<slh>
I'm still wary if that isn't a more generic issue, not restricted to ipq806x
Tapper has quit [Ping timeout: 480 seconds]
Rentong has quit [Ping timeout: 480 seconds]
<dangole>
fda: i tortured the image you sent me a lot now and weirdly i cannot reproduce the problem. i've set a password, changed a couple of settings, rebooted a couple of times, ...
figgyc has quit [Ping timeout: 480 seconds]
figgyc has joined #openwrt-devel
KGB-0 has quit []
danitool has quit [Ping timeout: 480 seconds]
Tusker has joined #openwrt-devel
Rentong has joined #openwrt-devel
Rentong has quit [Ping timeout: 480 seconds]
Tusker has quit [Ping timeout: 480 seconds]
guidosarducci has quit [Remote host closed the connection]
guidosarducci has joined #openwrt-devel
GraphicHealer has quit [Quit: Connection closed for inactivity]
Tapper has joined #openwrt-devel
valku has quit [Remote host closed the connection]
guidosarducci has quit [Remote host closed the connection]
guidosarducci has joined #openwrt-devel
<digitalcircuit>
If it *is* power supply related, that might correspond with me using a USB 3.0 HDD - perhaps writing to a USB-powered HDD coupled with using the CPU (including L2 cache) is just enough to tip everything over the edge.
<digitalcircuit>
slh: Hmm... Hypothesizing, might it be a timing issue or race condition that's only (more easily?) triggered when the L2 CPU cache is allowed to run faster on ipq806x (i.e. 1.4 GHz)?
<digitalcircuit>
Also I seem to have picked quite the fun issue for my first attempt at contributing to OpenWRT :)
<digitalcircuit>
Ansuel did respond via email on mobile (which didn't appear to be received by the mailing list for some reason) - remarking it might be an issue with chip degradation or bad power supply, suggesting to tinker with the OPP table to raise the voltage to the CPU as that reduced instability for some routers.
<digitalcircuit>
(Alongside firmware tweaks, I s'pose I could try getting a powered USB 3 hub or otherwise better isolate the router. I do have concerns this might impact others though, so I'd rather try to fix the issue instead of simply working around it just for myself.)
<digitalcircuit>
(Might be time to buy a controlled USB load tester/power resistor, or a device to measure the power draw of USB devices.)
Rentong has joined #openwrt-devel
Rentong has quit [Ping timeout: 480 seconds]
Tapper has quit [Ping timeout: 480 seconds]
KGB-0 has joined #openwrt-devel
Tusker has joined #openwrt-devel
Tapper has joined #openwrt-devel
Tapper has quit [Ping timeout: 480 seconds]
nitroshift has joined #openwrt-devel
goliath has joined #openwrt-devel
Tapper has joined #openwrt-devel
jordbaer has joined #openwrt-devel
jordbaer has quit [Quit: Leaving...]
jordbaer has joined #openwrt-devel
jordbaer has quit []
aleksander has joined #openwrt-devel
jordbaer has joined #openwrt-devel
rejoicetreat has joined #openwrt-devel
guerby_ has joined #openwrt-devel
guerby has quit [Read error: Connection reset by peer]
decke has joined #openwrt-devel
danitool has joined #openwrt-devel
<mangix>
turns out fedora has debootstrap and pacstrap...
<mangix>
is nspawn the lowest weight container?
Borromini has joined #openwrt-devel
<fda>
dangole: what? how could that be??? why do MY images dont work in MY device? but prebuilt images work and my image workd for you?
<fda>
maybe i sould start form begin again, go back to non ubi first? i noticed 3 files in ubi0_3 / boot_backup partition: mtd0=512kb mtd1=1280kb mtd2=1024kb
<dangole>
fda: yes, very odd. only difference is maybe that on my device i got serial terminal connected
<dangole>
fda: i don't think this problem is realted to anything you could fix by running the installer again. it's not that low-level (unfortunately, that'd be easy to fix), but rather complex high-level
<dangole>
attaching serial terminal would be the first thing i'd do to examine the device when it's in non-responsive condition
<fda>
hm, with freetz we had also different behaviour when serial was connected and when not
<fda>
i had attache an serial die fritz1200 at weekend to fix the DTS. could i also use this cable? its form an very old siemsnd s45 cell phone
<dangole>
fda: yes, exactly. 3.3V TTL level USB-to-serial. the old siemens cables work perfectly, just do NOT connect the VCC (!!!) but only RX, TX and GND
<fda>
hm, if i could remember the wirdes ... i buld it for avm7490 as the device was new ^^
<fda>
btw the ubi* bins in mtd-tools are build without-xattrs, but selinux is not set at all
<dangole>
ubifs-bins with or without xattr: that doesn't matter, we don't use those at all in normal run-time. format of ubifs happens implicitly on first mount of an empty volume
<fda>
could it be i formated ubi wrong? i used the initram-installer image from github to boot the device. then i flashed my sysupgrade...
<dangole>
fda: for 99% of things, SELinux should always be switched off even if we are building an image with SELinux-enabled. this is because we are NOT using the normal reference policy and anyway do NOT allow any runtime configuration of SELinux what-so-ever. it's more like SELinux on Android and requires only very little awareness from useland.
<dangole>
fda: no, if ubiformat was wrong, you would not see mounted filesystems, the bootloader would not be able to even load the kernel
<fda>
and as i extraced 14mtds with 150mb is also odd
<dangole>
fda: it's because they overlap. ie. there is "firmware" and then again "kernel", "rootfs", "rootfs_data" which is the same area on flash as "firmware", but split into parts
<dangole>
fda: ie. 14 mtds with weird overlaps (and gaps! even worse!) is just the vendor layout I was so desperate to get rid off
<fda>
with the default layout, was the device to switch boot partition on a new flash?
<fda>
maybe its a good thing
<dangole>
fda: i believe recovery vs. production dual boot is a MUCH better method. because it really prevents you from bricking, no matter how much you try. A/B dual-boot prevents you from bricking *once*. if you try again, that's it then, you destroyed both A and B parts.
<dangole>
fda: and that's how I implemented it in the bootloader of the UBI variant. just hold the reset button during boot and always end up in initramfs, and you can NEVER destroy it using sysupgrade, and you easily know that you are now in recovery and something went wrong (but yet, trying again won't ever brick the device(
<fda>
oh i forgot it. this small recovery partition is nice. but in webif it should be clear that this is booted! i checked the backup/restore backe if there is a "factory reset" button to see if it worked
<dangole>
fda: btw. now that i'm playing more with your image i notice that it take A LONG time for the device to become responsive on it's LAN IP. maybe switch doesn't learn or something, it's weird a bit. but a minute later it works then...
<dangole>
fda: did you modify kernel_menuconfig for anything in that image?
<dangole>
hda: (and i'm not into all that web stuff at all, i do stuff up to JSON RPC level and all above is not really my cup of tea)
<fda>
i thought this way: on header with hostname + "OpenWrt" there could be an "(Recovery)" after the openwrt label
<fda>
i changed the governor what "autofixed" the kernel .config
<fda>
if i have set openwrt-github envoronment i want to do some PR
<fda>
im very new to openwrt, and i noticed there are so many small bugs in luci! eg xinetd package has no ipv6 support at all! -> by webic created services listen always only ipv4, and the allow/block input allows no ipv6
<fda>
or samba page shows the big textfield only at the right 50% of the dispaly
<fda>
and xinetd packages does not allow the create internla-services like time+daytime
<dangole>
fda: ok, that explains a lot. because we recently figured '30 mhz' is a typo and should be '300 mhz'
<fda>
^^
<fda>
damn :)
<dangole>
fda: so if you change cpufreq to on-demand, it actually starts to use 30 mhz, which surprisingly even works a bit, but not for everything in the SoC...
<fda>
this i the problem? im using it since a week witih non-ubi
Tapper has quit [Ping timeout: 480 seconds]
<fda>
whats about the symbols in the .config which are deleted by defconfig?
<fda>
dangole: i just started a new build without the 2 patches, well see
<dangole>
fda: step by step. first clean build like we do, so we see your buildhost isn't the cause. then one change at a time...
<dangole>
fda: btw: please post the patch to disable xattr on squashfs if selinux isn't enabled, it may help other people as well
<fda>
done by just run "kernel_menuconfig" and save
<fda>
is there a way to speed up openwrt make? with freetz i could build a complete image within 3 minutes, openwrt need 30. if enabled ccache in both
Rentong has quit [Ping timeout: 480 seconds]
<fda>
with freetz could be selected if pre- or self-built tools and/or toolchain should be used
<fda>
dangole: with freetz the tools need to build ~5 minuten and the toolchain about 2min0: 30-20-5=5 ^^
<fda>
toolchain 20min
<dangole>
OpenWrt caches tools and toolchain by default. 30 minutes is a lot for a **rebuild**.
<fda>
complete builds each!
<dangole>
and if you just want to re-assemble binaries, there is the OpenWrt ImageBuilder for that
<dangole>
fda: what you mean by "complete"? you delete build_dir and staging_dir and all that? make distclean?
<fda>
delete complete dir
<dangole>
fda: why would you do that?
<fda>
do a new git-clone
<fda>
because i dont know the "clean" targets and what they do
<dangole>
that's too much. you really don't need to do that. do 'make clean' is the most you should need in normal development (unless you start patching the compiler)
<fda>
is "make target/linux/clean V=s" okay after i change kernel config?
<fda>
dangole: or what clean i should use after the changes of include/image.mk and package/utils/mtd-utils/Makefile ?
<fda>
imagebuilder is not what i want because "It downloads pre-compiled packages". i do recompile because i change packages :)
<dangole>
fda: "make target/linux/clean" is enough when changing kernel options. for package changes, make package/mtd-utils/clean will do the trick. for image.mk changes you don't need to clean at all.
<dangole>
fda: you can use the SDK to compile only packages you actually changed and use the binaries for everything else. that avoids diverging from official builds without wanting to (and all the hell of debugging that then)
<dangole>
fda: i mean the OpenWrt-SDK-... for that target/subtarget. not the buildroot as in `git clone ...`
<dangole>
fda: but to send patches, of course, they need to go on top of the buildroot, so you did right for development perspective. just in production I'd avoid to build from source, especially if the buildhost wasn't thoroughly verified to behave
<fda|>
but with non-ubi on-demand worked well for week
<fda|>
so for summer i should change back....
<dangole>
fda: or just set the 300 Mhz instead of 30 Mhz, then also on-demand should work. "should work" as in a theory that you will have to verify ;)
danitool has quit [Read error: Connection reset by peer]
danitool has joined #openwrt-devel
<rsalvaterra>
dangole: I just noticed, cpuidle is disabled in MT7622. It should be supported, no?
Tapper has joined #openwrt-devel
<dangole>
rsalvaterra: i think we still lack support in TF-A for it, it's intended to be there, Linux has support to handle it in that way, but TF-A lacks code to do that on MT7622
<rsalvaterra>
Oh, it's a TF feature? I had no idea. :/
<fda|>
@rsalvaterra CONFIG_CPU_IDLE ?
<fda|>
@dangole where could i change 30mhz?
<rsalvaterra>
fda|: Yes. If supported, I'd recommend the TEO governor.
<fda|>
rsalvaterra: teo? there is none with a T
<rsalvaterra>
fda|: What kernel version? 5.4?
<dangole>
rsalvaterra: i'm not sure it HAS to happen in TF, but TF and Linux have hooks to do it in that way.
<dangole>
fda: patch looks like i thought it should be, just testing is needed and that's the main part...
<dangole>
fda: the SDK is a tarball, so yes, you need to unpack it before you use it. but for the changes you do now, it's useless. it only helps you to build packages which match existing binaries which were built with that SDK
<dangole>
that also matches with my observation that LAN IP takes ages to be available and I blamed it on switch learning issues: turns out the switch just memorized the previous MAC correctly and that mtd-mac-address -> nvmem change caused problems...
<dangole>
fda: ah i see, you already noticed that as well
Rentong has joined #openwrt-devel
Rentong has quit [Ping timeout: 480 seconds]
<rsalvaterra>
Hm… is there an easy way (a script, or something) to factor out common subtarget kconfig symbols to the parent target kconfig?
* rsalvaterra
thinks ramips is getting a bit messy in that regard…
<fda>
seems my switch has no problems wirth different mac, the device reboots very quickly. but i have an very old and unmanaged gbit switch, luckyli with vlan forwarding
<fda>
but mac does not matter for me, i've static macs for wan and every vlan, only lan has original. if i have same mac for vlans and another device is also multihome in more vlans it causes sometime problems..
<fda>
another thing: i have a crappy german "hybrid" internet connection. Packets marked by client devices with DSCP/DiffServ are handled with the lower latency connection. With Openwrt it seems this flag is removed, fritzbox did not change them
<fda>
and german telekom voip does only work for by copper, so without dscp only normal sip provider work
<fda>
is default of linux behaviour to keep or remove dscp on routing?
<fda>
im setting it this way: iptables-t mangle -A OUTPUT -m owner --uid-owner asterisk -j DSCP --set-dscp 1
<rsalvaterra>
fda: You should take a look at the sqm-scripts package and the cake documentation. :)
<fda>
rsalvaterra: sqm? i have a "hybrid" internet connection, so the bandwith gos form 5-150 mbit ^^
Rentong has joined #openwrt-devel
Rentong has quit [Remote host closed the connection]
<rsalvaterra>
fda: Maybe autorate-ingress could help?
Rentong has joined #openwrt-devel
Rentong has quit [Remote host closed the connection]
<fda>
the "speedport hybrid" has dsl+lte, and everything marked wirh DSCP is routed only by dsl. this is requeired for voip
<fda>
the device automatically 1st used dsl and if more bandwith is needed (and if availabe, good weather, not much peaople around..) uses additionl lte. there is not much configurable
<fda>
this thing even can not do PD (even it got a /48) or accept ping from wan (so no hurrican ipv6 tunnel..)
<rsalvaterra>
Ouch… nasty.
<fda>
good idea, but not so good done
<fda>
so the dscp mark my client set should just be keep and forwarde on wan
<fda>
and so sadly every traffic shaping is unusable
<fda>
i've tried to disable "Software flow offloading" but it did not help. (i have then another pubic ipv4)
Rentong has joined #openwrt-devel
rsalvaterra_ has joined #openwrt-devel
Rentong has quit [Ping timeout: 480 seconds]
<fda>
of course the hybrid also blocks every "input" to ipv6 (GUAs) in the /64 on lan interface
rsalvaterra has quit [Ping timeout: 480 seconds]
<fda>
maybe i need to "mark" with openwrt on "input" dscp packages and set dscp again on "output" if marked
jordbaer_ has joined #openwrt-devel
jordbaer has quit [Ping timeout: 480 seconds]
rsalvaterra_ has quit []
rsalvaterra has joined #openwrt-devel
jordbaer_ has quit [Quit: jordbaer_]
dorf has joined #openwrt-devel
fda has quit [Remote host closed the connection]
fda has joined #openwrt-devel
rejoicetreat has quit [Ping timeout: 480 seconds]
philipp64|work has quit [Quit: philipp64|work]
philipp64|work has joined #openwrt-devel
<fda>
there is a new bug in dnsmasq: "Additional Hosts files" could be set by luci but syslog shows "failed to load names from .... Permission denied"
<dangole>
fda: dscp should be untouched for packets forwarded by Linux, but in this mt7622 hardware there is lots of offloading and maybe something goes wrong there.
<dangole>
fda: for asterisk, I setup tos/cos marking in the channel drivers (iax2, pjsip) and no need to mangle with iptables
<fda>
i tried to add logging, but neiter match noch target support is available in openwrt. so i need to add the modules first
<dangole>
fda: for dnsmasq with ujail: yes, this will need added logic in the init script to also add needed paths to jail based on configuration
<fda>
@dangole i have to look. maybe this works per connection?
<fda>
but i use DSCP not only for asterisk but also for some other things, even windows hosts
<dangole>
fda: dscp should be flow-based, which is why i'm suspecting offloading to be involved if it mysteriously gets lost or set to 0x0 for no reason
<fda>
i disable offloading in luci to test.is that enough?
<fda>
in case yes: does not work
<dangole>
fda: for static dscp assignemt, i use fwmark in mangle table and then apply the dscp using tc.
<dangole>
fda: and yes, disabling flow offloading should then also make DSCP fields pass transparently in case they were previously gobbled by flow offloading in hardware...
<dangole>
fda: i've just dealt with that a lot only a few weeks ago, installing new voip system on the sea watch 3, and i've watched all these DSCP markings in tcpdump... even had Toke help me with eBPF to copy DSCP from tunneled traffic to the outside packet :)
<dangole>
fda: and didn't see any markings getting lost ever unless that was the expected behavior. but there it's all x86 boxes (PCEngines and such)
<fda>
its not a bug of dnsmasq! i did a backuo-restore. in the tar all dirs had 755. But on the device 4 of 14 dirs had only 700
<Pepes>
Can someone bump base-files in OpenWrt 19.07 branch as someone forgot to do it or should I create a PR?
Acinonyx_ has joined #openwrt-devel
Acinonyx has quit [Ping timeout: 480 seconds]
fda has quit [Read error: Connection reset by peer]
fda has joined #openwrt-devel
fda has quit [Read error: Connection reset by peer]
fda has joined #openwrt-devel
<fda>
@dangole funny thing: dscp works. why? i dont know! i just rebootet everything. maybe switch, maybe changing mac, maybe the clients, maybe something other. windows asks on every new mac after every reboot if the network is public/private
<fda>
and without ondeman gov my temp rised from 52°C to 60°C. phy sensors are also a lottle higher
shibboleth has quit [Quit: shibboleth]
kabel has quit [Ping timeout: 480 seconds]
kabel has joined #openwrt-devel
fda has quit [Ping timeout: 480 seconds]
fda has joined #openwrt-devel
nlowe has joined #openwrt-devel
kabel has quit [Ping timeout: 480 seconds]
kabel has joined #openwrt-devel
Rentong has joined #openwrt-devel
Rentong has quit [Remote host closed the connection]
Rentong has joined #openwrt-devel
philipp64 has quit [Quit: philipp64]
Rentong has quit [Remote host closed the connection]