Yeah, there seem to be lot's of stuff pending from the hostapd side, but from the kernel side I believe that everything is ready.
I just saw this which adds cca:
https://patchwork.ozlabs.org/project/hostap/list/?series=270829
Also, what do you think about the dynamic MU-EDCA?
Just two more things, I managed to make the router go suicidal with an iperf stream, twice (so it's reprocucible), and roaming between bands seems to be broken, I don't mean connectivity, I mean all clients I have once they connect to the 2.4 they stay there until I manually switch them, I believe this worked previously in my wrt3200acm. I don't know if any's worth checking with the state of things.
Yep, that's the series I mentioned, It has been pending for a while.
Are you sure that CCA is in the kernel, cause I have not seen patches for it?
MU-EDCA is gonna be a good way to increase speeds a bit more as well as its offloaded to FW which dynamically adjusts that, but its gonna take some time until its merged.
BTW, roaming between 2.4 and 5G is completely up to the clients, there is no "band steering".
Newer and smarter clients should all prefer 5GHz anyway automatically
If I notice something more I'll notify you, but do you think Openwrt upstream will accept those colouring patches while they're pending in hostapd?
I believe this is it:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v5.17-rc4&id=5f9404abdf2ac31c8f4768c39714bfcaca389e3a
About the MU-EDCA, do you think it's worth porting it to our patches while it's not merged upstream?
Yeah I know, with 802.11k/v they should do that by themselves but it doesn't work, yet in my ISP's router it does... As for the devices mostly are windows 11 but my main one is a OnePlus 8 with android 11, I don't believe we do smarter than this 
They might actually do since it looks really likely they will get merged, especially since you linked CCA in-kernel support.
Might actually try backporting CCA support and adding the pending patches for hostapd to see whether it works.
I don't think MU-EDCA is worth the trouble of porting now.
Its weird that they choose 2.4G, even my phone always like always picks 5G
1 Like
Yeah, I believe that everything is ready except for hostapd, and they seem to be very slow, not to mention the times OpenWrt actually updates hostapd which is very scarce.
Great! Can't wait to try it! 
Ok, just asking since it appeared to be 1 patch for hostapd and another for ath11k/mac80211
Ok, just to clarify, they connect to the 5ghz by default, but for example if you go to the shell and issue the "wifi" command and let the radios restart, all devices connect to 2.4 and stay there for eternity
OpenWrt actually updates Hostapd semi-regularly, as hostapd releases are so far apart that hashes are used instead.
Even when it's not updated backport patches are used.
Ok, that could easily be some kind of failsafe behaviour, who can tell with so many wpa_supplicant variants
True, you're right.
If you need anymore help with any of this just say so, I'm always watching the forum.
Yeah, I have no idea what to tell you, it's driving me crazy trying to figure out what's wrong...
About the failsafe feature, I believe I have found the issue, I believe it's because, when failsafe is triggered only stuff in the preinit folder run, and I believe the ethernet driver isn't being started when this happens, could that be it? And what can be added to the preinit folder to fix this?
This is an example, but I need to know what's the basic setup that needs to be done so that networking can be configured.
Do you have any idea?
1 Like
Are you guys sure that SSDK and the rest of QCA soup gets loaded when failsafe is triggered?
That seems like the most likely reason why it wouldn't work, as well as ifname not being set
I believe they aren't, that isn't standard failsafe initialization, OpenWrt expects the kernel to have all the required drivers, as modules shouldn't be loaded for obvious reasons.
Now, I can create a preinit file that can runs only in case the preinit was triggered and load the required stuff for networking, but I need to know what that is and how to load it ie. What commands to run inside the script. Is it just an insmod for the nss-dp and ssdk?
Just modprobe the qca-nss-dp it will load ssdk as dependency
1 Like
You quoted this some time ago in other thread (Roaming Issues Xiaomi AX3600 - #83 by psi-c) relating the FBD problem, maybe helps:
It pretty much sums up the issue, but their solution is to use one more NSS module which requires even more kernel modifications to solve it.
This is an issue when you are trying to pretend that the underlying switch doesn't exist, I am kind of tempted to see around hacking that without the bridge module.
There has got to be a way of registering to the notifications about new entry and just deleting the existing one
1 Like
I see, thanks!
I'll give it a try and report back as soon as I can.
One question, if I can't make a script that loads ONLY if the failsafe is triggered, is there an issue if I just load the modules in preinit?
I can do:
- Load module
- If the system isn't in failsafe unload the module
- Let the system proceed as usual (and load the module which is already loaded)
Is this a problem?
Since we are dealing with extremely fragile things they must be in order, that is why they are staged one after another.
So, if possible it would be best to not mess with loading anything unless failsafe is triggered
Got it!
I'll give it a try maybe tomorrow!
Awesome, nice to see community engagement
2 Likes
Well we have lift-off! I managed to get wired working in failsafe.
set_preinit_net_module() {
# Load the required networking modules for failsafe
/sbin/modprobe /lib/modules/*/qca-nss-dp.ko
}
boot_hook_add failsafe set_preinit_net_module
This works in failsafe and doesn't break regurlar boot (from what I can tell it really only runs in failsafe.
Would you like me to do a pull request to your branch?
The thing is, it only works on the first lan port, in this case, lan1 or eth1, is that ok?
1 Like
You don't need:
/sbin/modprobe /lib/modules/*/qca-nss-dp.ko
modprobe qca-nss-dp should work just fine.
This is what worked for me this morning:
. /lib/functions.sh
load_networking_drivers() {
modprobe qca-nss-dp
}
boot_hook_add failsafe load_networking_drivers
Failsafe interface can be set per device like IPQ40xx does it
From my testing we do, in the boot log errors get thrown out in the log if modprobe qca-nss-dp is used though I didn't try modprobe /lib/modules/*/qca-nss-dp.ko only /sbin/modprobe qca-nss-dp
Do we need to set a interface? by default it gets the first lan interface from the board.json, this should work for all devices
EDIT: I don't know how you managed to do it, it didn't work for me 
xyyx
5945
I can confirm no leaking memory. After 24 hours I have more than 120 MB free memory. Awesome!
