<nbd>
yes, that's the repo i use for making the tarballs
<nbd>
since upstream backports is not actively maintained
<rmilecki>
thanks
<SpectreDev_01>
robimarko: can you help me fix this error I'm trying to get openwrt 23.05 working katb.in/itegedalixo
<robimarko>
SpectreDev_01: On a new device?
<SpectreDev_01>
robimarko: Yeah still the MX4200
<robimarko>
SpectreDev_01: I think its the old cpufreq driver that threw a BUG intentionally on unknown SoC
<SpectreDev_01>
Ah
<robimarko>
Honestly, why are you even trying to add a new device to 23.05
<SpectreDev_01>
Cause I keep getting OOM on snapshot
<SpectreDev_01>
After a few hours it'll just OOM and reboot
<robimarko>
You set the correct ath11k memory mode?
<SpectreDev_01>
robimarko: yeah I memory mode 1
<SpectreDev_01>
Worked on snapshot
<SpectreDev_01>
I tried nss cores, they loaded but no speed boost either and just eats up even more ram
<robimarko>
NSS obviously uses RAM
<robimarko>
Its a mess that is unsupported
<SpectreDev_01>
Yeah even without NSS it eventually just fills ram till 0 is free
<SpectreDev_01>
Could you add IPQ8174 CPU freq support?
<robimarko>
Its added to master
<robimarko>
I dont plan to add any new features to 23.05
<SpectreDev_01>
Ah
<SpectreDev_01>
Ok
goliath has joined #openwrt-devel
torv has quit [Remote host closed the connection]
torv has joined #openwrt-devel
torv has quit [Remote host closed the connection]
torv has joined #openwrt-devel
<neggles>
once more unto the breach, dear friends
<neggles>
trying to browbeat mikrotik into providing more up-to-date GPL sources
<neggles>
since the device i'm poking at is newer than the sources they provided and not mentioned at all :P
<neggles>
give me your cursed atl1c pcie endpoint driver you bastards
<robimarko>
Good luck :)
<f00b4r0>
mikrotik is a lost cause
<robimarko>
YAFFS FTW :)
<neggles>
"luckily" for me I am unemployed at the moment and used to bashing my head against brick walls :P
<neggles>
robimarko: ugh don't remind me... the CCR2004-1G-2XS-PCIe is extra cursed too
<neggles>
16MiB SPI NOR flash, almost entirely employ, contains only the annapurna bootloader that's an SPL-equivalent, routerboot's in... once again... parallel NAND
<neggles>
stop doing parallel nand ffs >:(
<robimarko>
Parallel NAND is perfectly fine
<robimarko>
Just use UBI instead of YAFFS
<neggles>
but why use parallel NAND when eMMC is cheaper, easier to route, faster...
<f00b4r0>
neggles: just stop giving mikrotik more business. They built it on opensource and won't give back, don't encourage that ;P
<robimarko>
Old school ONFI1 NAND cannot be more expensive than eMMC
<neggles>
unfortunately a lot of the time i don't really have a choice
<f00b4r0>
also what robimarko just said
<f00b4r0>
they're out to build at the lowest possible BOM cost
<neggles>
10ku prices for the NAND chip it's using are Not Low
<robimarko>
What is the model they are using?
<neggles>
im gonna have to go take the heatsink off again since i didnt write it down
<neggles>
but i need to do that anyway
<robimarko>
Its only 128MB of NAND, no way you can get eMMC cheaper than old school SLC ONFI NAND
<neggles>
(or rather i did write it down, 18 months ago, and lost my notes)
<robimarko>
Especially since they are buying everything in bulk
<robimarko>
Same reason why they are reusing 16MB NOR, because all of their boards use the same one
<robimarko>
But yeah, dont fund MikroTik
<robimarko>
I gave up on them completely
<neggles>
hm you're right i must have got this mixed up with something else
<neggles>
thinking about it i just remember being mad that it had very little storage even though it'd be great to run containers on
<neggles>
ah wait yeah. RB4011 is the one with the expensive 8Gbit NAND
<neggles>
I do generally try to avoid them these days, but I bought this card almost 2 years ago now :P at my old job i was pushing for little x86 boxes running OpenWrt or VyOS
<robimarko>
And then they picked Annapurna
<neggles>
I made some friends at annapurnazon recently
<robimarko>
Tell them to upstream their SoC-s
<robimarko>
Hell, not even a marketing flyer is available
<neggles>
they are trying but without a customer demanding it they cant get resources
rua has quit [Quit: Leaving.]
<neggles>
but one of them is looking into whether they can make the SDKs public again
<neggles>
from what i've been told management doesn't have any problems with upstreaming things, they just don't think it's worth spending any dev time on
<neggles>
nobody's quite sure how they ended up with mikrotik as a customer even
<robimarko>
As long as MikroTik is like the only large public customer it wont be upstream ever
<robimarko>
But they could at least provide specification sheets
<robimarko>
Cause currently, the only hits you get are for MikroTik products, no info on the SoC itself
<neggles>
yeah
<neggles>
only other uses of alpine chips i know of are some QNAPs and CheckPoint firewalls using first-gen
<neggles>
oh and a few netgears iirc
<robimarko>
All of that is EOL
<neggles>
yeah
<robimarko>
First gen was being upstreamed, then Amazon bought them and scrapped all of that
<neggles>
i get the feeling they're not really interested in being an SoC vendor or doesn't think there's much in the way of money to be made
<robimarko>
I thought that is kind of a thing they do, what else do they manufacture?
<neggles>
I mean, they make SoCs but they make them for AWS
<neggles>
the AL32400 in the CCR2004-1G-2XS-PCIe is also known as a 3rd-ish generation Nitro card, powering the Elastic Network Adapter
<neggles>
current-gen and last-gen Nitros use variants on the AL73400 that's in the CCR2116
<neggles>
im told they're kind of offended by how little effort MT have put into actually making use of the capabilities of these chips; the PCIe one could be doing so much more than pretending to be a weird nearly-20-year-old atheros nic
rua has joined #openwrt-devel
<robimarko>
Well, with the amount of new models that MikroTik makes and the pricing its not worth their time in customizing their SDK to support anything advanced
<neggles>
annapurna wrote none of that code
<robimarko>
I know, I am talking about MikroTik putting in the effortr
<neggles>
yeah
<neggles>
fair
<neggles>
idk i just find it strange that they'd go to all the effort of designing this piece of hardware, then do nothing to make it actually useful
<robimarko>
Its obviously selling just fine without any additional effort
<f00b4r0>
^
<neggles>
ehhh, not so much...
<f00b4r0>
marketing 101. Oversell and underdeliver.
<neggles>
the pcie card has been quite a flop, the regular CCR2000s are selling though
<neggles>
(here, at least, according to the two AU distributors)
<neggles>
they sold a bunch of them at launch (presumably to curious people like yours truly) and very few since
<neggles>
there is *some* hope though, Graviton2 is either already available to non-AWS customers or will be very shortly, and that's a chip which is actually powerful enough to be interesting to people other than the likes of mikrotik
nixuser has quit [Remote host closed the connection]
guidosarducci has quit [Remote host closed the connection]
<neggles>
hopefully someone who cares slightly more about upstreaming things (or is at least more inclined to meet their GPL obligations) will pick them up
nixuser has joined #openwrt-devel
guidosarducci has joined #openwrt-devel
fakuivan has quit [Remote host closed the connection]
<SpectreDev_01>
Uhmm is it a good idea to use Sqm on IPQ807x platform
<robimarko>
Its not going to be fast
<stintel>
I thought that was a modern platform? :P
<stintel>
I can do ~1000Mbps SQM on Firebox M300 (1.5GHz PPC)
<robimarko>
If we had a normal ethernet driver then yeah
<stintel>
I would assume IPQ807x would be able to do the same
<stintel>
I see :(
<robimarko>
We dont even have checksum offloading implemented
<robimarko>
If NSS was working then that would be another story
Mangix has quit [Read error: Connection reset by peer]
<SpectreDev_01>
robimarko: yeah it won't go fast but currently I get around 700mbps, but if sqm doesn't eat my ram up I can lower the bufferbloat at least
PaulFertser has quit [Remote host closed the connection]
philipp64 is now known as Guest10470
philipp64 has joined #openwrt-devel
Guest10470 has quit [Ping timeout: 480 seconds]
PaulFertser has joined #openwrt-devel
<SpectreDev_01>
Memory seems fine now
<SpectreDev_01>
Got past 2 hrs without OOM
autkin has left #openwrt-devel [#openwrt-devel]
mentalow has quit [Quit: :]]
mentalow has joined #openwrt-devel
PaulFertser has quit [Remote host closed the connection]
PaulFertser has joined #openwrt-devel
<f00b4r0>
ok so after testing 10 more devices it seems to be confirmed: WDR3600/4300 (ath79/ar9433) with Winbond RAM chips are fine, those with Zentel RAM are randomly getting stuck in reboot.
PaulFertser has quit [Remote host closed the connection]
PaulFertser has joined #openwrt-devel
<f00b4r0>
so I have to ask ye gurus of memory if they might have any clue what could possibly cause this, because this is quite beyond my skills?
SpectreDev_01 has quit [Quit: Connection closed for inactivity]
SpectreDev_01 has joined #openwrt-devel
<SpectreDev_01>
robimarko: yeah after about 6hrs it OOMs all I have is completely stock snapshot build and just DoT using dnsmasq and stubby
jkl has quit [Quit: Gone.]
jkl has joined #openwrt-devel
<SpectreDev_01>
Man this is weird
autkin has left #openwrt-devel [#openwrt-devel]
PaulFertser has quit [Remote host closed the connection]
PaulFertser has joined #openwrt-devel
<PaulFertser>
SpectreDev_01: are you looking at memory consuming userspace in "top" and probably at relevant parts of /proc/meminfo to see if the leak is inside the kernel? Also, if you suspect a kernel leak just enabling "kmemleak" can help a lot.
<SpectreDev_01>
PaulFertser: yeah top showed nothing out of the ordinary I think it was either software offloading or I need to check meminfo
<PaulFertser>
meminfo shows data from kernel allocators so if that consistently grows there must be a leak. And kmemleak is great at exposing the reasons for the kernel leaks.
fakuivan has quit [Remote host closed the connection]
PaulFertser has quit [Read error: Connection reset by peer]
<SpectreDev_01>
Ok thanks
PaulFertser has joined #openwrt-devel
hanetzer has joined #openwrt-devel
<robimarko>
I havent been able to track any leaks, my board has been running 24/7 as the AP and its not leaking
<SpectreDev_01>
robimarko: yeah I think it was software offloading causing my issue
<SpectreDev_01>
With Disney+ running on my TV with software offloading off it goes down to 76mb - 87mb ram free with about 22 devices connected
<SpectreDev_01>
Well now 70-77mb lol
autkin has joined #openwrt-devel
<SpectreDev_01>
Yeah I think it's a mem leak or something it's gradually decreasing in free ram
<robimarko>
There was a memory leak that was visible on 512M devices but I have not seen it in ages
<SpectreDev_01>
robimarko: yeah something isn't going right for me, as soon as I run a web game and Disney+ boom ram goes from 50-60mb to 32mb~
<SpectreDev_01>
PaulFertser: yeah Bluetooth module seems to be taking alot of ram up and doesn't even show up so I guess I'll drop that, there's also aquantia but I don't think Linksys mx4200 even has an aquantia chip https://katb.in/ocobabazojo
<PaulFertser>
SpectreDev_01: why not enable kmemleak?
<SpectreDev_01>
Yeah will do
noahm has quit [Read error: Connection reset by peer]
noahm has joined #openwrt-devel
autkin has joined #openwrt-devel
f23xhxxx has joined #openwrt-devel
<SpectreDev_01>
PaulFertser: after 5 mins of running kmemleaks scan, and then doing cat kmemleaks nothing showing so Ig there isn't a leak (so far(
<SpectreDev_01>
10 mins in nothing showing up
<PaulFertser>
SpectreDev_01: do you see numbers growing in /proc/meminfo ?
<SpectreDev_01>
PaulFertser: Yep caught a kmem leak now
<SpectreDev_01>
It's caused by hostapd
<PaulFertser>
SpectreDev_01: hope the backtraces there are useful.
<SpectreDev_01>
I'll send it to you
<PaulFertser>
To robimarco :)
<SpectreDev_01>
I'm no Dev in fixing stuff lol
<SpectreDev_01>
Sure
<PaulFertser>
I have no idea about ath11k
<PaulFertser>
I'd look at the traces because of curiousity
<SpectreDev_01>
katb.in/ubaxoqiquge
<PaulFertser>
color_collision_notify ! essential!
<PaulFertser>
You can try disabling colors in the wireless settings to see if it makes it stop leaking for you.
<SpectreDev_01>
Hmm ok
<PaulFertser>
But this problem should be reported to someone who can investigate the driver anyway, even if the workaround helps.
<SpectreDev_01>
Wait where is this colour collision
<SpectreDev_01>
PaulFertser: yeah after disabling 40mhz in 2.4ghz seems to have solved the problem 15 mins nothing coming from kmemleak
<PaulFertser>
SpectreDev_01: you mean you were using 802.11ax on 2.4 GHz and disabled it now? How is that related to HE colors?
<SpectreDev_01>
I was using 40mhz HE40 and switched it to HE20 and I haven't seen a logspam so far
autkin has left #openwrt-devel [#openwrt-devel]
<PaulFertser>
Interesting
<SpectreDev_01>
Yep very interesting lol
<PaulFertser>
SpectreDev_01: I meant to suggest trying "option he_bss_color_enabled 0" , unfortunately that option is not documented in the wiki
<SpectreDev_01>
Alright will try
<PaulFertser>
SpectreDev_01: you can check if it applied by looking into the generated hostapd config for he_bss_color_disabled=1
<PaulFertser>
SpectreDev_01: probably your conditions (with other APs working nearby on same channel) are special enough and so a generic mac80211 bug is exposed for you. Certainly worth reporting.
<SpectreDev_01>
PaulFertser: that config don't work hostapd don't recognise it and it gets killed
fakuivan has joined #openwrt-devel
<PaulFertser>
SpectreDev_01: are you adding that to wifi-device section?