06:54tagr: kherbst: yeah, I think chronologically they relate like this: gp100 -> gp10b -> gp102
07:53kherbst:should talk to nvidia to get the more expensive jetsons as well :D
08:04tagr: kherbst: which ones do you have?
08:05tagr: heh... trying a GP106 on Jetson AGX Xavier I get this: [ 23.894838] nouveau 0005:01:00.0: gpio: GPU is missing power, check its power cables. Boot with nouveau.config=NvPowerChecks=0 to disable.
08:06danvet: imirkin, did you figure out anything about why nouveau goes boom on your nv30something?
08:06tagr: are there external supplies for such GPUs?
08:08pmoreau: Could be that the detection code for power cables need some tweaks for such a platform.
08:08pmoreau: mmenzyns might have an idea, since they worked on that path IIRC.
08:13kherbst: tagr: only the nano, but that one I bought.
08:13kherbst: tagr: heh? fun
08:14kherbst: tagr: it's a check if those power PINs are connected
08:14kherbst: and we thought.. maybe we want to check that
08:14kherbst: tagr: but weird that you are hitting this as you shouldn't have such a kind of GPIO advertised
08:15kherbst: tagr: that's the commit adding it: https://github.com/skeggsb/nouveau/commit/a18d02063929160bf15a00c150f6d94fe7bd4338
08:16kherbst: skeggsb: uff.. seeing that commit: in the kernel we don't split up strings :)
08:16kherbst: even if they go beyond 80 lines
08:16kherbst: having the string in one line is always prefered over spliting it up to fit 80 chars
08:18kherbst: skeggsb: https://www.kernel.org/doc/html/v5.2/process/coding-style.html#breaking-long-lines-and-strings
08:18kherbst: "However, never break user-visible strings such as printk messages, because that breaks the ability to grep for them."
08:21kherbst: tagr: ohhhh wait, you try a nvidia PCIe on top of the AGX xavier?
08:21kherbst: does that gpu indeed has those power PINs?
08:21kherbst: if not... we are in trouble
08:22kherbst: sounds like a GTX 1060 and those go above 75W, so I'd imagine it has one :)
09:06tagr: kherbst: yeah, this is an ASUS GTX 1060 dual something
09:06tagr: ah... wait... DUAL-GTX1060-3G
09:07tagr: and yes, that's connected to the AGX Xavier and apparently requires that 6-pin to be connected
09:07tagr: so I'm thinking of maybe just getting a cheap PSU and one of those dual-4-pin-molex-to-6-pin adapters
09:08tagr: the PSU would also help with some other use-cases I need to occasionally test (like external SSDs for SATA)
09:08tagr: there's power supplies for just a 4-pin molex, but they are almost as expensive as a cheap PSU and the PSU has tons of connectors already
09:20karolherbst: tagr: I am more worried about the power supply of the xavier not being able to provide enough power.. so yeah
09:20karolherbst: I bet it's 100W max
09:20karolherbst: or well.. 20A@5V
09:21mmenzyns: nvidia claims maximum power of gtx 1060 cards is 120 W
09:21karolherbst: ohh, it uses 19V, like laptops
09:22karolherbst: also.. no bumblebee required anymore \o/
09:26HdkR: What solved bumblebee?
09:27karolherbst: HdkR: hybrig graphics with nvidia binary driver
09:27karolherbst: HdkR: ohh
09:27karolherbst: the other way around
09:27karolherbst: HdkR: 435
09:27karolherbst: nvidia now supports rendering offloading
09:27karolherbst: and it works quite nicely actually
09:27karolherbst: much faster than bumblebee as well
09:28HdkR: Interesting. Wasn't one of the issues before due to kernel symbols being tagged GPL?
09:28karolherbst: HdkR: apparently?
09:28karolherbst: HdkR: I think the problem was runpm rather
09:28karolherbst: but they solved it
09:28karolherbst: userspace has to enable it :)
09:29karolherbst: so nvidia never calls the GPL runpm smbols, but only provides the callbacks for runtime_suspend/resume
09:29HdkR: hah, cheeky
09:29karolherbst: oh well..
09:29karolherbst: runpm is turing+ only
09:29karolherbst: for... stupid reasons
09:30HdkR: That's annoying
09:30karolherbst: especially because it works with turing and nouveau
09:30karolherbst: so I can't reverse engineer what we do wrong...
09:30HdkR: So many features being limited to turing from Nvidia recently. Feels like a drive to get people to purchase the new hardware
09:31karolherbst: well... yes
09:31karolherbst: I mean, for most stuff it really makes sense, but for runpm it's just crappy reasoning
09:32karolherbst: imirkin: ohh, do you have old cards which require additional power supply? like tesla or something before that?
09:33karolherbst: did I ask that already?
09:33HdkR: It has been nearly a year. Still no Turing support in Nouveau? :)
09:34karolherbst: well.. you know.. secboot
09:35HdkR: Sad times
09:36karolherbst: there is only one proper solution to secboot: 🔥
09:37HdkR: Launch it in to the sun?
09:48karolherbst: mhh.. wouldn't really change anything, right?
09:50HdkR: Least you don't have to worry about the problem anymore, it's in the sun :P
10:01pmoreau: karolherbst: I should have some Tesla cards with additional power ports.
12:22imirkin: karolherbst: NV42 (Quadro FX 3450)
12:23imirkin: danvet: tagr sent me a patch. i haven't had time to test yet.
12:23imirkin: and nouveua just goes boom, no nv30 required
12:23imirkin: the nv30 was to test your changes
12:26danvet: imirkin, oh so it's all of nouveau that goes boom?
12:47karolherbst: imirkin: can you check if the vbios has a GPIO for checking for attached power cables? should be 0x10, might be 0x4c
12:48imirkin: karolherbst: https://people.freedesktop.org/~imirkin/traces/nv42-vbios.rom
12:51karolherbst: ah, so it has it
12:52karolherbst: "GPIO 6: line 8 tag 0x10 [ATX_POWER_BAD] IN NEG"
12:52karolherbst: imirkin: would be cool if you could test if the changes we added to master to detect a missing power cable actually does work on that GPU as well
12:52karolherbst: it has to be the secondary GPU for it though
13:18imirkin_: not 100% sure the system boots without it in
13:18imirkin_: that gpu is finicky - system won't boot with that gpu + any other in the other pcie slot
13:21karolherbst: imirkin_: mhh.. the bios usually only cares about the main GPU... but yeah. I think it depends on the system actually
13:21imirkin_: like ... doesn't POST
13:22karolherbst: there are some boards being less annoying with PCIe though
13:22karolherbst: like the one I have really doesn't care
15:58phillipsjk: SO I may have found a modern use for nouveau: using old nivida adpater as display adapters for modern cards without VGA outputs. Sort of like optimus in reverse. https://en.wikipedia.org/wiki/Nvidia_Optimus
15:59imirkin_: aka "reverse prime" -- using another GPU's outputs
15:59imirkin_: although as i recall, you have some older boards, so that is unlikely to work well there
15:59phillipsjk: I currently have it working on a Ubuntu Live DVD, but have no idea how much accelaration each driver is doing.
16:00imirkin_: should be fine with nv50+
16:02phillipsjk: Even did the math: 1280x1024x4bytes@60Hz -> 315MB/s. My new computer I am putitng together has PCI-E 2.0 slots, so even 1 lane at 500MB/s should be enough.
16:03phillipsjk:did the math becuase the first attempt was very laggy. Second boot, with a TV plugged into the modern card was very fast on the nv card.
16:04phillipsjk: Is prime what AMD calls it?
16:04phillipsjk: or is that a pun?
16:04imirkin_: it's a pun
16:04imirkin_: and it's the term used in drm
16:05phillipsjk: No specific questions at the moment I guess.
16:08phillipsjk: My nv card is a GF119 (GeForce GT 610) -- was surprised by the 2GB of VRAM
16:09imirkin_: welcome to the future.
16:10phillipsjk: My amd card is a VEGA56: it barely fits in the case/ motherboard
16:10imirkin_: when your new video card has more ram than all your previous computers put together
16:10imirkin_: btw, DP -> VGA active adapters are a thing
16:11phillipsjk: Yes, I have upgraded from when I was trying to get a mach64 working.
16:11phillipsjk: If game lag is a thing, I may have to look into that.
16:13phillipsjk: The computer I am typing this on has 20GB of RAM (old server, converted to work-station use) -- power hungry and noisy though.
16:13imirkin_: and your video card probably has 32GB :p
16:14phillipsjk: no, 8GB, same as the system RAM I have in the box
16:16phillipsjk: I am in the process of conveting most of my computers over to use ECC RAM. (Paraniod memory errors will cuase data-loss in ZFS (of bttrfs))
16:20phillipsjk: I noticed you are blcoking riff-raff from the channel.
16:54karolherbst: phillipsjk: that non ECC RAM is causing issues is a rumour though
16:54karolherbst: it's a bug inside the FS as far as everybody else is concerned
16:55karolherbst: there are fs which are good in handling errors cases (ext4, XFS) and others which totally aren't (btrfs). No idea how fault tolerant ZFS is
16:56phillipsjk: ZFS has incredible file integrity guarantees (SHA-256 has, IIRC): but they only work if your RAM is reliable.
16:58phillipsjk: I have not used btrfs yet, but on paper it should be better (or just as good) as the others.
16:58karolherbst: it's not
16:59karolherbst: I never had a broken ext4 fs
16:59karolherbst: I already had 5 broken btrfs
16:59karolherbst: also, there are some studies about fault tolerant of various fs
16:59karolherbst: and btrfs is even worse then vfat or ntfs
16:59karolherbst: might have changed over the last few years though
17:00karolherbst: but it was never good
17:00phillipsjk: I assume this is with no mirror?
17:00karolherbst: this is when you system crashes and you reboot
17:00karolherbst: normal operation is boring
17:00karolherbst: and has nothing to do with being fault tolerant
17:01karolherbst: if crap gets written to the disc, the driver has to handle it one way or the other... in most cases btrfs crashed the kernel just another time
17:01karolherbst: at least for me
17:02karolherbst: it's all fun if your kernel is in a solid state all the time and you never get memory errors
17:02karolherbst: but what if bits on the disc itself are in a broken state?
17:02karolherbst: then stuff becomes interesting
17:05phillipsjk:has had systemd try to break his ext2fs -- they seem to assume you upgraded to ext3 with journalling, so fs checking should take no more than 30s.
17:07phillipsjk: The machine inquestion had an installation dating back to 2004 or so.
17:07karolherbst: but these days the ext4 driver is used anyway
17:07karolherbst: so it shouldn't matter
17:07karolherbst: although I guess it would make sense to upgrade the fs at some point
17:08phillipsjk: I think you can even do it in place.
17:10phillipsjk: I retired the machine is a recent move due to a noisy disk.
17:10phillipsjk: (almost like they weren't designed to run 20 years)
17:11karolherbst: well.. that's always the issue with moving parts
18:39phillipsjk: karolherbst, sounds like btrfs may be in need of some fuzz testing if bad data crashes the kernel.
18:49phillipsjk: Maybe if I get btrfs crashes with ECC RAM, I can prove it was *not* a memory error.
22:39imirkin_: Lyude: for your 3/4 "use less encoders", i don't know if you're trying to make my head explode or if it's just a happy coincidence...
22:39imirkin_: but please change that to "fewer" :)
22:39Lyude: imirkin_: it's coincidence I promise :P
22:40Lyude: There are some grammar rules my brain just does not like to remember for some reason. Happens with apostrophes sometimes too
22:40imirkin_: fewer vs less is a very common point of confusion for native speakers, for some reason
22:41imirkin_: however a native speaker would under no circumstances mess up "much" and "many"
22:42imirkin_: (whereas for non-native speakers whose native language does not differentiate between countable and uncountable things, it feels like completely randomly selected)
22:44imirkin_: does the possible_crtcs thing have to get fixed in -nouveau ?
22:44Lyude: ...also, does anyone know if I respond to that patch with a v3 if patchwork will actually properly pick it up along with https://patchwork.freedesktop.org/patch/330699/?series=66691&rev=2
22:45Lyude: imirkin_: I didn't actually check, but I wouldn't be surprised if it did
22:45Lyude: it is likely one of those bits of copy pasted code
22:45imirkin_: i know i copy/paste early and often
22:45imirkin_: and sometimes end up with copy-pasta
22:47Lyude: imirkin_: happen to know the answer to my patchwork question btw? ^ wondering if I can avoid resending the series to fix the s/less/fewer/ thing in https://patchwork.freedesktop.org/patch/330675/?series=66691&rev=2
22:50imirkin_: i dunno
22:50imirkin_: either way, you can wait for actual feedback
22:50imirkin_: before doing lots of resends
22:51Lyude: mm, true