<Ansuel>
\x well pppoe + natting + wifi usage combined for sure will cut some mbps
<Ansuel>
for qos no chance for ipq40xx
<Ansuel>
the switch supports some policy but not to the level of sqm
<\x>
as for wifi, its like this, I can put an AX radio in, then if i run the iperf server right on that radio, I can get 900Mbps, but when I run the server on another server on the lan its like 450~550
<\x>
kinda yeah, I guess large hit from bridging
<Ansuel>
it's all magic once you tranfer packet on the same "subsystem" once you start moving things across them there goes your perf
<\x>
yeah
<Ansuel>
on ipq807x they fixed this by putting nss cores in the middle... when both ethernet and wifi are offloaded everything pass with the nss cores 99% of the time
<Ansuel>
problem is that you need tons of ram LIKE A LOT
<Ansuel>
that's why ipq807x works bad with less than 1gb
<\x>
like what, i still havent experienced havin an 807x system
<\x>
actually deciding if its 807x or mt7986
<\x>
or maybe ill go cheap ass and go 60xx
<Ansuel>
\x driver is shit, we have no documentation of it so we can't fix it... we are stuck with qsdk driver
<Ansuel>
that's all folks :D
<stintel>
mtk or gtfo :P
<\x>
so whats the difference with qsdk builds?
<Ansuel>
60xx is not supported
* stintel
hides
<\x>
just newer kernel?
<Ansuel>
\x no xiaomi bs and a working wifi
<Ansuel>
also in theory we in the future we can consider wifi offload
<Ansuel>
had some fun with it and works good but i need to fix some small thing...
Guest236 has quit [Ping timeout: 480 seconds]
<\x>
<Ansuel> it's all magic once you tranfer packet on the same "subsystem" once you start moving things across them there goes your perf
<\x>
that iperf run is the reverse on mt7621 stuff
<\x>
when you run iperf server inside the ap, its slower
<\x>
when you run it on another server wired toi the ap its faster hehe
<Ansuel>
\x that's an exact example of how things are offloaded under the hood... traffic from one port to another don't even reach the kernel
<Ansuel>
they are handled by the switch directly
<hurricos>
neoraider: talk to me about https://git.openwrt.org/?p=openwrt/staging/neoraider.git;a=commitdiff;h=bec095597197165e2962f2979157dfca01c4dc14#patch8
<nbd>
regarding perf: with mt7622 i was able to do ~1.8+ Gbps LAN->WLAN purely in software, and it only has 2 cpu cores. mt7986 has 4 cores, so i expect the non-offload performance there to be very good as well
<Ansuel>
nbd wow nice perf o.o
valku has joined #openwrt-devel
<nbd>
and on top of that, we will have offload support for it soon
<nbd>
it's already being worked on
<nbd>
with proper upstream code, not qsdk hackery
<Ansuel>
i assume offload won't require a special firmware
<nbd>
right
<nbd>
the offload core doesn't have firmware
<nbd>
it will remain opt-in via module parameter (because enabling offload support slightly reduces perf of the pure software path)
<Ansuel>
was thinking of a real way to make an upstream driver for nss core
<Ansuel>
but can't be...
<Ansuel>
nightmare to handle different firmware version
<Ansuel>
the struct HAS to match nss firmware version or everything is silently ignored
<nbd>
with qualcomm firmware stuff, everything's a nightmare, not just the offload path
<nbd>
heard a lot of stories from john about BDF incompatiblities between SDK versions
<nbd>
so the BDF data needs to match the SDK version, otherwise the radio doesn't work or perf just sucks
<nbd>
and there's no proper way to convert
<Ansuel>
about that
<Ansuel>
xiaomi bdf are ancient
<Ansuel>
guess what crash with new firmware ?
<Ansuel>
:D
<nbd>
:D
<Ansuel>
still have to find time to create a bufilla
<\x>
oh yeah Ansuel, I think youre also one of the guys that did this, real good. wifi5 on 2.4! https://i.imgur.com/uJac0EN.png Thanks!
<Ansuel>
bugfilla* and be ignored
<Ansuel>
\x you mean vht on 2.4?
<\x>
yeah
<Ansuel>
it's a non standard feature but practically every device supports it...
<Ansuel>
IMHO we should reconsider adding support for it in openwrt
<nbd>
makes sense
<\x>
Ansuel: lots of phones do support it
<\x>
snapdragon ones
<Ansuel>
well the hack is abused by brcm and qcom mainly so
<\x>
instead of 72 on 1x1 20, they get 96, 150 becomes 200
<Ansuel>
guess what most device have?
<Ansuel>
but i think even intel supports it
<\x>
Ansuel: actually they do as an AP
<Ansuel>
nbd it's really a simple patch for mac80211
<\x>
i tried it on ax200 on 2.4GHz, with that vht patch, they allow clients to connect as vht on 2.4GHz
<Ansuel>
the drives has to support it but for example ath10k-ct always have the change baked in
<\x>
i reported on pr 2522
<\x>
idk how to make intel as a client support it though
<\x>
i tried enabling it on 8260 sometime ago, it complains on some weird thing I forgot, basically something "2 < 4" limit blah blah
<stintel>
Tapper: can't seem to find a lot of info about what cpu is actually in the r7800 - I find dual core 1.7 ARM v7 - if that's the case yes rt3200 CPU definitely better than r7800
<\x>
i can retry sometime if you guys need more investigations on pr 2522
<Ansuel>
\x pls do if you have time
<stintel>
but this could be an armv8 with the qca retardedness they only support arm32 in qsdk
<\x>
Ansuel: ill just write the error, i dont do any programming really
<Ansuel>
stintel with new qsdk they dropped 32bit support entirely
<nbd>
Ansuel: i think i'd prefer not setting has_80mhz for 2.4 ghz, that seems rather hackish
<stintel>
Ansuel: but did they manage to stabilize it yet :P
<stintel>
Tapper: so ipq806x is krait is armv7 so yes, anything in the mediatek target > r7800 I'd say
<Tapper>
OK thanks.
<Tapper>
I am going to look on ebay now.
<Ansuel>
stintel fun thing is that ipq807x hw assume things are still in 32bit and because of this all the switch dma descriptor needs specail hack to be handled cause system is 64bit but descriptors expect a 32bit address
<Tapper>
Any ones to look for besids the rt3200?
<Ansuel>
nbd ok i will write that down and resping a new patch for it
<stintel>
Tapper: well I've a unifi 6 LR - it has 2.5GbE uplink - the only problem with everything in mediatek afaik is that 2.4GHz is wifi4
<nbd>
also, i think the patch should be posted upstream for review
<nbd>
and then pulled into openwrt from the upstream submitted patch
<Ansuel>
mhh don't know at times they tried but it was NACK as it was not standard for 2.4
<nbd>
if pretty much everything out there supports it, i think that's a good reason to also support it in mac80211
<\x>
Ansuel: ill demo ax200 first. how do i list clients again and infos from a partcular radio?
<nbd>
just need to argue it well in the submission
<stintel>
if it improves airtime efficiency this is a very good argument to give
<stintel>
especially since 2.4 is already a total disaster
<stintel>
*any* effort to improve airtime efficiency there is a good thing
<nbd>
i can also help with discussing this with johannes if he objects again
<\x>
for 8260, I have to find my ufl pigtails first so it might be tomorrow or something ahaha
MAbeeTT5 has joined #openwrt-devel
<Tapper>
hahahaha Some one has the Linksys E8450 flashed with OpenWrt on Ebay and is selling for £256.00 must be smocking crack if they think I would pay that price!
<\x>
though yeah I dont like these intel ones, no mumimo unlike the 2.4GHz radio on my ipq4019
MAbeeTT4 has quit [Ping timeout: 480 seconds]
<schmars[m]>
downloads.openwrt.org not reachable?
<schmars[m]>
trace stops at hetzner's router for both ipv4 and ipv6
soxrok2212 has joined #openwrt-devel
<\x>
<nbd> if pretty much everything out there supports it, i think that's a good reason to also support it in mac80211
<\x>
so far, its mostly phones with snapdragon on it, not sure with mediatek dimensity ones, but yup even lowest end snapdargon 4xx chipsets supports the feature
goliath has quit [Quit: SIGSEGV]
<Ansuel>
aparcar can confirm
<Ansuel>
doing traceroute with ipv6
<aparcar[m]>
Ansuel: you summoned me?
<Ansuel>
OH WAIT
<\x>
anyway Ansuel, I kinda dont want to test the 8260 tonight, its 11PM here and I need to find a ufl antenna pigtail for it first
<Ansuel>
got confused by the m
<\x>
but theres your AX200 and 9260AC confirmation that the feature works
<Ansuel>
schmars yes can confirm
<Ansuel>
(anyway aparcar if you want IMHO we can merge the CI pr)
<\x>
tomorrow I will show the issue
<aparcar[m]>
there are about 5 CI PRs, which one do you mean?
<Ansuel>
the tools container one
<Ansuel>
also jow ping ?
soxrok2212_ has joined #openwrt-devel
soxrok2212_ has quit [Remote host closed the connection]
soxrok2212_ has joined #openwrt-devel
soxrok2212 has quit [Ping timeout: 480 seconds]
cbeznea1 has quit [Quit: Leaving.]
MaxS0niX has joined #openwrt-devel
MaxSoniX has quit [Read error: Connection reset by peer]
<\x>
Ansuel: so 2522 is gonna be rebased? ahaha, I follow the one on lean, they rebase it as needed
<Ansuel>
\x think i will open another pr as we need to make some changes to it also i have to propose it to upstream
<Ansuel>
so long way to go
<\x>
yeah its fine if you take your time, its easy to do for me as i use ath10k-ct so, hostapd and mac80211sh edit and that one patch
<\x>
i just edit locally
<\x>
would be nice if accepted upstream yeah, lots of 256-qam capable radios are wasted being 64-qam only on 2.4GHz
<\x>
being used as 64-qam only on 2.4GHz*
<Ansuel>
problem is that i really need to write a good submission text
<Ansuel>
need to give tons of reason to have that accepted
<\x>
well, lots of android devices support the feature, thats one
<\x>
newer radios come with support but unused, like MT7622's built-in radio
<\x>
with vendor you can get 4x4 40 at 800Mbps max PHY, with openwrt you get 3/4 of that.
<\x>
I think theres some talk on RT3200 thread about it. some people dont want it though even when it really doesnt affect anything so idk really
tohojo has quit [Quit: Client limit exceeded: 20000]
noltari has quit [Ping timeout: 480 seconds]
goliath has quit [Quit: SIGSEGV]
<ynezz>
another evening, another cmake compilation
<Ansuel>
ahahahah
<Ansuel>
ynezz do you have access to git server?
dangole has quit [Remote host closed the connection]
dangole has joined #openwrt-devel
ekathva has joined #openwrt-devel
ekathva has quit [Remote host closed the connection]
<Ansuel>
will ansuel manage to make the git server push commit as openwrt-bot instead of jow? ahahha
<hurricos>
Slimey: rip
shibboleth has joined #openwrt-devel
borek has quit [Ping timeout: 480 seconds]
goliath has joined #openwrt-devel
Borromini has joined #openwrt-devel
<ynezz>
Ansuel: what do you need?
<ynezz>
tmn505: could you please refresh the tegra 5.15 config and send v2?
<Ansuel>
ynezz there is currently a problem from a long time. our git server push commit to github with jow username. We should change it to github-bot as when issue are autoclosed jow- is the user that close them (and this is wrong)
<Ansuel>
aparcar have access to github-bot but wanted to know what keys are linked to jow account on git server
<Ansuel>
(ideally a correct fix would be make the user who commited the commit the one who push the commit to github.com)
<aparcar[m]>
I imagine the easiest way is to drop the public key from Jows account and add it to the bot
<Ansuel>
aparcar yep
<aparcar[m]>
Im assuming jo does not use that key anywhere else
<aparcar[m]>
So it’s actually a thing to manage between me and Jo
<Ansuel>
another solution would be add another key while jow removes it but i think moving the key permit to don't touch the git server
<ynezz>
tmn505: Generic device cooling support (DEVFREQ_THERMAL) [N/y/?] (NEW)
<ynezz>
tmn505: I wanted to refresh that config, but it needs toolchain and I'm lazy to solve that
minimal has quit [Quit: Leaving]
<jow>
aparcar[m]: can you grant me access to openwrt-bot, then I can add the key there
<jow>
and drop it from my account. it is indeed a dedicated key pair, so transferring it should be no issue
<tmn505>
ynezz: I'll do the refresh.
tlj has joined #openwrt-devel
xian07 has joined #openwrt-devel
tlj_ has quit [Ping timeout: 480 seconds]
soxrok2212 has quit [Read error: Connection reset by peer]
soxrok2212 has joined #openwrt-devel
shibboleth has quit [Quit: shibboleth]
xian07 has quit [Quit: Page closed]
<slh>
Tapper: if your r7800 is coping well with your current WAN speed and everything you throw at it, I'd defer shopping a little bit. ipq807x support for master isn't far away - and filogic 830 is coming as well - on the other hand, in your particular geographical location, you might be lucky to find cheap linksys rt3200 on the second hand market
torv has quit [Remote host closed the connection]
torv has joined #openwrt-devel
soxrok2212 has quit [Read error: No route to host]
soxrok2212 has joined #openwrt-devel
dangole has quit [Remote host closed the connection]
<jow>
seems openwrt-bot isn't invited to the latter
<jow>
can be easily fixed though
<zorun>
jow: I'm writing up about this conntrack/fw4 issue, and I'm looking to reproduce without reboot: do you know what is the state of nftables before fw4 starts?
<zorun>
I can't reproduce with a simple "fw4 stop; sleep X; fw4 start"
<zorun>
while fw4 is stopped, "nft list ruleset" just returns nothing
<Ansuel>
jow thanks a lot for the change. Can finally flag pr as merged wonder if I should add a guide on how to do that on the wiki
<jow>
Ansuel: do you have something else to push?
<Ansuel>
yes
<jow>
can you push it and see if the denied issue is solved?
<Ansuel>
ok give me a sec
Borromini has joined #openwrt-devel
<Ansuel>
yep worked
<Ansuel>
fixed
<jow>
great
<Ansuel>
mhhh where should i write the steps on the wiki....
<jow>
Ansuel: how do you flag a PR as merged?
<jow>
I usually write Fixes: #1234 or Closes: #1234 before my S-o-b
<jow>
this is enough for github to auto-close the mentioned issue/pr
<jow>
but then it is closed as unmerged. I suppose you found a way to close it as merged eventhough it wasn't merged through GH
<jow>
?
<Ansuel>
oh the closes works also for pr? currently my way is checkout the branch of the user, rebase on top of master, force push on his branch, checkout master and then git merge --ff-only the branch and finally push... this way the commit match and github autodetect that the pr has merged... quite a trip and i think only me will follow this way ahahah
<Ansuel>
(but i guess it's easier if i explain this with commands)
<jow>
ah yes. Can be maybe automated with some script