shoragan has quit [Remote host closed the connection]
shoragan has joined #openwrt-devel
<tersono>
I would appreciate reviews of this small enhancement to the git download method: https://github.com/openwrt/openwrt/pull/13000 Please let me know if there's a specific process for requesting reviews or for moving ahead contributions, thanks!
minimal has quit [Quit: Leaving]
danitool has quit [Quit: Cubum autem in duos cubos, aut quadratoquadratum in duos quadratoquadratos]
slh has quit [Remote host closed the connection]
slh64 has quit [Quit: gone]
slh has joined #openwrt-devel
tSYS has quit [Quit: *squeak*]
tSYS has joined #openwrt-devel
rmilecki has quit [Quit: Konversation terminated!]
<rmilecki>
it seems phy0-ap0 got reconfigured after all, encryption seems ok
<rmilecki>
network.wireless notify has failed though
<nbd>
yep, i can reproduce that
<nbd>
should be harmless, but needs to be fixed anyway
slh64 has joined #openwrt-devel
<rmilecki>
nbd: ok, I've type of bug you're not gonna like
<rmilecki>
when after first boot (wifi devices OFF) I do: uci set wireless.radio0.disabled=0; uci set wireless.default_radio0.encryption=psk2; uci set wireless.default_radio0.key=password12345; - it works all fine
<rmilecki>
I can connect to my "OpenWrt" network using "password12345"
<rmilecki>
however if I do that in 2 steps: first I enable wifi device (then reload_config); then I add psk2 + key (then reload_config), I can't connect to my OpenWrt ssid
<rmilecki>
i didn't dump wireless traffic yet
<rmilecki>
however I don't even see association info on router side
<rmilecki>
I can see Mon Jul 17 08:01:27 2023 daemon.notice hostapd: Reloaded settings for phy phy0
<nbd>
can you please do this: /etc/init.d/wpad stop; sleep 1; hostapd -d -g /var/run/hostapd/global > /tmp/log.txt 2>&1 &
<nbd>
then please trigger the problem again and send me the logfile
<rmilecki>
should I do that right after first boot, or after brinding interface up with no encryption?
<nbd>
wifi needs to be restarted after hostapd restart anyway
<nbd>
typically that happens automatically
<Ansuel>
rmilecki ping ?
<rmilecki>
Ansuel: pong, but i'm a bit busy with nbd' stuff atm
<Ansuel>
just a quick question
<Ansuel>
can i use negative values for the nvmem layout ?
<rmilecki>
yes, you should
<rmilecki>
<&mac (-4)>
<Ansuel>
ok thanks
<rmilecki>
let me know if that doesn't work
<rmilecki>
nbd: i can nofirm this issue does NOT affect # strings /proc/device-tree/compatible
<rmilecki>
netgear,r6220
<rmilecki>
mediatek,mt7621-soc
<rmilecki>
*confirm
<rmilecki>
so this is brcmfmac think apparently
<rmilecki>
*thing
<nbd>
maybe something in the log will indicate which command sequence brcmfmac doesn't like
<rmilecki>
i'm switching back from mt7621 to broadcom but I'll continue testing after dinner
<bluecmd[m]>
rmilecki: thanks for the patch to vEdge 1000 for the Mac address. I'll try to make time to test it tomorrow
<bluecmd[m]>
We had it on our Todo to do this change either way so it was nice to find your patch
<rmilecki>
bluecmd[m]: great to have someone to test it :)
<neggles>
hello friends
<neggles>
is wpad-openssl still required for wpa3 + 802.11k/r/v
<neggles>
or will wolfssl do it
schwicht has joined #openwrt-devel
<Ansuel>
rmilecki can confirm the -1 works correctly
schwicht has quit [Read error: Connection reset by peer]
<stintel>
anyone notice stability issues with the latest hostapd bump? my phone seems to be losing connectivity regularly
<neggles>
hi stintel o/
<neggles>
cant help you on the hostapd front im afraid
<neggles>
tho actually while i remember, did you ever dig into those Xirrus XD2s?
<stintel>
yes!
<stintel>
they have Xirrus Boot Loader and I fear it's going to be a hard one :P
<stintel>
also the antennas are fragile, I noticed 1 broke off of the PCB
Danct12 has quit [Quit: WeeChat 4.0.0]
<rmilecki>
Ansuel: good ,thanks
minimal has joined #openwrt-devel
xavifr has quit [Ping timeout: 480 seconds]
<neggles>
stintel: interesting. I never got around to hooking a console up to mine, but, at last check the octeon they use doesn't actually support any meaningful secure boot so it shouldn't be horrible
<neggles>
i am somewhat confused by the presence of SPI NOR, parallel NAND, *AND* eMMC though. w-why
<neggles>
guess i should pester someone at cambium for a gpl dump
<stintel>
they do have a TTL header
<stintel>
TTL pinout:
<stintel>
GND VCC RX TX LAN2 LAN1
<stintel>
[Boot ] Address : 0x20000000 ... ### Wrong Image Format for boot Command ###
<stintel>
is as far as I got
<stintel>
didn't have much time to dig further
xavifr has joined #openwrt-devel
<neggles>
alrighty, i shall do some poking at some point
<neggles>
been off in AI shenanigans land building chatbots
<neggles>
and messing around with shady chinese NPUs
<stintel>
haha I've been doing ESP32 development
<neggles>
I bought two nvidia A40s for this stuff because I am insane :P oh? whatcha building?
<rmilecki>
nbd: logging doesn't show anything interesting on reload after enabling psk2: https://pastebin.com/26N6r30J
<neggles>
I knew I'd regret not buying one of the "I can't believe it's not M5Stack" ESP32-S3 things
<rmilecki>
it's from -dd and with "procd_set_param stdout 1" and "procd_set_param stderr 1"
<stintel>
we will add support for M5Stack S3 before end of the month
<rmilecki>
nbd: i tried your command too but didn't get anything more
<neggles>
oh there's an m5stack s3? (I meant the ESP-BOX, which is Absolutely Not The Same Form Factor No Sir Nuh Uh)
<stintel>
ah :)
<stintel>
well 1 week after we launched Willow mouser adafruit and digikey were sold out
<stintel>
but right now digikey has stock
<neggles>
i think espressif's official aliexpress store has stock too
<stintel>
yeah quite possible
<neggles>
whisper's quite good
<stintel>
I just tend to stick to mouser, very fast delivery and I don't have to deal with customs
<neggles>
it's partially integrated into my chatbot stuff, but dealing with parsing text chat is difficult enough without adding multiple people talking at once into the "ok so when do I trigger a response?" logic
<stintel>
hah I can imagine
<stintel>
I'm running a local inference server on a P4
<neggles>
yeah if I was in the EU i would too, but down here our customs' attitude to things under AU$1k declared value is "meh"
<stintel>
considering getting a T4 but waiting until price drops a bit
<neggles>
T4 suffers a bit from power limits
<neggles>
but it ought to be about 4x the speed of a P4, or something along those lines - P4 can't actually do fp16 math, it lies
<stintel>
hmmm wait the T4 is x16
<neggles>
you can't argue with how stupid cheap they are, mind you, but Pascal's a complete wash for ML stuff
<stintel>
I thought it was x8 and would fit in my r515 without molesting a slot on the PCIe riser
<neggles>
A2 is x8
<stintel>
(they're x8)
<neggles>
but A2 is kinda mid tbh, it's slower than a T4
<stintel>
I see
<neggles>
not by a lot but by a little
<neggles>
(I have one)
<stintel>
well I already molested one riser for the P4
<stintel>
so a T4 would fit :P
<stintel>
and I found 3 risers in some German shop so decided to buy them all 3
<neggles>
right now in the US/AU at least, a used A4000 is better value than a T4, for some reason T4s still go for ludicrous prices
<stintel>
I have 2 r515 - now with spare risers I can risk messing one up without issue :P
<neggles>
though that said, I do have a friend who's trying to sell one rn for uhhh
<neggles>
~365EUR+postage, whatever that'd be
<stintel>
o_O
<stintel>
is that friend in Europe ?
<neggles>
Perth, sadly, but he's happy to ship, should be about 40EUR
<stintel>
and customs :(
<neggles>
he's from sweden lmao
<neggles>
eh, would it be that bad if it's marked with a low value and non-commercial?
<stintel>
yes
<neggles>
booo
<stintel>
the xirrus took 4 months because retarded
<neggles>
*yikes*
<stintel>
1 month to get from UK to BG
<stintel>
then it was shipped from Northern Ireland which should be handled as from within EU
<stintel>
but nobody here knows that
<neggles>
of course not...
<stintel>
so I declared the value and said fuck it i'll pay the tax
<stintel>
then 3 months later I get an SMS " it's in the post office "
<stintel>
had almost forgotten about them
<neggles>
h-uh. well, fwiw he does still work for a company in sweden so might be able to pull a string or two, i can ask
<neggles>
man makes more money than is remotely reasonable, dropped AU$50k on a shiny new dev system without even really thinking about it
<neggles>
but yeah uh. it helps to remember that the T4 is almost literally a 2070 with a 75W power limit and 16GB
<stintel>
I've been looking at getting a beefy epyc workstation at some point but damn
<neggles>
it's not super zippy
<neggles>
heh
<stintel>
it's ok, it's going to be better than the P4 ;)
<stintel>
and I can keep the P4 in the "testing" r515
<neggles>
yeah and if you're stuck with HHHL+75W it doesn't get much better
<stintel>
and setup a redundant inference server!
<stintel>
I even limit the P4 to 60W
<neggles>
at least not without an L4, and, well, they're US$6k
<stintel>
because limited power on the PCIe riser in the r515
<neggles>
ah
<stintel>
and I do have an HBA and x540-t2 in there also
<neggles>
wait r515 as in the cursed opteron dell?
<stintel>
yes
<stintel>
didn't know it was cursed :P
<neggles>
...whyyyyyy
<neggles>
all those opterons are cursed afaik
<stintel>
because 12x LFF disk bay and could get them for cheap :P
<neggles>
worse power efficiency than westmere
<stintel>
but it's ~550ms to STT a typical home assistant intent
<stintel>
the P4
<stintel>
so it's quite usable
tetienne has joined #openwrt-devel
<neggles>
interesting
<neggles>
i've found current STT stuff to be very underwhelming
<tetienne>
Hi all
<neggles>
TTS is cool and good but speech synthesis is... iffy
<neggles>
o/
<stintel>
o/
<stintel>
yeah. "turn on desk light" -> "turn on this light", "turn on disc light"
<stintel>
you need to choose your words carefully :P
<neggles>
depends on how big of a model you run ofc
<stintel>
I should probably add an alias "workstation" so I can say "turn on workstation light"
<stintel>
even on large it's not reliable enough
<stintel>
or maybe my English just sucks :P
<neggles>
ohhhh right of course
<neggles>
it's probably the accent
<bluecmd[m]>
<neggles> "someone introduced me to Xianyu,..." <- Huh, really cool. You have now transferred the curse to me. Do you have any experience with things like dropshipping/mailboxes/or just simply asking sellers to ship to e.g. Europe?
<neggles>
superbuy
<neggles>
are who i use
<neggles>
xianyu won't even let you talk to sellers without a verified chinese ID, and it's only accessible on mobile app or via the link share webui
<neggles>
so superbuy hands your link over to a local agent who handles the local seller
<neggles>
probably going to pick up one of the mATX Kunpeng 920 boards that are in the $200-300 region, just because funny 7nm chinese arm chip; been trying to find one of the 24-core W510 boards but they're rare
<tetienne>
I have a small question (sorry to disturb your discussion). Are there people building their OpenWRT image bins on a gentoo OS? On my gentoo the package libnl is always failing, whatever the version 3.5.0 or 3.7.0. In a Debian virtual host it's working fine, I would be curious if other people got the same problem.
<neggles>
stintel: you're up :P
<neggles>
tetienne: iirc stijn rolls with gentoo+musl so may be able to help you out, otherwise I can't think of a reason why that wouldn't work, do you have a dump of the output from a failure with `make V=sc` ?
<bluecmd[m]>
I had some issues on NixOS with libnl because NixOS enables some extra warnings
<tetienne>
neggles: seems to have to do with bison stuff byacc -d -o lib/route/cls/ematch_syntax.c lib/route/cls/ematch_syntax.y
<tetienne>
byacc: e - line 28 of "lib/route/cls/ematch_syntax.y", syntax error
<bluecmd[m]>
format-security
<bluecmd[m]>
Is what I had issues with
<bluecmd[m]>
Fwiw
<tetienne>
I'm trying to make back to back comparisons to understand where it comes from
<tetienne>
but I'm also surprised to get that kind of issue since the toolchain is supposed to be the same with the same tolls versions
<neggles>
yeah that's the whole point of building and using the local toolchain
<neggles>
i guess it doesn't BYO yacc though
<neggles>
`%name-prefix "ematch_"` seems like a weird place to fail
<rmilecki>
nbd: hostapd is also very silent on mt7621 when calling "reload_config" after seting psk2
<neggles>
tetienne: bison versions?
<neggles>
> Prior to Bison 2.6, a feature similar to api.prefix was provided by the obsolete directive %name-prefix and the option --name-prefix.
<tetienne>
neggles: I was thinking about bison versions, but they are compiled in the toolchain, it's supposed to be the same ?
<tetienne>
on gentoo I have 3.8, on Debian it's 3.7
<tetienne>
I've just recompiled a 3.7 local version on my gentoo to see if it can influence the whole process
<hauke>
I am using debian stable with bison 3.8.2 and it works fine
<hauke>
maybe you deactivated some options in your bison build
<neggles>
¯\_(ツ)_/¯ sounds like it, there's probably a flag to enable the apparently-long-deprecated name-prefix thing
<tetienne>
good to know hauke, on gentoo there are not that much USE flags, but again, the toolchian should use it's own bison install, no? I've seen during the toolchain compilation a bison line
<neggles>
might be a config file thing?
<tetienne>
tetienne@P15 ~/OpenWRT/openwrt-22.03.5 $ ls dl/bison-3.8.2.tar.xz
<tetienne>
dl/bison-3.8.2.tar.xz
<tetienne>
from what I can compare, the ematch_syntax.y files are exactly the same from my build on a gentoo and another working build on debian
<Slimey>
giggity
<Slimey>
careful, deviwiki is trying to serve spicy malware from ads
<tetienne>
from libnl Makefile I suspect it is using my system byacc instead of the one built during the toolchain compilation
<Znevna>
ads? :p
<Slimey>
=0
<tetienne>
neggles: think I've found, on my gentoo it's using YACC = byacc (probably because byacc is installed on my system) where as on Debian it's using YACC = bison -y (byacc not present on that debian install)
danitool has quit [Remote host closed the connection]
<stintel>
neggles: tetienne: my dev workstation is Gentoo glibc and usually builds OpenWrt fine. I do hit host lib leakage more often than most others because everything installs with headers
<tetienne>
stintel: check if you have byacc installed on your gentoo, which is the root cause of the issue
<tetienne>
I've just checked that export YACC='bison -y' fixes the problem for libnl
<stintel>
no byacc and I also don't have libnl enabled in any of my builds
<stintel>
but I guess the libnl Makefile needs some change to use bison from toolchain
<tetienne>
I could not find any option in the configure script, but it seems to be sensible to YACC variable, so export YACC='bison -v' seems to do the job, I'll post a remark on git to robustify the whole process
<nbd>
rmilecki: don't change the init script to start hostapd with -d - please run it from the command line with output redirected to a log file
<nbd>
oh, wait
<nbd>
i also forgot something
<nbd>
you need to set CONFIG_WPA_MSG_MIN_PRIORITY=0 in the config
<nbd>
i mean in the openwrt .config
<nbd>
otherwise it gets compile without debug messages
goliath has quit [Quit: SIGSEGV]
<stintel>
I'll upload an OEM bootlog soon - it's very silent and shows some non-standard things
<stintel>
[Image ] Type : MIPS Linux Multi-File Image (uncompressed)
<stintel>
oh
<stintel>
so Multi-File Image seems to be a standard U-boot thing - had not seen it before
<stintel>
let's see if I can just feed it a uImage