoxRox: [OR they should contribute to nouveau]
15:20AndrewR: imirkin_ just a bit like ati DDX used to print at startup, just for kernel: "Your hardware may need additional tuning, visit nouveau.freedesktop.org for more info'
15:20imirkin_: AndrewR: like i said, i'm not against an error message
15:20imirkin_: suggest one.
15:20imirkin_: and when it should be printed.
15:21AndrewR: imirkin_ at probe time, along all this tech info ?
15:21imirkin_: (or info or whatever message)
15:21imirkin_: right ... but for which gpu's?
15:21imirkin_: the ones where manual reclocking works?
15:21imirkin_: and my error message for the ones where it doesn't? :)
15:22HdkR: What about the ones where reclocking just doesn't work? :)
15:22imirkin_: skeggsb: airlied: what do you guys think?
15:23AndrewR: imirkin_ depend on family you may ddd warn about secboo, manual relcocability..but generic line should be like above" Your device will be slow due to low boot clocks, in any case visit nouveau.freedesktop.org for more information"
15:23imirkin_: [adding an info message for manual reclocking availability, and an error message for gm200+ saying they should go back to the store and buy amd]
15:24AndrewR: imirkin_, actually i tend to think this is socio-technical problem - so far nearly any talanted developer who can fix some of those issues sucked into some company before they had chance to do something for nouveau ...
15:25imirkin_: AndrewR: some after they had the chance :)
15:25AndrewR: myself apparently have some low ceiling when iit comes to programming :/
15:26AndrewR: imirkin_ still, some things tend to dangle around ..like those perf counters... they were RE'd yet wait for kernel side changes killed ininciative, at least for my generation of hw
15:27AndrewR: imirkin_ problem from last year, when I tried to get more info via Gallium HUD trying to understand why Mafia2 via wine was so slow ....
15:27imirkin_: should work...
15:28imirkin_: (probably same thing i said then too...)
15:28AndrewR: it worked, but not at level I was assuming it worked ..moent, try to reding this
15:29AndrewR: imirkin_, this part never landed https://www.phoronix.com/scan.php?page=news_item&px=gallium3d-perfkit-nvidia
15:30imirkin_: hakzsam: what happened with that perfkit thing? i totally forgot about its existence.
15:31hakzsam: imirkin_: never landed
15:31imirkin_: was it usable?
15:31RSpliet: AndrewR: "socio-technical problem"... oh yes, the age old problem that life costs money, and very few people willing to give money to those in a position to accept that money to work on nouveau, because as it turns out there's no solid business case for nouveau... As opposed to a user case. There's a very solid user case to be made for nouveau.
15:31AndrewR: last commits ...they never landed in main nouveau tree
15:31hakzsam: imirkin_: https://cgit.freedesktop.org/~hakzsam/mesa/log/?h=nouveau_perfkit
15:31hakzsam: imirkin_: it was still experimental
15:31imirkin_: hakzsam: yeah, but like ... how would it be used by someone who wasn't you?
15:31imirkin_: could one use nvidia perf things?
15:32imirkin_: or were you expecting a whole application suite to spring up around this new interface?
15:32hakzsam: imirkin_: it was usable with a demo from NV, I don't remember which one
15:33AndrewR: RSpliet, well, at least attempt at solving this probably requres making more developers ... and this is ust as easy as making humans in general (no jokes please, because making human requres like 10-15 years of effort ...with often not-so-stellar results ..as I figured out ....at my ~37 age)
15:39AndrewR: i had this 'brilliant' idea about contacting those ppl: https://bootlin.com/blog/tag/vpu/
15:40AndrewR: or may be some developers can chnage their preffered projects for some months, so they will walk in someone else shoes ...considering project i'm thinking about also basically one-dev project ..this might be not as good idea as i think about it
15:40imirkin_: they're a solid group of contract developers
15:40imirkin_: although they tend to focus more on embedded stuff
15:40imirkin_: afaik they were contracted to do some vc4-related items
15:41AndrewR: imirkin_ wasn't tegra soc actually nvidia gpu within arm system? but GPU there seems to be different from desktop cards .. :/
15:41AndrewR: imirkin_, i mean VP* part of it ...
15:42karolherbst: AndrewR: they have two GPUs
15:42karolherbst: one 2D and one 3D chip
15:43AndrewR: karolherbst, with video decoder hanging sideways? :}
15:43karolherbst: no idea, but I think they use the real nv GPU for that
15:44AndrewR: well, my idea - may be they can handle 'campaign' part of it...like many users hoped to donate money for nouveau yet run into problem most important thing here seems to be dev time ... so, may be those users can pay bootlin devs, and as sideproduct of those efforts nouveau will get some code, too
15:44RSpliet: AndrewR: Google paid for open source Tegra support at some point. From a "DVFS" point of view they were able to skip the most difficult part (DRAM clock changes), because they use system RAM rather than dedicated video RAM on those chips.
15:45karolherbst: AndrewR: the problem is also rather the lack of people than anything else. And it takes a lot of money until that would get useful
15:45AndrewR: RSpliet, yeah .... may be you can level up your finding on reclocking in nouveau wiki?
15:45karolherbst: minimum 4k€ a month, otherwise it's wasted
15:46karolherbst: or to rephrase is: what would be possible with less money?
15:46RSpliet: AndrewR: I'm one of those unfortunate souls who's time is sinked into a PhD. I'm sure it'll pay out on the long run, but there's 0 reserves for doing nouveau stuff.
15:47karolherbst: AndrewR: also, if somebody wants to fund nouveau, it should probably go through the Xorg foundation with some internal budget which gets only spent on nouveau stuff
15:49karolherbst: AndrewR: or do you have any ideas how this "paying bootlin" would even work out?
15:51karolherbst: thing is, sponsoring a company is never a good way to finance an open source project or the community around it as this money doesn't get fully to the project
15:51karolherbst: it's always a waste of meny
15:52karolherbst: if we want users to be able to donate, we can ask the xorg foundation to provide for this
15:52karolherbst: and if the donations are enough, the foundation could even hire people working on nouveau
15:52karolherbst: without any commercial interests
15:53karolherbst: also, who are those users who would be willing to donate? I never heard of any
15:53karolherbst: but anyway, maybe we could talk about this on XDC
15:53RSpliet: there's a decent subset of users happy to donate old hardware. As for money... well
15:54RSpliet: Speaking of which, I pulled a GTX760 off the street today.
15:54karolherbst: hardware donations are not the issue
15:54karolherbst: it's already spent money, so it's much easier to do
15:54karolherbst: donating meonry regulary is what's difficult
15:55karolherbst: one time donations are also useless
15:55karolherbst: depends on the amount
15:55karolherbst: but everything below 1k€ is only symbolic
15:55HdkR: One time donations have an amount of appreciation to them but it doesn't go nearly as far as long term sustained donation subscriptions P
15:55RSpliet: One-time donations lead to one-time investments. Like X.org EVOC and GSOC
15:56karolherbst: I also totally dislike this sponsoring approach Xorg is doing right now
15:56karolherbst: it's more annoying than it actually helps
15:56orbea: i would not be surprised if people donated if a donate button would appear.
15:56HdkR: Same thing is happening with emulators with bounty culture as well
15:56karolherbst: maybe I should talk about this on XC?
15:57orbea: HdkR: and patreon
15:57karolherbst: patreon is bad
15:57karolherbst: it's waste of money
15:57RSpliet: the general consensus has tradisionally been that we don't have a good place for these donations to be managed and spent.
15:57orbea: i agree, but I see people throwing money at it regardless...
15:57karolherbst: but you use patreon if you have no infra
15:57karolherbst: Xorg has
15:57karolherbst: Xorg even has a bank account for stuff
15:58karolherbst: and normal people don't as providing the bank account data is harmful
15:58orbea: yea, obviously using Xorg would be better in all regards
15:58karolherbst: and stuff like this
15:58karolherbst: there are donation button though
15:58karolherbst: on xorg
15:58RSpliet: You could ringfence nouveau donations for nouveau, but it seems dodgy to then funnel it to the few "professional" contributors even if it's for purchasing hardware or the likes.
15:58RSpliet: Dividing over contributors is equally undesirable and questionable tax-wise
15:58karolherbst: RSpliet: ohh, we should ask people if they donate to Xorg to write it's for nouveau
15:58karolherbst: then we can ask Xorg how much we actually get
15:59karolherbst: I have no idea how many donations there actually are
15:59imirkin_: i've asked x.org if they would handle nouveau donations
15:59imirkin_: they said "no"
15:59karolherbst: imirkin_: they have to
15:59karolherbst: how can they say no
15:59imirkin_: they'll happily handle donations to x.org
15:59karolherbst: so either they _are_ the xorg foundation or they are not
15:59RSpliet: Quite easy, it's only 2 letters
15:59karolherbst: imirkin_: so nouveau isn't part of the xorg foundation
15:59karolherbst: glad to hear....
16:00imirkin_: but they won't earmark them
16:00imirkin_: and they won't manage distribution of funds to reimburse ... stuff
16:00karolherbst: the Xorg foudnation becomes pointless by every day, doesn't it?
16:00imirkin_: which i totally get - it's a pain.
16:00karolherbst: of course it's a pain
16:00karolherbst: but the foudnation has money
16:00imirkin_: tbh i don't remember precisely what i asked about
16:00RSpliet: XDC isn't free either. Nor is infra.
16:00imirkin_: but it was something along those lines.
16:00karolherbst: if it's painful for the members, then we hire somebody.. it's just a waste of meny
16:01orbea: might be worth asking again, maybe you'll get a different answer?
16:01imirkin_: like in Shawshank Redemption? :)
16:01karolherbst: anyway, I see this as one of Xorgs responsibilities to handle all thise berucrazy crap
16:01imirkin_: send a letter a week?
16:01karolherbst: otherwise, what's the point of a foundation
16:01orbea: i'd imagine you asked a while ago and time can change things
16:01imirkin_: at least a week ago
16:02imirkin_: maybe even more
16:02imirkin_: like ... 100 weeks ago
16:02imirkin_: or 200?
16:02karolherbst: I guess it depends on who is the board
16:02karolherbst: but, seriously
16:02imirkin_: full of reasonable people
16:02imirkin_: with no ill will towards nouveau
16:02imirkin_: or me, at the time
16:02karolherbst: what would even be a better "task" for the foundation than to manage project specifics donations as well
16:02karolherbst: and maybe even hiring people
16:04karolherbst: I am already upset enough about this stupid way of sponsoring
16:04karolherbst: never saw it that bad anywhere else I went
16:06karolherbst: ohh Lyude is a board member now
16:36Lyude: karolherbst: yep! :)
16:42karolherbst: can we get those sponsoring rules changed? :D I don't mind the general idea of sponsoring, I just think this daily mentioning is just annoying and pointless
16:44karolherbst: not a big fan of sponsoring in general, but it does give us a significant amount of money. Just wondering if we can make it less crappy
16:44karolherbst: or at least remove the entirely pointless parts of it
17:06Lyude: karolherbst: I beg your pardon?
17:06karolherbst: Lyude: ohh, we have this sponsoring rule, where at each begining session each day, we tell everybody how wonderful our sponsors are and name each by name and with big pauses so everybody can clap
17:07karolherbst: it kind of felt like we are trained monkeys to do so in the audience
17:07karolherbst: but maybe that's just me
17:07imirkin_: DisplayPort 2.0 enables up to 3X increase in video bandwidth performance (max payload of 77.37 Gbps)
17:07Lyude: I think that's usually how sponsorships work :p
17:07imirkin_: that's a lot of bandwidth.
17:07karolherbst: Lyude: each day?
17:07karolherbst: never saw it anywhere each day
17:07karolherbst: usually just the opening or closing session of the full conference
17:08karolherbst: and sometimes just on the slide
17:08karolherbst: without even saying out the names out loud
17:08karolherbst: with our audience it is even questionable if there is any benefit at all to do it like we do
17:08imirkin_: gotta make the sponsors feel all warm and fuzzy inside
17:08karolherbst: and I highly doubt any of those sponsors would go: "ohh, if you don't do it each day, we cut our sponsoring by half" which wouldn't hurt us
17:09Lyude: karolherbst: you're welcome to send an email to the board if you wish :p, this kind of stuff usually needs to be discussed with other board members and voted on
17:09karolherbst: imirkin_: sure.. but XDC is for us
17:09karolherbst: not for doing advertisment for companies
17:09imirkin_: then they won't sponsor.
17:09karolherbst: ohh, I am not arguing for it in general
17:09karolherbst: to tell the names once per conference is enough
17:09karolherbst: and have their labels somewhere
17:09karolherbst: which I don't like, but I see why that might be required
17:10karolherbst: but to do that each day was just too much
17:10karolherbst: I felt uncomfortable
17:10karolherbst: and calpping would have made me even fell worse
17:10HdkR: imirkin_: I've been waiting for DP 2.0. Need something to drive my planned 8k displays :P
17:11karolherbst: I am fine with it doing it on the first session first day, and the last session last day
17:11karolherbst: but what happens in between is for us
18:14imirkin_: Lyude: did you track down when that GPU started getting lockups?
18:15Lyude: imirkin_: got interrupted because of issues building one of the older kernel versions with gcc9. All I know is that it's present before 4.18
18:15Lyude: will continue investigating today
18:15imirkin_: Lyude: ah. so it was broken with 4.18 though?
18:16imirkin_: fwiw i do have a G84 plugged in at home
18:16imirkin_: and i'm fairly sure it's fine
18:16imirkin_: with 5.0 at least
18:16Lyude: imirkin_: curious, have you tried suspending/resuming while in fbcon?
18:16Lyude: or just in X/wayland
18:16imirkin_: oh, this is after suspend only?
18:16imirkin_: ok, i didn't realize that
18:17imirkin_: i'm a whole lot less surprised
18:17imirkin_: try going back to pre-4.3 for that.
18:17Lyude: big bisect
18:17imirkin_: well ... i'd just grab some distro liveusb's
18:17imirkin_: to get a rough bound on it
18:17imirkin_: i'm concerned it might be the big 4.3 rewrite
18:18imirkin_: is this s2ram or hibernate?
18:20Lyude: imirkin_: s2ram
18:23Lyude: karolherbst: do you remember how to specify compiler flags to the kernel makefile?
18:29karolherbst: HOSTCFLAGS I think?
18:29Lyude: karolherbst: it ended up being EXTRA_CFLAGS
18:29karolherbst: ohh, HOSTCFLAGS is for the binaries ran on the host
18:46AndrewR: sorry was away from keyboard because I asked my friend to look at what was wrong with my back (i developed some pain there)
18:57AndrewR: RSpliet, I think you (or Karol?0 omnce remarked on how uneasy it was to reclock modrn high-speed gpu memory - (gddr5 ?). So i thought this line of thinking probably can be moved closer to fron pages, if of course anyone still read this wiki ...
19:00AndrewR: may be I'm confused nouveau with other projects (re: monetary donations). Or may be it wa in really early ties... not sure how 'audience' users and devs supposed to communicate nowaydays? I think earlier testing requests were posted on ml and on site, with phoronix picking them up ....
19:05RSpliet: AndrewR: You're right, changing memory clock frequencies is tedious work.
19:05RSpliet: I think the wiki should still be the primary source of information for users.
19:06imirkin_: tedious work for the software developer who implements it
19:06imirkin_: fairly simple for the end user...
19:11AndrewR: I tend to agree with R. Stallman's ideas about users becoming (with time) developers at somewhat better rate than in usual closed-source environment...but actually making this potential reality into real reality not so simple ..for example, why most huans actually can't become developers? Basics are simple enough, yet real fixes not as frequent, from passing-by users/devs as one hoped ...
19:11imirkin_: that's like asking the square root of a million...
19:11imirkin_: no one will ever know
19:13AndrewR: http://libregraphicsworld.org/blog/entry/interview-with-developer-of-akira-free-interface-design-tool and probably other articles at the same blog (re: funding). I also thinl PS3 development currently funded via patreon (PS3 uses nv40-ear GPU, right?)
19:14imirkin_: most people with the skill level of developing nouveau are either college students or have full-time jobs
19:15AndrewR: square root of million must be 100, no? 100^2 = 1000 * 1000 = 1000 000 (million)
19:15AndrewR: 1000, not 100 (typo)
19:15imirkin_: that was from the simpsons.
19:15imirkin_: but it's also true - ask random people in your life the square root of a million. unless you're in a math department, no one will have a clue.
19:16karolherbst: truth is, there doesn't seem enough people interested to work on nouveau out of their spare time
19:17karolherbst: and nobody really has time to address this, or would even know how to address it
19:23AndrewR:scratches head and has no real idea to try ....
19:24AndrewR: I have time, but no skill, and probably this is something i can't just fix in my head (i really tried to wrap my head around how nouveau drives hw, yet come up with nearly nothing ..useful)
19:25karolherbst: usually the best approach to such things is to just start
19:25karolherbst: there is way too much to learn
19:26AndrewR: karolherbst, well, i compile mesa/kernel often, read changelogs and discussions, yet even simple plumbing was not something i actually finished :/
19:26karolherbst: I get the feeling that most wanting to start always have this need to be prepared for this, which is entirely pointless. Sure everybody has to learn stuff, but usually having a task first and then learn what's required in this particular case is the best approach for super complex areas
19:28karolherbst: AndrewR: then you weren't motivated enough
19:28karolherbst: or dedicated
19:29karolherbst: Lyude: is this actually fixed? https://bugs.freedesktop.org/show_bug.cgi?id=101778
19:29AndrewR: karolherbst, it seems I'm dedicated enough to whole linux (slackware) thing, was doing it since 2006, with bugreporting and pathes ..and special interest in gpus ..yet .....
19:29karolherbst: or maybe simply the wrong approach or something
19:30karolherbst: what I can only say is that a lot have this "I have to learn everything about Nvidia GPUs" urge
19:30karolherbst: and I am sure that's not helpful at all
19:33AndrewR: karolherbst, ma be just trying to add some vdec clock printing will help me ..next time when I'll compile my new kernel (soon).
19:33karolherbst: ohh, you wanted to fix reclocking in your first attempt to help nouveau?
19:33karolherbst: only insane people do this (like me) :p
19:34Lyude: karolherbst: note, I don't actually have a P51
19:34Lyude: I don't actually think anyone in RH does
19:34karolherbst: I guess I can ping on the bug
19:34karolherbst: but this was tested on a 4.11 kernel
19:34karolherbst: and we did tons of stuff since then
19:34Lyude: yeah-I wasgoing to say
19:35Lyude: if it still happens I might know a fix, there's been some very obscure mst bugs I've been finding while finishing up the s/r reprobe (+hotplug fixes, +more debugging tools, etc.) series I've been working on
19:35AndrewR: karolherbst, also do you run your kernel with preemption , or without? there seems to be three different levels in kconfig, and no/voluntary (my default for years) apparently were not uberstable over days/weeks ....last uptime was just 5 days but because power dies for second..so I'm gaining confidence for my card/mobo combination full preeempt kernel actually more stable ... (but is this default for other distro kernelrs or not?)
19:36AndrewR: karolherbst, no, just how I want to see real clocks for video decoder part ..it not printed anywhere ..even if internally set ....
19:36karolherbst: because that's usually not really useful for end users
19:36karolherbst: we could add an ioctl to retrieve GPU related information though
19:36karolherbst: it's just super hard to make that stable
19:37karolherbst: or we version it and just add new fields
19:37AndrewR: karolherbst, I was thinking about same file in debugfs where other clocks displayed
19:37karolherbst: mhhh, maybe we could add it to the pstate output
19:37karolherbst: because it's also changed through the perf levels
19:37karolherbst: so yeah.. we could just change that
19:37karolherbst: debugfs stuff isn't supposed to be stable anyway
19:37karolherbst: so we can just change the layout
19:38karolherbst: it's just a bit difficult to decide what's the actual clock as it's usually not just one engine and their clocks might differ
20:01AndrewR:mostly disappears in direction of bed ....
21:50Lekensteyn: karolherbst: by tracing ACPI calls you will see the value of CPEX (done here and in the next comment: https://github.com/Bumblebee-Project/bbswitch/issues/115#issuecomment-218551781)
21:50Lekensteyn: karolherbst: setting up kernel debugging can be done as described here: https://docs.microsoft.com/en-us/windows-hardware/drivers/debugger/setting-up-a-network-debugging-connection
21:51Lekensteyn: karolherbst: for various ACPI debugging extensions (requires a checked/debug version of Windows), see https://docs.microsoft.com/en-us/windows-hardware/drivers/debugger/acpi-debugging
21:52karolherbst: doesn't seem like if the value of CPEX ever changes