<rsalvaterra>
Whoever gets to add support for the next LTS kernel is going to have a bad time.
robimarko has joined #openwrt-devel
<f00b4r0>
yum
<robimarko>
It was damn time for that
<robimarko>
Currently they are all in one folder and its just a mess
<robimarko>
BTW, anybody here knows how to use DT overlays with FIT images and U-boot?
<robimarko>
I tried following the upstream docs but it just fails to do its magic
schwicht has joined #openwrt-devel
schwicht has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
schwicht has joined #openwrt-devel
schwicht has quit []
schwicht has joined #openwrt-devel
schwicht has quit []
tersono has joined #openwrt-devel
<tersono>
Hi đź‘‹ I'd appreciate reviews of this small PR that makes the submodules to clone configurable for the git download method: https://github.com/openwrt/openwrt/pull/13000 Thanks!
<gch981213>
robimarko: dangole used that for BananaPi R64 images in mediatek.
<robimarko>
gch981213: Thats a good tip, thanks
<gch981213>
yw :)
schwicht has joined #openwrt-devel
schwicht has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
schwicht has joined #openwrt-devel
schwicht has quit []
schwicht has joined #openwrt-devel
schwicht has quit []
danitool has joined #openwrt-devel
minimal has joined #openwrt-devel
schwicht has joined #openwrt-devel
schwicht has quit []
Mangix has joined #openwrt-devel
schwicht has joined #openwrt-devel
schwicht has quit []
schwicht has joined #openwrt-devel
schwicht has quit []
schwicht has joined #openwrt-devel
<nbd>
rsalvaterra: ping
<nbd>
rsalvaterra: in a thread that's investigating a mt7621 performance regression, somebody mentioned that enabling cpuidle support and dynticks was causing a significant performance regression
<nbd>
do you by any chance remember if you measured any real benefits to enabling those?
<nbd>
i will run some performance tests soon to see if i can reproduce the issues
<rsalvaterra>
nbd: At the time I haven't measured, and I haven't noticed any difference either. If it causes such a performance decrease, it should be disabled of course.
<nbd>
thx
<nbd>
i'll see what my measurements turn up
<rsalvaterra>
It'd be unfortunate, though, as cpuidle and dynticks should save a bit of power.
<rsalvaterra>
Oh, well…
<nbd>
not sure how much of a difference it would make in practice on these devices
<Znevna>
how about heat?:P
<nbd>
so it does seem like there's a really nasty performance regression somewhere on mt7621
<nbd>
while it's not caused by cpuidle, cpuidle does contribute to it
<nbd>
i ran some NAT tests without flow offloading
<nbd>
with cpuidle: 110-130 Mbit/s, without cpuidle: 155 Mbit/s
<nbd>
will test dynticks next
<rsalvaterra>
nbd: Maybe cpufreq makes a difference too, then.
<nbd>
will check
<rsalvaterra>
nbd: Thanks!
<rsalvaterra>
nbd: By the way, what about those ath10k-ct threaded NAPI patches? I'm running them on my Omnia basically since they exist, and I haven't observed any ill side-effects. :)
<rsalvaterra>
Also, what GCC/Binutils combo are you using? I can't build ramips/mt7621 at the moment, some libraries are failing to build due to MIPS16 relocation issues (liblua, libselinux, for example).
<rsalvaterra>
(I'm using GCC 13 and binutils 2.40.)
* rsalvaterra
wonders if it's LTO and/or gc-sections related…
schwicht has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<nbd>
rsalvaterra: dynticks doesn't seem to make much difference
<rsalvaterra>
Ok, that's nice to know, at least.
<rsalvaterra>
Actually, the MT7621 doesn't have cpufreq, it seems.
ptudor has joined #openwrt-devel
<Christophe[m]1>
ndb: Would you share some details on how you're benchmarking?
schwicht has joined #openwrt-devel
schwicht has quit []
schwicht has joined #openwrt-devel
<rsalvaterra>
I think I got it… LTO really doesn't play along with MIPS16 ASE.