05:56 anholt: RSpliet: I don't think an intel atom board would be much use for testing.
06:03 anholt: imirkin: what kind of power does that imac mini take?
06:05 anholt: so close. https://gitlab.freedesktop.org/anholt/mesa/-/jobs/17967715#L1783
06:05 anholt: nouveau drm has probed. wonder why we're not getting novueau
06:15 anholt: nvc0_screen_create:1084 - Base screen init failed: -19
06:15 anholt: oh, this is firmware isn't it.
06:16 imirkin: anholt: most likely ... you can check dmesg to see if it didn't init properly
06:16 anholt: yep
06:16 imirkin: anholt: not sure precisely what kind of power, iirc there's a power brick + DC power of some sort
06:17 anholt: imac seems really likely to come in under 15 watts
06:17 anholt: *unlikely
06:17 imirkin: you know which thing i'm talking about right?
06:17 anholt: no clue
06:17 imirkin: it's the little mini-pizza box thing
06:17 imirkin: https://www.apple.com/mac-mini/
06:17 imirkin: like that. but 10 years older.
06:17 anholt: I interacted with mac mini in the first iteration of a thing with that name.
06:18 imirkin: anholt: https://www.ebay.com/itm/124967454184?hash=item1d18a405e8:g:J-AAAOSwiyRhcwCi
06:19 anholt: https://www.amazon.com/Compatible-Apple-Supply-Adapter-Charger/dp/B07LB52QHS slightly over 15w.
06:20 imirkin: wow... https://support.apple.com/en-us/HT201897
06:20 imirkin: all detailed
06:20 imirkin: 13-15W - 110W
06:20 imirkin: so _slightly_ outside your power budget
06:21 anholt: the sharp drop on transition to m1 there :)
06:21 imirkin: anholt: as for the firmware issue, you just need stuff in /lib/firmware/nvidia/nv12b (or gm20b or whereever it's shipped)
06:21 imirkin: anholt: if you built nouveau in, then the firwmare has to be added to the kernel too
06:22 imirkin: i.e. it just needs to be available whenever nouveau loads
06:22 anholt: yeah, I'll probably need to do a build as a module, so we can load late enough
06:22 imirkin: you can also do CONFIG_EXTRA_FIRMWARE
06:22 anholt: I really like how msm can load the gpu firmware late
06:23 imirkin: (not highly recommended, but just fyi about it existing)
06:23 imirkin: patches welcome :)
06:23 anholt: Yeah, I don't think we should dump 524k of tegra firmware into everyone's tftp payload.
06:24 imirkin: more of an fyi to get you going
06:24 anholt: yep. there are definitely a few hacks here
06:25 anholt: (like a hand-built kernel still because there's no way I'm debugging this all through container rebuilds)
06:25 imirkin: hehe
09:40 RSpliet: anholt: Sure. That atom board is essentially a Mac Mini with a shittier processor but same GPU. The atom 330 only has an 8W TDP (25W for the Mac Mini Intel P7350). I think the total system still came in around 20W.
09:41 RSpliet: IIRC that gen of Atom was really bad at DVFS
09:46 RSpliet: Oh no, it's 20W idle, over 40W under load O.o
16:42 karolherbst: anholt: so.. for CI the main thing is I don't really have much time. I do have hardware though which I could host at home (but super unreliable, because.. ISPs are crappy and everything)
16:42 karolherbst: I planned to follow mupuf guide and do at least post merge testing or sometihng
16:42 karolherbst: but not sure when
16:45 karolherbst: RSpliet: running propber CI on that hw still takes ages though
16:45 karolherbst: *proper
16:53 karolherbst: imirkin: btw.. I don't think that the vbios patch is correct
16:53 imirkin: pretty sure it is.
16:53 karolherbst: soo.. let's assume you call nvbios_read with bios, 0xff and 0x1
16:54 karolherbst: then it checks 0xff + 1 for OOB
16:54 anholt: karolherbst: if you wanted to send a batch of suitable hw my way, I should be able to get it up as pre-merge testing within a week or so. or I can go order some.
16:54 karolherbst: which 0xff clearly isn't if the size is 0x100
16:54 karolherbst: there is an implicit -1 hidden in the >=
16:54 karolherbst: size is just the "size" of the access in byes
16:54 karolherbst: *bytes
16:55 karolherbst: anholt: I can ask nvidia to send you tons of it :D
16:55 karolherbst: mostly tegras only though
16:55 karolherbst: in case you think this would be helpful I can ask them
16:55 imirkin: karolherbst: check my email
16:55 anholt: karolherbst: tegra is what I can take in this setup, basically. "<15w and I can find a poe splitter to plug it into"
16:55 imirkin: karolherbst: where i gave the proper example
16:55 anholt: karolherbst: you could ask, or I would love an introduction if you have a contact.
16:55 karolherbst: anholt: yeah.. that was my plan as well. I have a jetson nano at home
16:56 karolherbst: anholt: yeah, I will have a meeting next week about jetson with nvidia, but I will write an email before that
16:56 anholt: thanks!
16:56 karolherbst: the reason we discuss mesa CI anyway is, because it caused fedora blocker bugs twice in a row :D
16:57 karolherbst: (with nvidia that is)
16:57 karolherbst: but they don't want to host it themselves...
16:57 karolherbst: anyway, so much for background
16:57 anholt: context helps!
16:58 karolherbst: anholt: is 15W a hard limit?
16:59 anholt: I think there's some higher end POE that can go higher, but 15w is easy at least
16:59 anholt: (and it's what this switch can do)
17:00 karolherbst: mhhh... mhhh
17:00 karolherbst: not enough I guess
17:00 karolherbst: the nano itself already needs 15W and that's like the lowest end.. let me check
17:01 karolherbst: yeah... 4A@5V max
17:01 anholt: tk1's 15w tdp it looks like
17:01 karolherbst: well.. normally it's good enough
17:01 karolherbst: 3A is also good enough for the nano
17:01 karolherbst: but it can draw more
17:01 karolherbst: and for the tk1 it mostly depends on having a PCIe card or not
17:02 karolherbst: soo.. if you are fine connecting devices which can potentially draw more than 15W then that's fine
17:02 anholt: right, tdps are for no addons, but for this setup we're talking about no peripherals.
17:02 karolherbst: I know that 10W isn't enough for the nano
17:02 imirkin: karolherbst: imagine you have a bios of size 4, and you're trying to access a 4-byte thing at offset 0
17:02 karolherbst: and under high load it crashed due to not enough power
17:02 imirkin: karolherbst: 0 + 4 == 4. so the "invalid" check has to be 0 + 4 > 4
17:03 karolherbst: imirkin: you read byte 0. 1. 2 and 3
17:03 imirkin: exactly.
17:03 karolherbst: mhh
17:03 karolherbst: but still
17:03 imirkin: the current check is >=
17:03 imirkin: it needs to be >
17:03 imirkin: the current check disallows reading the last byte
17:04 karolherbst: ohhh, right.. I thought the wrong way around
17:04 karolherbst: *was thinking
17:05 karolherbst: I thought the patch is limiting it by another byte :D
17:06 imirkin: yeah, that's what's confusing -- it's the condition for the _error_
17:06 karolherbst: yeah
17:06 karolherbst: I replied to the patch
17:11 karolherbst: anholt: anyway, what are you using to control PoE? custom scripts tight to the switch or something else?
17:11 anholt: https://gitlab.freedesktop.org/anholt/mesa/-/commit/7dad8c5dc5034863b7d476e3545954852383794d
17:12 anholt: (WIP docs, obvs. questions appreciated so I can fill them out some more)
17:12 karolherbst: ahhh
17:12 karolherbst: I do have a netgear switch here :D not even sure it has an API besides doing shitty HTTP calls
17:12 anholt: if it's "managed" then it probably has snmp
17:12 karolherbst: probably there is something, dunno
17:12 karolherbst: anholt: yeah it is
17:13 karolherbst: I think...
17:13 anholt: (note: making assertions about networking here not actually backed up with any relevant experience)
17:13 karolherbst: might be "stupid" managed
17:13 karolherbst: netgear GS310TP here
17:13 karolherbst: it does have 30W per port though
17:13 imirkin: if there's a web ui, there's snmp
17:13 karolherbst: imirkin: you'd think that...
17:13 imirkin: at least on every switch i've seen
17:14 anholt: https://community.netgear.com/t5/Smart-Plus-and-Smart-Pro-Managed/GS310TP-snmp-poe-disable/td-p/1796094 nice
17:14 karolherbst: these days you also have non managed ones wiht web UI
17:15 anholt: my networking friend was like "you don't really want to use snmp, for configuring network you really want to use some combination of this collection of python libraries"
17:15 karolherbst: anholt: that's why I have a machine standing here which would do that
17:15 karolherbst: with two network cards
17:15 imirkin: nobody wants to use snmp :)
17:15 karolherbst: worst case... I have some docker containers managing the second network :D
17:17 karolherbst: but this switch is strange
17:17 karolherbst: it can only do 55W in total for Poe...
17:17 anholt: yeah, finding lots of total poe was one of the hard parts. thus the big previous-gen cisco.
17:18 karolherbst: yeah.. I wanted something cheap
17:18 karolherbst: paid like 100€ for that one
17:19 anholt: https://www.ebay.com/itm/253355655581?epid=82152190&_trkparms=ispr%3D1&hash=item3afd2c819d:g:FfcAAOSw6pRhJpJ4&amdata=enc%3AAQAGAAACkPYe5NmHp%252B2JMhMi7yxGiTJkPrKr5t53CooMSQt2orsSg3Ye8yTWgOW7pmE1t838dsSLc6jbI7Z2qPP3hT7zruiwlZ87Y9NRcYH3uFjHC8W%252B09secFnjPMCHO8beW7ZO91sw%252FcKOMLUQZfsniIQJ%252BB%252FKRL9r7TUhOV0g%252FheZP6b86sHvBmwuq42v%252FzwOnuTA4gMElwM4j68LF%252BVgFulv63iVb0EcNrEM14yaHOsSOvw76J54xfKQgIa5yeKNlQ31k7jPWl%252B6eyBUpslSfT9NwED
17:19 anholt: 7Lx5fwhW8wmwJSRiUdiPoZCgZoEybmOCSNipVJwIQJWCDKlAutfl4XjXTgdsCj%252B5sD1XYEXJr8uZ5DLPjlNMr%252BM8ybrvgTNa1RPoe6lrN%252FJxydSiKZMASepH7ahZPKYC%252BzIfaXqdn7fJJ9InxNcr5J5vS5eLere2CMDchN2A5vZBJfB64f89ypAASsNztDUHl6f1eakRoNMgfYui1df8bBxj5L5CtnqIrSnG9hqzC2faz%252Fazb%252FZT09Nnc0VAtuzynbb5y4MMKSJYEugAY37%252BrRv2RqJeUt4ojcvISOxImyPpoaCby70dqRs%252FtlqGfi2GtmiqZWyZzI%252FuXmmoZtQgMMtf2iBWbyvd4M0qUVGPORfHsUGGsEItLy5sz8M0CSTF7uW4g%252FXDUU0X%252FE9lhJ
17:19 anholt: XW7BXYgHucaAsNyxuCgGLsAyy9aUKo3Tzi6c0F8jW2Dq%252Fi5cvzzyOjW9meI%252FMYLf47rE55XEuaReOOnuCPhQ8cIfUwiqiMHB%252FekVTBlu%252Bp%252BebF5ntfzc9Lo2tdrSpUSqfMMxwrC%252FTkNNJT%252BWkrZ8bQsb4epgF2hU%252FmvAmz%252BKYLaDR6dtW5mgW00fPPBiZxljMXi%7Cclp%3A2334524%7Ctkp%3ABFBM3OXF0dFf <-- basically what I'm on, one with a 750w power supply
17:19 karolherbst: anyway.. having a nano and the tk1 one and my plan was to just install whatever mupuf tells us to and "it should work"
17:19 anholt: eep! sorry for spam.
17:19 karolherbst: anholt: 750W, but only 15W per port? sound slike a 64 port beats
17:19 karolherbst: *beast
17:19 anholt: 48. I think there's some overhead allocated too.
17:20 karolherbst: ahh
17:20 karolherbst: well.. the switch itself also needs power
17:20 karolherbst: but given that my switch is 8 port and 30W _per_ port, but only 55W total, yours might also allow more per port
17:21 anholt: afk
17:22 karolherbst: wow... ciscos page is as shitty as my opinion about them
17:23 karolherbst: "Each of the PoE ports on the Catalyst 2960 switches deliver up to 15.4 W of PoE." okay....
17:23 karolherbst: whtvr
20:36 Lyude: imirkin: yep! if you need my attention on something feel free to poke me via email or irc
20:38 ccr: PoE = Poke over Ethernet
20:38 imirkin: still waiting on this one: http://bash.org/?4281
20:39 ccr: I remember headlines from years (could be a decade now) about a company that was developing "smell over internet" or so
20:39 imirkin: uhh ... smell-a-vision?
20:39 imirkin: i thought that was a SNL thing
20:40 imirkin: or something along those lines
20:40 imirkin: scratch & sniff
20:40 ccr: https://en.wikipedia.org/wiki/Digital_scent_technology#1990s%E2%80%932000s
20:41 ccr: one can only imagine the possibilities for hacks/attacks on such technology :P
20:41 imirkin: wow, i didn't realize these things were real
20:42 imirkin: "audience members in the balcony complained that the scents reached them several seconds after the action was shown on the screen"
20:42 imirkin: making the sound "show up" at the same time across a whole movie theater is actually a surprisingly difficult problem. smell must be doubly so.
20:43 ccr: I'd be willing to bet that some day someone will get the idea of combining such thing with AR/XR/VR .. assuming the current AR/XR/VR hype survives long enough.
20:44 imirkin: "Current obstacles of mainstream adoption include the timing and distribution of scents, a fundamental understanding of human olfactory perception, the health dangers of synthetic odours, and other hurdles."
20:44 imirkin: so ... yeah. other than that, it's almost there :)
20:44 ccr: :)
22:26 anholt: karolherbst: yeah, looks pretty solidly like this switch can do 30w with a fancier splitter.
22:26 anholt: good news: more hardare support. bad news: more hardware support.
22:55 karolherbst: anholt: what do you mean by fancier splitter?
22:55 karolherbst: with its own power supply or something?
22:55 anholt: poe+ (.at rather rather than just .af)
22:56 karolherbst: ohh you meant in general
22:56 karolherbst: I thought with your switch
22:56 karolherbst: yeah.. you can do way more over poe if you really want to
22:56 karolherbst: 100W or something?
22:57 karolherbst: my netgear one does 30W as I mentioned
22:57 HdkR: poe++ is 100w :D
22:57 karolherbst: yep :P
22:58 karolherbst: W per port isn't the huge issue overall
22:58 karolherbst: overall power is a problem
22:58 karolherbst: so in the end it gets cheaper to buy few smaller switches than a 2kW monster
23:01 HdkR: At least microtik and unifi have some nice PoE switch options these days if you do want to go that route
23:02 karolherbst: HdkR: the huge issue is just, that with the devices we discuss here, you need a poe splitter for each anyway :(
23:02 karolherbst: or at least _something_
23:03 HdkR: Right, they don't support PoE
23:03 karolherbst: although there are ways to get poe support the jetson, just not at 48V
23:04 HdkR: Shame
23:04 karolherbst: huh.. there are also 48V hats...
23:05 HdkR: oh hah
23:05 karolherbst: or smaller modules like this one: https://www.iotamy.com/20W-PoE-Module-for-Jetson-Nano
23:05 HdkR: I guess you can't slap one on a Xavier though
23:05 karolherbst: but that's just a splitter
23:06 HdkR: Might as well as just buy a regular splitter at that point
23:07 karolherbst: yeah