00:22cockroach: I've got an issue where my Xorg keeps freezing as soon as I boot it with a GT 710 (GK208B) ("FB_FLUSH_TIMEOUT"). I can't seem to find anything about such an issue on the gitlab tracker but figured I'd quickly ask in here if this was a known thing before I report it.
00:22cockroach: So, is this a known issue? Is there anything I can do?
00:23cockroach: also, dmesg at http://paste.debian.net/1189297/
00:24cockroach: happens with kernel 5.4, 5.10 and 5.12
00:26imirkin: cockroach: looks like accel is weirdly broken
00:26imirkin: is there anything particularly odd about this system?
00:27cockroach: not that I'm aware of.
00:27imirkin: it's a 32-bit kernel, that's the only even slightly odd thing that jumps out at me
00:27imirkin: is core2 duo 64-bit capable? i forget
00:27cockroach: yes it is
00:27cockroach: did I compile a 32bit kernel? that would be silly of me, hold on
00:27imirkin: i'd definitely recommend running a 64-bit kernel, even if you have 32-bit userspace
00:28imirkin: i could be reading this wrong too
00:28imirkin: it's been known to happen
00:28cockroach: menuconfig says "[*] 64-bit kernel"
00:28imirkin: can you just run 'uname -m' ?
00:28cockroach: x86_64
00:28imirkin: cool
00:28imirkin: i just can't read then.
00:29cockroach: :)
00:29cockroach: too bad, would have been nice if that had been the issue
00:29imirkin: (or rather, don't know which things to read)
00:29imirkin: so
00:29imirkin: [ 41.032044] nouveau 0000:01:00.0: fifo: CHSW_ERROR 00000002
00:29imirkin: that's basically the problem
00:29imirkin: that means the (nouveau-written) context switching firmware failed somehow
00:30imirkin: there's one thing to check
00:30imirkin: can you pastebin /sys/kernel/debug/dri/0/pstate ?
00:30cockroach: sure, hold on
00:30cockroach: hmm
00:30cockroach: No such file or directory
00:31cockroach: I don't have /sys/kernel/debug
00:32cockroach: any idea which kernel option would give me that?
00:32imirkin: CONFIG_DEBUGFS
00:32imirkin: and you have to mount it
00:33imirkin: although it's usually mounted
00:33imirkin: it's root-only, of course
00:33imirkin: debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)
00:34cockroach: ok, hold on, need to recompile.
00:34imirkin: so the theory is that it boots up in a super-low-perf state
00:34cockroach: might take a while :)
00:34cockroach: ah right
00:34imirkin: and something about that causes our ctxsw fw to timeout incorrectly/etc
00:35imirkin: i wanted to see what clocks you were set to, and if low, to suggest to force a reclock
00:35cockroach: sounds like a plan
00:35imirkin: although ... ctxsw timeout is usually what's reported when the firmware hangs
00:35imirkin: this is a legit error
00:35imirkin: i know all of nothing about this
00:35imirkin: =/
00:35cockroach: heh
00:36imirkin: but i'm what you got right now
00:36imirkin: unless skeggsb_ is back?
00:36imirkin: let me grep around to see if i can get some more detail on that error
00:36cockroach: I'll take any help I can. I think I would rather enjoy getting this little card to work.
00:36cockroach: aye
00:37cockroach: I have some more errors in dmesg now but I've been trying some stupid things. Still, I'll pastebin them, maybe they help.
00:37imirkin: well. FB_FLUSH_TIMEOUT suggests it could be a RAM-related issue
00:37imirkin: (FB = VRAM in nvidia parlance)
00:37cockroach: http://paste.debian.net/1189298/
00:38imirkin: interesting
00:38imirkin: that first error is ... impossible. well done.
00:38cockroach: interesting. I think the GT 710 exists both in DDR3 and GDDR5 variants, maybe I should have bought the DDR3 kind...
00:38cockroach: heh, thank you :)
00:39imirkin: i mean .. it's easily achieved, but requires someone to write some obviously bad code
00:39imirkin: unless you've been writing your own software to interact with nouveau directly, that wouldn't be the case
00:39cockroach: most certianly now
00:39cockroach: not*
00:39cockroach: video drivers scare me
00:39imirkin: UNSUPPORTED_KIND is like ... you're using the wrong memtype iirc
00:39cockroach: heh
00:40imirkin: which is easy enough to do. but our configuration for this stuff is pretty fixed and works.
00:40imirkin: anything else odd you can think of? lots of monitors? oddly connected? etc
00:40cockroach: two screens, one connected to DVI, the other one to HDMI (with an HDMI-to-DVI cable)
00:41imirkin: and the size of those screens?
00:41imirkin: looks like 1600x1200 for at least one of them?
00:41cockroach: 21" i think
00:41cockroach: yeah, both of them
00:41imirkin: (my favorite resolution ... brings back some fond memories of a Sony Trinitron 19")
00:41cockroach: mmh, Trinitron, i *HATED* carrying those around
00:42cockroach: still, amazing screens
00:42imirkin: i acquired a 21" of some sort around that time too, and it was WAY heavier than the trinitron
00:42imirkin: i had a legitimately hard time lifting it
00:42imirkin: vs the trinitron which was just very heavy
00:42cockroach: heh. I can imaging that, a 21" CRT, yikes.
00:43cockroach: imagine*
00:43imirkin: thankfully they made solid desks back in those days, none of this cheap ikea crap :)
00:43cockroach: :)
00:43imirkin: (so the desk didn't cave under the weight of 2 large CRTs...)
00:44cockroach: aye. also, my back was a *lot* less ... old ... back then.
00:44imirkin: cockroach: one thing i'd recommend is making sure you're using xf86-video-nouveau and not -modesetting
00:44cockroach: I am
00:45imirkin: you can quickly check by running 'xrandr --listproviders'
00:45imirkin: that should say nouveau at the end
00:45cockroach: if I use -modesetting the crashes / lockups happen a bit later, but they also happen
00:45imirkin: hmmm ok
00:45imirkin: weird.
00:45cockroach: hold on
00:45imirkin: it sounds like we might be missing some sort of init
00:45cockroach: need to restart X
00:45imirkin: nah, don't worry about it
00:45imirkin: you can also check in Xorg log
00:46imirkin: should be obvious - the lines will either be NOUVEAU(0) or modeset(0)
00:46cockroach: lots of NOUVEAU in there
00:46imirkin: ok
00:46imirkin: so that's like ... super-validated software
00:47cockroach: must be crappy hardware then
00:47imirkin: although i guess it could be a downstream glx client
00:47imirkin: well, it's definitely crappy hardware
00:47imirkin: GT 710 isn't exactly top-of-the-line
00:47imirkin: but this experience is very sub-par
00:48imirkin: i don't think we ever double-checked that these GK208B's come up in the same way as the non-B's
00:48imirkin: in terms of init
00:48cockroach: hmn, all I really need are some video outputs that deliver reasonable 2d performance without generating a lot of heat, so I figured a GT710 might be the right thing for me.
00:49imirkin: in general, you're right
00:49imirkin: there's just something odd going on here
00:49imirkin: we're missing something.
00:49cockroach: (still compiling that kernel, btw)
00:49imirkin: if it's something you feel like doing
00:49imirkin: you could also load up the nvidia blob driver, and get a mmiotrace
00:50imirkin: which would allow us to verify whether we're missing some bit of init
00:50cockroach: oh yikes, that thing.
00:50imirkin: you're not the first person to run into issues with the GT 710
00:50imirkin: i have a GK208B too, and also had stability issues
00:50imirkin: however those issues tended to manifest after many days
00:50cockroach: heh, that's because my PC is so very fast
00:51imirkin: mine's not _much_ newer...
00:51imirkin: i7-920
00:51cockroach: ah nice, almost as old indeed
00:52imirkin: i don't update often ... only when i need to
00:52cockroach: makes sense
00:53cockroach: okay, here's what /sys/kernel/debug/dri/0/pstate says
00:53cockroach: 07: core 405 MHz memory 810 MHz
00:53cockroach: 0f: core 653-954 MHz memory 5010 MHz
00:53cockroach: AC: core 405 MHz memory 810 MHz
00:54imirkin: ok, so it seems to come up in a reasonable state
00:54imirkin: (last line)
00:54imirkin: you can try doing "echo 0f > ..."
00:55imirkin: and see if that helps matters at all
00:55cockroach: that would put it in max power mode?
00:56imirkin: yes
00:57cockroach: I do have dmesg errors but Xorg seems to work. Currenly playing a video.
00:57imirkin: what are the errors about?
00:57cockroach: they might have been from before, hold on
00:57cockroach: I'll try setting 0f before Xorg is started
00:57imirkin: ok
00:57imirkin: another thing
00:57imirkin: is that that summary isn't perfect
00:58imirkin: there are like 1000 settings
00:58imirkin: and it only prints a handful
00:58imirkin: so even if the 07 level looks exactly like AC, it might still be different
00:58imirkin: so if the 0f thing works out well
00:58imirkin: the next test woudl be to see if echo'ing 07 has the same (positive) effect
00:59cockroach: ah right, that would make sense
01:01cockroach: so far I'm not getting any errors in dmesg (with 0f)
01:01cockroach: I'll try switching to 07 now
01:02cockroach: looking good so far
01:02cockroach: ah
01:02cockroach: now it crashed
01:03cockroach: "find /" did it :)
01:08imirkin: find / might crash it for other rasons
01:09imirkin: not nouveau-realted ones
01:09imirkin: once you start pokign around in /sys or debugfs
01:09cockroach: maybe 'find /usr' would be less stupid
01:09imirkin: there's a same-filesystem thing
01:10cockroach: in find? i thought that was in tar :)
01:10imirkin: definitely somewhere :)
01:10cockroach: ah '-mount' looks reasonable
01:11imirkin: maybe it was rsync?
01:11cockroach: heheh
01:11cockroach: could be
01:12imirkin: anyways - if setting it to 07 still has fail
01:12imirkin: that means we're probably not doing some power thing it expects us to be doing
01:12cockroach: okay, with "0f" I don't seem to get any issues, as soon as I return to "07" I get errors in dmesg
01:13imirkin: and it's right at some power budget / clock budget / etc limit
01:13cockroach: time to install the evil driver then?
01:13imirkin: and so nouveau's lack of some functionality breaks it
01:13imirkin: no
01:13imirkin: that won't help
01:13cockroach: \o/
01:13imirkin: i mean
01:13imirkin: it would help if you wanted to figure out what we were missing
01:13imirkin: but it'd be a lot more than a single mmiotrace to solve it
01:13cockroach: ah, yeah, that's the plan
01:13imirkin: i think it's a more dynamic problem
01:13imirkin: but i dunno - maybe not
01:14imirkin: Lyude: did you ever push the BLCG stuff? were there bits leftover?
01:14cockroach: i will *not* be using the nvidia driver for everyday stuff, I'm too proud for that :)
01:14imirkin: (that's "clock gating", for those following at home)
01:15imirkin: cockroach: tbh, i'd recommend you find someone who's interested in consuming those traces
01:15imirkin: before yuo go to the trouble
01:15imirkin: to be up-front about it, i won't be looking at them
01:15cockroach: imirkin: makes sense
01:15cockroach: aye
01:15cockroach: in case noone cares, any recommendations about fanless, low-power cards that just work? :)
01:16imirkin: oh whoa, this is fanless?
01:16cockroach: ah, yep, it is
01:16imirkin: and gddr5? keep an eye on the temp of the card
01:16cockroach: bad idea?
01:16imirkin: it might not be designed to run at the highest clock 100% of the time
01:16imirkin: we don't have automatic reclocking
01:16cockroach: oh yikes, it's running hotter than my radeon. 66°C atm
01:17imirkin: i'd be more concerned around 90
01:17cockroach: slowly getting there, 67 now :)
01:18cockroach: putting it on 07 mode helps
01:18imirkin: so that 0f mode actually has a range of frequencies
01:18imirkin: and we set the highest one
01:18imirkin: we could allow one to set a lower freq in the range
01:20imirkin: one could reasonably assume that lower freq = less temp
01:22cockroach: indeed
01:22cockroach: and be "we could allow" you mean, modify the driver? or is there an option somewhere?
01:23cockroach: (it's down to 56°C now)
01:23imirkin: modify the driver, i'd think
01:23imirkin: i don't remember it allowing this option
01:23imirkin: but it's been quite a while since i've looked at that logic
01:23imirkin: karolherbst: --^
01:23cockroach: hmn, somehow I don't think I would like to keep maintaining my own personal nouveau branch...
01:23imirkin: you'd upstream it
01:24imirkin: there's no reason for it to work the way it currently does - just easier that way in terms of config
01:24imirkin: otherwise you have to come up with a way to let the user specify the frequency, for example
01:24cockroach: yep
01:25imirkin: i don't know if hacking on kernel drivers is within your capabilities
01:25imirkin: but it shouldn't be very difficult, i'd think
01:25cockroach: I've done it before but it scares me a little
01:26imirkin: what's the worst that could happen? you screw up swapfile and corrupt your FS? oh wait, that was 5.12-rc1
01:26cockroach: I'm quite comfortable programming in C though, so I should be fine
01:26cockroach: heh :)
01:26cockroach: good thing I wasn't trying to debug this with the latest kernel a few days ago
01:34imirkin: i think you can also find some GT 1030's which are fanless. those use nvidia-supplied firmware which doesn't support reclocking at all.
01:35cockroach: yeah I saw those and was tempted, figured they'd run hotter than the GT 710 though.
01:36cockroach: also, when you say reclocking, that would imply that they always run of "0f" mode? or is that something else entirely?
01:37imirkin: it implies that it runs at whatever it runs at
01:37imirkin: which in the case of modern GPUs is the lowest clocks
01:37cockroach: that doesn't sound too bad then
01:38cockroach: you mentioned other issues with the GT710 before, things that might crash it after several days etc. - are you aware of similar problems with the 1030? or is that a slightly less bad card in general?
01:39imirkin: well, more bad in that it requires signed firmware
01:39imirkin: but IME it's been more stable than the GK208B that i have (which iirc is a GT 730 in terms of marketing)
01:39imirkin: although that one has DDR3. surprising that you'd have GDDR5 which is marketed as GT 710
01:40cockroach: bad marketing?
01:40imirkin: :)
01:40imirkin: marketing is always bad.
01:40cockroach: heh indeed
01:41cockroach: signed firmware *does* sound stupid
01:42cockroach: *sigh*
01:44cockroach: makes me wish there were alternatives.
01:44imirkin: i hear intel is about to come out with a board
01:44imirkin: the AMD fanless options are *very* old GPUs (terascale)
01:44damo22: signed fw isnt a bad idea, so long as the key is replaceable by the end user
01:45cockroach: imirkin: I *do* have a very old fanless radeon sitting here, but it seems to be too slow even for normal desktop stuff
01:45imirkin: do you know what chip it has?
01:46cockroach: damo22: heh yep, now that would be a fun world
01:46imirkin: iirc the "latest" ones are like evergreen family (CEDAR or whatever)
01:46cockroach: imirkin: RV570 IIRC
01:46imirkin: ah yeah, that's old.
01:46imirkin: that's pre-r600
01:46cockroach: yup
01:46imirkin: which is beyond ancient
01:46imirkin: the ones i'm talking about are DX11-capable
01:46imirkin: but are supported by the r600 mesa driver
01:46damo22: i have a Navi RX5700 but i think opencl does not work on it yet
01:46imirkin: HD 6540 or whatever
01:46cockroach: I was doing some audio stuff and kept getting buffer underruns when I had that video card installed. that was unexpected.
01:47imirkin: modern desktops have gone cuckoo for gpu accel
01:47cockroach: imirkin: ah indeed, I recall someone having a HD 6something with passive cooling.
01:47cockroach: well, it's just i3. but still
01:47imirkin: ok, but like gtk3/gtk4, qt, etc
01:48imirkin: damo22: i'm guessing that RX5700 isn't exactly fanless.
01:48cockroach: imirkin: ah yep, those
01:49cockroach: damo22: looks like a fun little card. it would most certainly be very bored in my PC, waiting for the CPU to do something
01:50damo22: lol
01:51damo22: i was hoping to use it to learn opencl and write some gpu accellerated stuff for work
01:52cockroach: imirkin: to me it's a good sign when a card with passive cooling does not require the additional power cable, that sort of implies that it might not waste a whole lot of power. I think the passive radeons do have those.
01:52damo22: but the AMD opencl stack is a bit crazy
01:53cockroach: don't you need their proprietary driver for that?
01:53damo22: one day, no
01:53cockroach: nice
01:53damo22: its just not completed for my particular model
01:56damo22: i refuse to use binary drivers
01:56damo22: i guess that means im stuck with hardware i cant use yet
01:57cockroach: heh. well I was trying to get away from the silly binary blobs required by the radeon driver but so far without much success
01:58damo22: considering i am contributing to a GPLv3 operating system, it would be nice to have free drivers for all the hardware it supports
02:00cockroach: indeed
03:15cockroach: imirkin: I opened a bug report at https://gitlab.freedesktop.org/drm/nouveau/-/issues/74 -- just in case you have any further ideas.
03:15cockroach: also, I'm off, it's getting a bit late.
03:15imirkin: cockroach: thanks
03:15cockroach: thank you for all the help
03:22imirkin: yw
08:47karolherbst: imirkin: yeah soo... this crashing in default state is a bug which is quite rare, but not new
08:48karolherbst: I talked with skeggsb_ about and it is probably an initialization issue and we are simply forgetting something to initialize or something
13:35imirkin: karolherbst: which one?
13:36imirkin: karolherbst: the GK208B one above? it doesn't work when he reclocks to 07, so it's not about init
13:51karolherbst: ohh, so stock clocks are okay?
13:51imirkin: not sure what you mean
13:52imirkin: both init and 07 cause problems
13:52imirkin: 0f works fine
13:52karolherbst: like if no reclocking happens
13:52imirkin: init boots to clocks that are a lot like 07 (perhaps identical, but hard to tell just by looking at pstate)
13:54karolherbst: imirkin: right, but we had the case where somebody complained about default clocks not working and having to clock to 0f as well, but turns out, that this wasn't sufficient afterall
13:54imirkin: karolherbst: the GM108 guys right?
13:54imirkin: https://gitlab.freedesktop.org/drm/nouveau/-/issues/61
13:55imirkin: for the person in question, booting to 07 + disabling runpm (which isn't a concern here as it's a desktop) was enough to fix it
13:57karolherbst: yeah.. in this case our init is borked somewhere
13:57imirkin: right.
13:57imirkin: er
13:57karolherbst: but could also be that our reclocking code isn't all that perfect either
13:57imirkin: sure
13:57imirkin: but the fact that 07 AND init have the same issues...
13:58imirkin: leads one to believe the issue isn't with clock config
13:58imirkin: that's why i'm thinking BLCG/etc
13:58imirkin: i forget if Lyude landed the stuff
13:58karolherbst: ahhm yeah maybe
13:58karolherbst: I think it isn't enabled for maxwell
13:58imirkin: that's ok
13:58imirkin: this is GK208
13:58karolherbst: but I was having issues with it on one board
13:58karolherbst: so yeah.. maybe disabling it could fix issues
13:59imirkin: oh hm
13:59imirkin: i was thinking it'd be the solution
13:59imirkin: but i guess it could be the problem
13:59karolherbst: on a maxwell board I mean
13:59imirkin: was there a config setting?
13:59imirkin: oh
13:59karolherbst: the kepler code could be fine
13:59imirkin: i'm talking about the GK208 guy
14:00karolherbst: yeah, but just saying that the blcg stuff can cause issues
14:00imirkin: a bunch of people have had issues with these GT 710's
14:00imirkin: right
14:00karolherbst: mhh, it is off by default I think..
14:01imirkin: seems on by default
14:01karolherbst: really?
14:01karolherbst: clkgating_enabled
14:01imirkin: nvkm_therm_clkgate_enable is called from nvkm_device_init
14:01karolherbst: check the nvkm_therm_ctor
14:02imirkin: oh right
14:02imirkin: cool
14:02karolherbst: ehh, wait
14:02karolherbst: there are two flips
14:02karolherbst: but yeah
14:03karolherbst: if (!therm || !therm->func->clkgate_enable || !therm->clkgating_enabled) return;
14:03imirkin: right
14:03imirkin: clkgate_enable is whether it's supported by the driver
14:03karolherbst: anyway, looks off by default
14:03imirkin: right
14:04imirkin: left a note in the issue
14:04karolherbst: the clocking stuff was never fully understood anyway, and kepler2+ have additional stuff we don't do :/
14:05imirkin: easy enough to test it
14:06karolherbst: yeah well. The things I know about are breaking things at super high clocks
14:06karolherbst: but our clock calculation is also not correct on those gens
14:06karolherbst: so... could be anything really
14:06karolherbst: worst case we need mmiotraces from the blob and nouveau
14:07karolherbst: but mhh...
14:07RSpliet: That's not a terrible worst-case...
14:07karolherbst: it looks like memory related
14:07karolherbst: skeggsb_ mentioned that there are a few bits we don't do there as well
14:10imirkin: ok. well they're not exactly rare boards.
14:10karolherbst: imirkin: oh, I was refering to bits in the vbios
14:10karolherbst: some config we can't deal with
14:10imirkin: ah
14:11imirkin: but it doesn't work at init
14:11imirkin: so ... :)
14:11karolherbst: so.. there are three things: 1. clock reading/writing function is broken on newer args 2. unhandled vbios bits in regards to mem reclocking. 3. potentially an init parsing issue
14:12imirkin: or the most likely
14:12imirkin: 4. we're missing some kind of init
14:12imirkin: (like e.g. the blcg bits, etc)
14:12karolherbst: yeah well.. I included that in 3.
14:12imirkin: ah, that sounded like vbios parsing
14:12imirkin: of the init scripts
14:12imirkin: which i don't think is likely a problem
14:13karolherbst: well.. the init scripts should handle that in theory
14:13imirkin: yes
14:13karolherbst: they also reclock, soo...
14:13imirkin: but in practice, many are tested in conjunction with driver
14:13imirkin: so they wouldn't notice they screwed something like that up
14:13imirkin: if the driver just fixes it up for them
14:16karolherbst: right
18:00kreyren: still struggling with https://github.com/Kreyren/kreyren/issues/75
18:00kreyren: apparently it's getting me a terribile performance too
18:02kreyren: (12FPS +38ms frametimes in Witcher 2 that previously worked on 230FPS )
18:02kreyren: even with:
18:02kreyren: kreyren@dreamon:~$ cat /proc/cmdline
18:02kreyren: audit=0 intel_pstate=disable BOOT_IMAGE=/boot/vmlinuz-5.11.0-6.2-liquorix-amd64 root=UUID=0afb1f96-f2b7-4c1b-8042-4da16a84fb60 ro nouveau.config=NvBoost=2 quiet
18:05diogenes_: kreyren, try also: nouveau.config=NvClkMode=15
18:05kreyren: what is that doing? O.o
18:05RSpliet: kreyren: multithreading is as broken as it was in 2018
18:06kreyren: RSpliet: eh?
18:06diogenes_: putting the card in a higher power state.
18:06karolherbst: kreyren: not sure if you really want "intel_pstate=disable" as this will probably reduce your performance
18:06RSpliet: kreyren: nouveau does not support multithreaded GL workloads. Other bugs you linked to implied that this is the culprit
18:06karolherbst: falling back to ACPI disables quite a lot
18:06kreyren: <karolherbst "Krey: not sure if you really wan"> this is set in a kernel as afaik it handles it differently
18:06kreyren: (liquorix kernel)
18:07karolherbst: sounds even worse
18:07karolherbst: depending on the CPU it is all in hw anyway
18:07kreyren: will use debian-distributed on reboot then
18:07kreyren: <karolherbst "depending on the CPU it is all i"> i7-2600K
18:07karolherbst: ahh.. thats old enough that it won't matter
18:07kreyren: o.o
18:08karolherbst: just digged into it recently with my CPU and using the ACPI one made everything worse
18:08karolherbst: but it's also 10th gen
18:08kreyren: what do you want me to do then? just add nouveau.config=NvClkMode=15 ?
18:08karolherbst: yeah
18:09karolherbst: but you can also change it at runtime
18:09RSpliet: karolherbst: there's an absolute metric shitton of errors in the log. I suspect they're more of a bottleneck than the CPU
18:09kreyren: <karolherbst "but you can also change it at ru"> how?
18:09karolherbst: kreyren: as root: echo 0xf > /sys/kernel/debug/dri/0/pstate
18:10karolherbst: RSpliet: yeah.. probably multithreading
18:10kreyren: root@dreamon:/home/kreyren# echo 0xf > /sys/kernel/debug/dri/0/pstate
18:10kreyren: bash: /sys/kernel/debug/dri/0/pstate: No such file or directory
18:10karolherbst: good that I don't have time to continue working on the fixes at least this week :/ maybe I will but last week was annoying
18:11karolherbst: kreyren: is this a laptop?
18:11RSpliet: karolherbst: that's my first hunch. Not sure if Witcher 2 does that. Also, they all fault on the exact same class/method
18:11karolherbst: probably not
18:11karolherbst: kreyren: ohh.. is debugfs even mounted...
18:12kreyren: <karolherbst "Krey: is this a laptop?"> desktop
18:12kreyren: <karolherbst "Krey: ohh.. is debugfs even moun"> what's debugfs
18:13kreyren: doesn't sound like FHS3_0 to me
18:13karolherbst: it's linux specific
18:13karolherbst: a filesystem mounted under /sys/kernel/debug/
18:14kreyren: < https://matrix.org/_matrix/media/r0/download/matrix.org/GhkWPRKrKJdEPAXZWEbZidtd/message.txt >
18:16karolherbst: ahh yeah.. ehh
18:16karolherbst: yeah, maybe adding the kernel parameter is easier than
18:23kreyren: oke
18:26kreyren: testing agains glxgears with vlank_mode=0 seems to get me same performace, trying the game
18:29kreyren: 17FPS ~60ms frametimes
18:29karolherbst: yeah.. glxgears usually bottlenecks on the CPU
18:29karolherbst: mhh
18:29kreyren: somewhat of an improvement i believe
18:29karolherbst: wait... what GPU is that even? forgot to check that
18:30kreyren: GTX970
18:30karolherbst: ahh mhh, crap...
18:30kreyren: (FWIW target is 30FPS <25ms frametime)
18:30karolherbst: okay.. soo...
18:30karolherbst: here is the deail
18:30karolherbst: *deal
18:30karolherbst: since 2nd gen maxwell, which the 970 is part of, Nvidia prevents us from controlling the fans
18:30karolherbst: we require signed firmware for that
18:30karolherbst: sooo... we don't enable reclocking on those GPUs
18:31karolherbst: because users might turn it on and constantly overheat their GPUs
18:31kreyren: so i need to get that binary from nvidia sources for my PMU?
18:31karolherbst: (and then run into thermal throttling, which reduces perf even more)
18:31kreyren: x.x
18:31karolherbst: well.... in addition to that, we would also need to wire up support for that in nouveau, which we didn't do to this point because we never needed it
18:32karolherbst: before that we can just handle that in the kernel
18:32karolherbst: no firmware required at all
18:32kreyren: so what do you suggest
18:32karolherbst: yeah.. it's a stupid situation
18:33karolherbst: well... I have no good solution as it would be equally terrible like the video acceleration firmware every user has to download as well on supported GPUs
18:33kreyren: worst case scenario i have equipment to work on electrical components and 4 years or experience with SMD soldering
18:33kreyren: seems more appealing to me then using garbage from nVidia
18:33karolherbst: externally controlled fan would do the trick + patching the kernel
18:33kreyren: patching the kernel how
18:33HdkR: Watercooling block is a solution as well :)
18:33karolherbst: I have patches somewhere...
18:33karolherbst: but yeah.. cooling is the more pressing issue
18:34kreyren: is there something i can put jumper wires to that would allow me to control the fans so that it's not as loud on demand?
18:34karolherbst: uhm... depends on how much you want to mess with your GPU
18:35karolherbst: I am sure the connected fans are just normal 2/3/4 pin connected fans
18:35kreyren: whatever is needed as long as i don't have to use nVidia
18:35kreyren: *nvidia drivers
18:36kreyren: i can change the PMU even!
18:36karolherbst: imirkin: seems like I have all my stuff on this branch, but there is more than just the reclocking thingies: https://github.com/karolherbst/nouveau/commits/clk_cleanup_for_real
18:36karolherbst: kreyren: on the GPU? I doubt that
18:36kreyren: got a new debugger for SPI flashing PMUs recently ^-^
18:37kreyren: <karolherbst "Krey: on the GPU? I doubt that"> i meant the chip itself O.o
18:37kreyren: for something that is compatible with the PCB while us having the firmware for it
18:37karolherbst: well, good luck finding the PMU on the GPU then
18:37kreyren: x.x
18:38kreyren: could do a custom PCB if i knew the pinning for the core too
18:38karolherbst: also, everything is highly integrated, so not as easy
18:38karolherbst: and the issue is still, that there is crypto stuff going on
18:38karolherbst: probably easier to just throw water cooling onto the GPU and be done with it
18:39kreyren: can you recommend a water pump for that? I can make a waterblock at house
18:40karolherbst: never done that myself
18:40karolherbst: I am sure HdkR has :p
18:40kreyren: the current cooler is trash anyway.. it doesn't have sufficient cooling for VRAM that is causing overheating issues
18:40HdkR: Watch out, you want to make sure to cool both the die, memory, and VRMs. Basically duplicating EKWB's designs https://www.ekwb.com/shop/ek-fc970-gtx-acetal-nickel
18:41kreyren: <HdkR "Watch out, you want to make sure"> i have access to a machine shop so i want to make it custom
18:41karolherbst:should really clean up all the reclocking patches *ugh*
18:41HdkR: Which you would need a radiator and fans strapped to that radiator as well. Just running water over it would soak the water and not move the heat out of the case
18:41karolherbst: kreyren: worst thing is just, that I probably won't be able to find enough time to port my old patches to the newest kernel
18:42kreyren: karolherbst: If you can provide what needs to be done then i might be able to finish it or provide a quote to bug bounty and bug.
18:43karolherbst: well, this branch, but it has a ton of other stuff in it, which is... more or less related: https://github.com/karolherbst/nouveau/commits/clk_cleanup_for_real
18:43karolherbst: but some can be dropped as well
18:43kreyren: dropped?
18:43karolherbst: well, similiar patches landed
18:44kreyren: i see
18:44HdkR: If you do pumps, you don't really want to stick with aquarium grade things these days. Most PC watercooling parts use "D5" or D5 clone pumps. Looking at custom water cooling bits to get an idea what is necessary. https://www.corsair.com/us/en/Categories/Products/Custom-Cooling/Pump-Reservoir/Hydro-X-Series-XD5-RGB-Pump-Reservoir-Combo/p/CX-9040006-WW
18:46kreyren:doesn't like pumps made for computer use, he prefers quiet industrial as they are more robust and economical
18:48karolherbst: just use 50l of this cooling oil :p
18:48HdkR: ugh. I never want to do an oil cooled pc
18:48karolherbst: it is silent though
18:48karolherbst: if you can ignore the bulbbing
18:48HdkR: Maintenance is a nightmare though
18:48karolherbst: *bubbling
18:49karolherbst: pffft
18:49HdkR: and the oil eats /everything/
18:49karolherbst: buy better oil
18:49karolherbst: or well.. use 99.99% Isopropanol :D
18:49HdkR: I mean, it leeches in to the case material over long periods of time
18:50karolherbst: get rid of the case then
18:50HdkR: After a year, yeet the whole PC. It's about the only thing that can be onde at that time :D
18:50HdkR: done*
18:50karolherbst: :D
18:51juri_: nah. sell it on ebay. :)
18:51kreyren: karolherbst for what kernel version did you made those patches?
18:51karolherbst: uhm.....
18:51karolherbst: 4.... 20?
18:52HdkR: 3M has some liquid that can be used for whole PC dunking as well. Just needs a condensation chamber which is neat
18:52HdkR: https://www.3m.com/3M/en_US/novec-us/applications/immersion-cooling-of-power-electronics/
18:53karolherbst: good luck if your SSD nreaks
18:53karolherbst: *breaks
18:53karolherbst: I mean, it really doesn't matter what you use, once you have to replace something it's messy
18:54HdkR: aye
18:57kreyren: Almost like it was made for this!
18:57kreyren:uploaded an image: (93KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/OyBPLxtuNTAPVxyRISfSzFPe/IMG_20210314_195703.jpg >
18:57karolherbst: kreyren: also.. probablye heaper to just buy a fast AMD GPU... just saying
18:57karolherbst: *probably cheaper
18:57karolherbst: :D
18:58karolherbst: or that
18:58karolherbst: I mean... the driver even reports the GPU temp, so userspace cooling should be possible
18:58kreyren: <karolherbst "Krey: also.. probablye heaper to"> Fast AMDGPU I have on the table with a hugged up BIOS on PCB this is backup
18:59karolherbst: the reason the branch is that huge is, that I was also implementing throttling and all this fun stuff
19:02kreyren: Hehe
19:02kreyren:uploaded an image: (103KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/ppQRgQjdoTuTZhXNweFYwZcn/IMG_20210314_200204.jpg >
19:02kreyren: 12V1A enough u think ?
19:03HdkR: Watch out if those are 5v fans
19:03RSpliet: yeah immersion cooling is cool... until it isn't. Contaminants and high-temperature can make a liquid conductive, at which point your machine is essentially EOL
19:03kreyren: <HdkR "Watch out if those are 5v fans"> How do I know
19:03RSpliet: *high temperature chemical reactions
19:03kreyren: Looks like 12V based on the intensity of the 9V battery
19:03karolherbst: hum.. mhhh
19:03HdkR: Rating is usually on the underside of the motor. Which would require taking the whole fan out
19:03karolherbst: HdkR: why would it be 5V though?
19:04karolherbst: so PCIe gives you 12V and it kind of is simplier to just use that directly for the fans? but yeah
19:04karolherbst: they could downconvert for... stupid reasons
19:04HdkR: stupid reasons, yes
19:05kreyren: 12VDC
19:05kreyren:uploaded an image: (128KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/hzyRIVzLGpJOtTbMnoQGygnM/IMG_20210314_200447.jpg >
19:05HdkR: Safe then :)
19:05karolherbst: kreyren: your motherboard has 12V fan connectors, no?
19:05karolherbst: just abuse one of them :D
19:06karolherbst: (and see if you can control them via linux)
19:06kreyren: Spare PSU for my router is 12VDC1A what a coincidence
19:06karolherbst: ohh, I mean a motherboard usualy has like 6 connectors for fans :p
19:06karolherbst: why deal with something external
19:07karolherbst: or... connector it to the CPU one, because chances are high that when the CPU spins up, the GPU should as well
19:07karolherbst: mhh
19:07karolherbst: maybe better not
19:08kreyren: Ah that PSU is 12VCD5A
19:08kreyren: Is that bad
19:08karolherbst: you want something you can control
19:09karolherbst: because full power all the time is quite annoying
19:09kreyren: >0.30AMPS
19:09kreyren: Probably very bad lol
19:10kreyren: <karolherbst "you want something you can contr"> Proof of concept is good enough atm I will add a knob to it after
19:12karolherbst: ahh
19:12karolherbst: next step will be to port the patches over :D
19:12karolherbst: although you really don't need much
19:12karolherbst: https://github.com/karolherbst/nouveau/commit/1883fba75c633e84fa9f8ab687af4793f2650d06
19:13karolherbst: https://github.com/karolherbst/nouveau/commit/241342504559eb76bba49fe6e3bd906dd9ea32df
19:13karolherbst: that's enough to enable it for your board
19:13karolherbst: the rest is just... for safety
19:15kreyren: 56K Ohm resistor enough u think?
19:16karolherbst: dunno
19:36kreyren: Yay beauty
19:37kreyren:uploaded an image: (107KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/jwEcJmBdfAFMqFGSGPwOFcbV/IMG_20210314_203620.jpg >
19:46kreyren: karolherbst: What patches do you want me to use? just https://github.com/karolherbst/nouveau/commit/241342504559eb76bba49fe6e3bd906dd9ea32df ?
19:46kreyren: and again 4.20 tag on linux kernel?
19:47kreyren: *against
19:48karolherbst: yeah.. dunno
19:48karolherbst: might be easier to port them over to the newest kernel
19:48kreyren: x.x
19:49karolherbst: the patches aren't that complicated, I just don't know if they work as intended anymore
19:49karolherbst: but.. they could
19:49karolherbst: maybe I will find some time over the week, because it's not the first time people ask about this
19:50kreyren: aaaa
19:53kreyren: i try the 4.20 in the meantime
19:54kreyren: Can you confirm that this issue it caused by nouveau being unable to control the fans? https://github.com/Kreyren/kreyren/issues/75
19:54kreyren: Or is it something different?
19:54kreyren:wants to update tracking
19:55karolherbst: nope, has to be something else
19:55karolherbst: probably just multithreading
19:55karolherbst: I have a mesa branch for this though
19:55kreyren: o.o
19:55karolherbst: https://gitlab.freedesktop.org/karolherbst/mesa/-/commits/nvc0_threading/
19:55karolherbst: untested (tm)
19:55kreyren: i like to live dangerously (tm)
19:56karolherbst: well, I did actually test it, but never installed it systemwide
19:56karolherbst: so it might just work
19:56karolherbst: or you could run into random crashes...
19:56karolherbst: dunno
19:56karolherbst: kreyren: what applications are you running? anything chromium based or so?
19:57karolherbst: like slack or... whatever other webpage wrappers people install
19:57kreyren: <karolherbst "Krey: what applications are you "> not that i am aware of i try to avoid chromium at all costs, but steam is using it
19:58karolherbst: mhhhhh
19:58karolherbst: dunno if steam can trigger those issues, but might
19:58karolherbst: it is a bit hard to pinpoint those crashes :/
19:58karolherbst: could also be missconfigered memory or something
19:58HdkR: Becoming more and more likely as more the UI gets replaced with CEF
19:59karolherbst: yep
19:59karolherbst: hence me focusing on the mt bug if nothing more important comes around and steals my focus from this bug :p
20:00karolherbst: "more important" == "we really have to fix this other bug"
20:01karolherbst: so apparently somebody thought it is a good idea to accept tegra/nouveau related bugs as beta blockers for Fedora :/ *sigh*
20:03ccr: why do you care?
20:03karolherbst: because work stuff
20:03HdkR: Work means you care about what work cares about :)
20:03karolherbst: :p
20:03karolherbst: well, kind of
20:03karolherbst: the bug was quite trivial
20:04karolherbst: but still
20:05karolherbst: the other one I am dealing atm is nouveau preventing kdump to dump stuff :)
20:07kreyren: kreyren@dreamon:~/Repositories/linux$ find . -type f -name nouveau
20:07kreyren: kreyren@dreamon:~/Repositories/linux$
20:07kreyren: ehh where this should be in the linux kernel hierarchy? x.x
20:07kreyren: or should it be a loadable module ?
20:07karolherbst: drivers/drm/ but this is an out of tree module tree
20:07karolherbst: so you should be able to jsut compile against your kernel
20:08kreyren:has no idea how linux works, but he knows programming
20:08kreyren: kreyren@dreamon:~/Repositories/linux$ ls drivers/drm
20:08kreyren: ls: cannot access 'drivers/drm': No such file or directory
20:08karolherbst: uhm..
20:08karolherbst: drivers/gpu/drm
20:08HdkR: drivers/gpu/drm/nouveau
20:08kreyren: thanku
20:09kreyren: that file hierarchy is different
20:09kreyren:uploaded an image: (118KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/PmnrlLgVCLVNgLawBoHRjXes/image.png >
20:10kreyren: x.x
20:10kreyren: ah i see
20:10kreyren: it's drm/nouveau in the repo
20:10karolherbst: hence me saying drivers/gpu/drm :D but you can probably just symlink nouveau or so :/
20:10karolherbst: yeah...
20:11karolherbst: but it will be easier to compile as an external module
20:11karolherbst: most likely
20:11kreyren: so just make install will work?
20:11karolherbst: potentially
20:12karolherbst: requires rebuilding initramfs
20:12karolherbst: and then you fight with load ordering and stupid crap :/
20:12karolherbst: sadly none of this is very straightforward or easy
20:12kreyren: x.x
20:14kreyren: kreyren@dreamon:~/Repositories/nouveau$ make 2>&1 | curl -F 'f:1=<-' ix.io
20:14kreyren: http://ix.io/2SQP
20:14kreyren: what GCC did you use for it?
20:15karolherbst: ehh...
20:15kreyren: x.x
20:16karolherbst: let's see
20:16karolherbst: mhh
20:16karolherbst: heh
20:16karolherbst: list_for_each_entry_from_reverse should exist
20:17kreyren:replaced it with suggested and it continues to build
20:17karolherbst: it was added in 4.11
20:17karolherbst: yeah well
20:17kreyren: > /bin/sh: 1: envyas: not found
20:17karolherbst: replaciung it is wrong
20:17kreyren: ah
20:17karolherbst: ohhh
20:17karolherbst: kreyren: you have to run make inside drm/
20:18karolherbst: not in the top root
20:18kreyren: ah oke
20:18kreyren: <karolherbst "replaciung it is wrong"> what do you suggest then?
20:18karolherbst: nothing, it should just compile when inside drm/ D
20:18karolherbst: LD
20:18karolherbst: ;ds.d.asd.
20:19karolherbst: :D
20:19kreyren: o.o
20:19kreyren: like first building drm and then `make` from root?
20:19karolherbst: nope
20:19karolherbst: just run make inside the drm dir
20:19karolherbst: or make -C drm
20:21kreyren: o.o
20:21kreyren: oke
20:21kreyren: kreyren@dreamon:~/Repositories/nouveau/drm$ make 2>&1 | ix
20:21kreyren: http://ix.io/2SQR
20:21kreyren: implicit declaration of function ‘pci_platform_rom’;
20:21karolherbst: okay..
20:22karolherbst: that might be too new then.. let's see
20:22karolherbst: or too old
20:22kreyren: o.o
20:23karolherbst: mhh, looks like that function got removed
20:24karolherbst: kreyren: sure you are on 4.20?
20:24karolherbst: by default make picks up the build dir of the currently running kernel
20:24kreyren: kreyren@dreamon:~$ uname -r
20:24kreyren: 5.11.0-6.2-liquorix-amd64
20:24kreyren: want me to get 4.20 ?
20:24karolherbst: so you have to trick it into using something older
20:24kreyren: o.o
20:24karolherbst: yeah.. could be.. I am not 100% sure if it was 4.20 or not
20:24karolherbst: could have been 4.19 or 5.0
20:25karolherbst: let's see
20:25kreyren: 4.17 from the github
20:25kreyren: wait no
20:25karolherbst: 4.20
20:25kreyren: ah oke O.o
20:26kreyren: > ifdef LINUXDIR
20:26kreyren: eh
20:26kreyren: > ifndef LINUXDIR
20:26kreyren: so i guess just `LINUXDIR="$HOME/Repositories/linux" make` ?
20:27karolherbst: yeah.. something like that
20:27karolherbst: or make LINUXDIR=...
20:27kreyren: wants me to make kernel config
20:28karolherbst: ye[
20:28karolherbst: need to compile a full kernel for it
20:28karolherbst: or install 4.20 and use its build dir
20:28kreyren:did localmodconfig
20:29kreyren:is compiling
20:37kreyren: built, building the drm thing
20:41kreyren: < https://matrix.org/_matrix/media/r0/download/matrix.org/ZFteZYDrpaAqkdpdJUnKQzSA/message.txt >
20:41kreyren: karolherbst:
20:41karolherbst: mhhh
20:42karolherbst: mhhh
20:42karolherbst: this got renamed in 4.19.. strange
20:43karolherbst: sure you are on the right branch of the nouveau tree?
20:43kreyren: kreyren@dreamon:~/Repositories/nouveau/drm$ git branch
20:43kreyren: * master_4.17
20:43kreyren: a
20:43kreyren: changing x.x
20:44kreyren:would swear that he changed it
20:44kreyren:is on clk_cleanup_for_real
20:48kreyren: built successfully
20:48kreyren: so now just make install?
20:48karolherbst: now you just need to install the new kernel and nouveau
20:48kreyren: karolherbst:
20:49kreyren: ah the changes are already in the linux dir?
20:49karolherbst: no, but you need to install the other kernel so you can load the custom nouveau module
20:49kreyren: -> just make install in linux dir?
20:49karolherbst: installing the kernel is a bit more annoying
20:49karolherbst: make modules_install; make install
20:50karolherbst: and then you can do make install in the nouveau tree at drm/
20:50karolherbst: then you need to regenerate the initramfs for the 4.20 kernel
20:50karolherbst: well.. not a good idea to run an EOL kernel tree anyway, so the patches should be ported to the newest kernel anyway :/
20:51kreyren: x.x
20:51karolherbst: and then you even have to boot with nouveau.config=NvFanless=1 to force enable reclocking
20:52kreyren: > cp: cannot stat 'kernel//home/kreyren/Repositories/nouveau/drm/nouveau/nouveau.ko': No such file or directory
20:52kreyren: when installing nouveau from drm
20:52kreyren: ah LINUXDIR
20:53kreyren: installed
20:53kreyren: initramfs generated too
20:54kreyren: trying to boot it ^-^
20:55kreyren: Kernel panic x.x
20:56karolherbst: upsi
20:56kreyren: Kernel stack is corrupted in start_secondary+0x1c7/0x1d0
20:56karolherbst: are you on 4.20.0 or the latest from the branch?
20:57kreyren: 4.20 tag in linux dir
20:57karolherbst: ahh, you might want to pull the stable remote and use the latest 4.20 tag from there
20:58karolherbst: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
20:58karolherbst: seems like 4.20.17 was the last release there
20:58kreyren: O.o
20:59karolherbst: get remote add stable https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
20:59karolherbst: get fetch stable
20:59kreyren: checking
20:59karolherbst: uhm
20:59karolherbst: s/get/git/
20:59ccr: :)
20:59kreyren: did you meant `git` ?
20:59kreyren: ah
21:00kreyren:got confused and started to write `go get remote add..`
21:01ccr: shell the text adventure game
21:03kreyren: lol
21:11kreyren: finished git fetch stable
21:11kreyren: so just `make` now?
21:11karolherbst: well, have to switch the branch and stuff
21:12karolherbst: also need to install and rebuild nouveau and and end :/
21:13kreyren: oke
21:31kreyren: Same kernel panic x.x
21:31karolherbst: sad
21:31karolherbst: was worth a try
21:32kreyren:uploaded an image: (80KiB) < https://matrix.org/_matrix/media/r0/download/matrix.org/EMAlibxToQtudSgIAhrVJHtM/IMG_20210314_223136.jpg >
21:32karolherbst: could also be that the kernel stack protector stuff wasn't ready back then :/
21:32kreyren: <karolherbst "was worth a try"> Yep
21:33kreyren: <karolherbst "could also be that the kernel st"> Seems like it x.x
21:33kreyren: Will you be finishing it then? Would be happy to use it
21:33karolherbst: yeah... depends on if I find time or not
21:33kreyren: Willing to pay too
21:34karolherbst: some of the patches should land anyway, so a good idea anyway
21:34kreyren: O.o
21:34karolherbst: yeah well.. at least I can't accept any money :D or could be too risky, because it is also my job more or less
21:34kreyren: O.o
21:35kreyren: I can draw you an encouraging picture then!
21:35karolherbst: would have to do it as a spare time only thing and it is hard to split it from my work related work
21:35kreyren: X.x
21:35karolherbst: yeah well.. money isn't the problem anyway :p
21:35karolherbst: I'd rather see more nouveau contributors, especially new ones
21:35karolherbst: but sadly it's not so easy to find new people
21:36kreyren: Would be happy to contrib if I knew what to do
21:36karolherbst: yeah... we are kind of planning to move to gitlab for kernel patches, so we can better track MRs and stuff
21:36kreyren:is currently working on a bigger project for gitea though
21:36karolherbst: yeah, no worries
21:37kreyren: Why not gitea instead ? Gitlab is painful mostly
21:37karolherbst: we already have a gitlab instance for all of fdo
21:37kreyren: I see x.x
21:37karolherbst: yeah.. already tons of trees there
21:37karolherbst: it's quite massive
21:38kreyren: Or like if you have a bug that explains the issue so that I know what needs to be done then I could probably contribute easily
21:38juri_: karolherbst: good luck on finding more hackers.
21:39karolherbst: yeah well... :/
21:39karolherbst: kreyren: well, thing is, nothing of that is really that simple sadly
21:39karolherbst: so even if we have a bug, figuring out what needs to be done is like 99% of the work
21:39kreyren: What can we do to make it simpler then
21:40karolherbst: full gpu docs would help :p
21:40kreyren: In-code docs seems like solution there
21:40karolherbst: it will help, but is also not the biggest problem
21:40kreyren: <karolherbst "full gpu docs would help :p"> Ah right x.x
21:40karolherbst: the problem is, we have close to noe GPU docs
21:40juri_: karolherbst: have you had anyone do a fuzz map of the code that is working? :)
21:40karolherbst: so.. we have to reverse engineer most of the bits from the blob driver
21:41kreyren: <karolherbst "the problem is, we have close to"> Was there any effort from FSF about that ?
21:41karolherbst: juri_: well.. fuzzing the driver is a fun project a tleast Lo
21:41karolherbst: :p
21:41karolherbst: kreyren: well.. why would nvidia listen to the FSF?
21:41kreyren: <karolherbst "Krey: well.. why would nvidia li"> FSF is founded to be concerned about these issues
21:41karolherbst: hehe.. yeah well.. no
21:41juri_: I was recently looking at a HDMI -> CSI capture device for the pi.
21:42karolherbst: none of the GPU docs are in any kind related to whatever the FSF did
21:42juri_: the FSF is much smaller than people think.
21:42kreyren: <karolherbst "hehe.. yeah well.. no"> O.o
21:42kreyren: Well FSFE is... FSF is kinda a joke now x.x
21:42karolherbst: FSF have nothing to do with AMD publishing docs
21:42karolherbst: nor intel
21:42karolherbst: nor anybody else
21:42kreyren: O.o
21:42kreyren: I though the the Intel was FSFE in part
21:43karolherbst: why would you think that?
21:43karolherbst: that we have that many open source SoC drivers is at least googles fault :p
21:43juri_: karolherbst: fuzzing hardware is fun, any way you look at it. I wish i had time to do that stuff.. but between dayjob, and my horribly large/complicated projects, things are busybusybusy.
21:43karolherbst: that's more or less a given
21:44karolherbst: juri_: yeah..
21:44karolherbst: fuzing from the GPU perspective, but also fuzzing ioctls would both be very helpful
21:44kreyren: <karolherbst "why would you think that?"> I remember them promoting something related to Intel in 2018 o.O
21:44karolherbst: D:
21:44karolherbst: yeah well..
21:44karolherbst: we have open source intel drivers long before that
21:45karolherbst: and I think heavy investment started around 2015 ?
21:45karolherbst: maybe before?
21:45karolherbst: dunno
21:45karolherbst: ask intel folks :D
21:46kreyren: O.o
21:47karolherbst: I mean.. we also have nvidia GPU docs, but the FSF also has nothing to do with it :D
21:49kreyren: O.o
21:49kreyren: <karolherbst "I mean.. we also have nvidia GPU"> That was redhat right
21:50ccr: "what have the Romans ever done for us anyway?"
21:51karolherbst: yep
22:02imirkin: build roads
22:02imirkin: clean water
22:02imirkin: schools
22:03imirkin: (i have to assume that was a life of brian reference...)
22:03karolherbst: it probably was
22:05ccr: :)
22:05ccr: the meaning of that pun-reference was that FSF has probably done something, just not those things
22:06karolherbst: imirkin: do you remember why intel started to invest into mesa?
22:06karolherbst: or well.. open source linux graphics
22:06imirkin: it was before my time
22:06karolherbst: that long ago already?
22:06imirkin: yes
22:07karolherbst: I mean it feels like there was a bigger push like 5 years ago, but maybe it was always like this?
22:07karolherbst: or well.. 6
22:07ccr: they still have this division between the more open development of Mesa/kernel and then there's the .. media .. and stuff like Beignet and so on
22:07imirkin: karolherbst: afaik they contributed the initial glsl compiler
22:07karolherbst: ahh
22:08imirkin: so ... a while ago.
22:08karolherbst: insane
22:08imirkin: which is also why there's a mesa ir
22:08imirkin: and a glsl ir
22:08imirkin: which are different
22:08imirkin: because intel can't reuse other people's things
22:08karolherbst: sure
22:08karolherbst: :D
22:08imirkin: that'd be death
22:08karolherbst: oh well
22:08imirkin: eventually they did grudgingly make a gallium driver
22:08imirkin: but they fought the good fight with i965 for a long time
22:09karolherbst: well
22:09karolherbst: somebody just had to do it and show how much faster it would be
22:09karolherbst: but yeah, that was stupid
22:09ccr: I thought there were/are some technical reasons for hanging to i965, something about vec4 and so on
22:09karolherbst: nope :D
22:09ccr: oh
22:09karolherbst: they share the same compiler
22:09karolherbst: so...
22:10imirkin: ccr: iris can't support the earlier gens
22:10imirkin: it's designed for softpin, which is not supported on gen7 and earlier
22:10ccr: imirkin, that's what I meant basically
22:10karolherbst: yeah, but that was by choice
22:10imirkin: has nothing to do with gallium per se
22:10imirkin: just the design of the driver
22:10imirkin: there's an effort to make another driver for the earlier chips on gallium
22:10imirkin: but it's slow-going
22:11imirkin: (and not intel-backed)
22:11karolherbst: yeah.. companies tend to not care about old devices :D
22:11imirkin: weird, right
22:11ccr: how odd, indeed
22:12ccr: then again I've been slightly amazed by people who come here, running on some TNT2-level stuff.
22:13imirkin: eh, why not
22:14ccr: well, I'm not against it of course .. and from hobbyist perspective it can be fun
22:14ccr: not that my opinion means anything :)
22:16ccr: sometimes I just get the feeling that some of those people feel a bit too entitled to have their ancient stuff supported.
22:17ccr: considering the resources available etc
22:20airlied: though sometimes fixing retro stuff is less effort than trying to work out modern hw :-p
22:21imirkin: airlied: i tried some crocus stuff, with limited success
22:21imirkin: gen6 is pretty unhappy
22:22ccr: airlied, undoubtably :)*
22:23imirkin: airlied: could use some hints :)
22:28airlied: imirkin: moved to other channel!
23:06karolherbst: imirkin: you might like this one: https://lichess.org/training/wshWl :D
23:07imirkin: yeah, not a bad one
23:07imirkin: relatively common pattern
23:07imirkin: took me a few seconds to work out tho
23:07karolherbst: yeah...
23:08karolherbst: I am out of the game for too long :/
23:08imirkin: more like i've been obsessing over it
23:08imirkin: there's a "puzzle storm" thing now too
23:08imirkin: my top score is in the 30's
23:08karolherbst: yeah, I see
23:08imirkin: oh wow, 39. but i think something weird happened, and i semi-cheated
23:08karolherbst: still trying to get back my old strength through puzzeling :D
23:09imirkin: i.e. they gave me the SAME sequence of puzzles a second time
23:09karolherbst: ahh
23:09karolherbst: weird
23:09imirkin: but i get up to 30 regularly
23:09imirkin: 35 is a good run.
23:09karolherbst: I am still too slow for it :/
23:09karolherbst: let's see
23:09imirkin: so ... there are some tricks
23:09imirkin: (a) make the board small
23:09imirkin: (b) use a mouse :)
23:13karolherbst: ahh yeah.. firest try at 9 :)
23:13karolherbst: *first
23:13karolherbst: I am just thinking too long
23:13imirkin: the first ~10 are easy solutions
23:13imirkin: so once you see something that wins, you can just go for it
23:14karolherbst: yeah..
23:14imirkin: there's not going to be funny business
23:14imirkin: once you get into the 20's ... funny business is to be expected :)
23:14karolherbst: :D
23:14imirkin: i just spent like 30s on puzzle #30 and still didn't see it
23:14karolherbst: I will move to the harder puzzles shortly
23:14karolherbst: still stable around 1800 in puzzles
23:14imirkin: https://lichess.org/training/OtlTe
23:14karolherbst: some cetageroies better, some worse
23:15karolherbst: let's see
23:15imirkin: ooooo
23:15imirkin: i see it now
23:15karolherbst: you get the queen, no?
23:15imirkin: i was focused on the wrong thing
23:15imirkin: no.
23:15imirkin: you're focused on the queen and rook, like i was
23:15imirkin: and there's no winning thing there
23:15karolherbst: ahh
23:16imirkin: (sack the bishop)
23:19karolherbst: yep
23:19karolherbst: takes a while
23:20imirkin: i've been maintaining a ~1500 bullet rating. gotta see that stuff moderately fast
23:20karolherbst: :D
23:20imirkin: obv still very weak, but ... improving requires actual studying
23:21imirkin: or at least playing longer games
23:21karolherbst: never played fast paced chess
23:21imirkin: it's basically not chess
23:21karolherbst: I know
23:21imirkin: you just move pieces around randomly.
23:21imirkin: and wait for the other guy's time to run out ;)
23:21karolherbst: what I played at tournament was usually 30 mins
23:21imirkin: yeah, that's standard rapid, esp at the lower levels
23:22karolherbst: yeah well.. once I get to my past level :D
23:22karolherbst: I am still doing dumb mistakes
23:29imirkin: and you think i'm not? :)
23:30karolherbst: :D
23:33karolherbst: the painful part is, that I know I was better :D
23:37imirkin: well, if you want, we can play some games unrated (or rated, whatever)
23:38karolherbst: yeah, at some point. This week will be a bit annoying. More furnature arriving :) And I still feel like I need to prepare better :D But once I feel ready I can ping you
23:40imirkin: sure