Adding OpenWrt support for Xiaomi AX3600 (Part 1)

diego123 · 2022-01-24T17:07:13.488Z

Today I wanted to try something and by accident realised this box has no USB...

Is there internally a way to add USB?

sqrwv · 2022-01-24T17:12:56.575Z

I have 2 ax3600, wired by LAN port and configured as dumb APs.
Both have last robimarko prebuild image and never had that problem.
I have DHCP server with static leases in my tplink archer C7 main router and disabled DHCP server in both AX3600, using the ignore interface option in DHCP server under LAN interface.

I also disabled IPV6 and deleted IPV6 interface in both ax3600 and tplink, only because I decided not to use IPV6 since it came out, I plan to use it in future but don't need it for now.

Gingernut · 2022-01-24T17:35:17.911Z

Some wireless devices connect fine but sit there trying to obtain an IP, fail after a while and connect to a further AP.

This is on 5.15, on 5.10 it just took a few minutes sometimes.

Have you experienced the same thing?

xabolcs · 2022-01-24T20:04:40.791Z

Yes, I experienced, but with a very old build.

This is an interesting bug ... let's rewind back to 2021. november:

robimarko:

It wasn't even the backports fault, it's the stupid QCA switch driver using hardcoded DTS values for how to set up forwarding.
I was using one of the LAN ports in DHCP client mode for uplink but since its isolated with the LAN bitmap the WLAN interface didn't have an uplink and then Poco F2 would reconnect due to "No networking" every couple of seconds, now it works fine when using the dedicated WAN port

The other solution is porting pull request #20: "Add NSS bridge-mgr support" to your favourite branch.

Gingernut · 2022-01-24T20:27:29.359Z

Your not using the newer 5.15 branch I gather.

sqrwv · 2022-01-25T17:12:49.197Z

I build and flashed the 5.15 branch in AX3600 and could replicate your problem.

Then I made a static IP conf. instead of DHCP in the mobile phone.
The problem still persists, randomly as before, but now the phone says it has no internet connection, but stays connected to the wifi.

With static IP in the phone now, when I get the no internet problem, I can ping the 5.15 router from the phone and I can login to the lucy gui, but I can't see anything else in the wired LAN.
From the LAN, I can see the 5.15 router, but not the connected wifi phone.

Looks like the problem is not DHCP, but WIFI-LAN.
Restarting the LAN interface didn't help.

Other times, with the same configuration, everything works.

I don't have previous experience building openwrt, so I selected my ax3600 makemenu config options based in robimarko .config for ax9000 (posted here: OpenWrt support for Xiaomi AX9000 - #485 by robimarko). I need to compare the .config files again to see if I missed anything.

lokisp · 2022-01-25T19:01:47.688Z

I made an image from the Github of robimarko 5.15 for AX3600, and for my use it is very stable.It is a very simple version , I only have running the original dns , ddns , prometheus and wireguard. I have about 12 wifi clients , and several network devices , TV , PC and others.
I tend to give it a regular use , with video conferencing over wifi and netflix network and has not given me any connection failure or disconnect issue. It has enabled the 160Mhz channel and through pppoe I get the 500/500 that I have contracted with the ISP.
I have copied the configuration from rodriguezst's github and I have configured it to my preferences, so that it creates an image with the repo 5.15 and the packages I mentioned before.
here is the github , if someone wanna play with it : https://github.com/ns2switch/ipq807x-openwrt-builder/releases

this is what it looks like after 24 hours:

Captura de pantalla 2022-01-25 a las 19.50.521828×1062 162 KB

sqrwv · 2022-01-25T20:35:48.822Z

Thanks for sharing, I flashed it.
I still have the random connection problem in wifi that Gingernut asked about.

lokisp · 2022-01-25T20:42:55.529Z

Then it is a problem in any configuration file. I am glad to share my configuration if needed.

etzisim · 2022-01-25T21:10:18.769Z

I just had big problems with memory, when i had two SSIDs on the phy now i delete one and it's perfekt again. with my build ( https://cloud.etzi.at/s/2RJkdWNo3QFdSZn ) it's just the basic with luci and wireguard (nearly the same then @lokisp)

sqrwv · 2022-01-25T22:27:38.574Z

I went back to 5.10 for now.
Will try 5.15 again when I have a bit more time.

dspalu32 · 2022-01-26T10:48:47.461Z

5.15 is working well for me (ipq807x-5.15 a8f5769). Thanks again, as always.

One AX3600 working as router/DHCP server etc. with two slave AX3600s connected through ath11k AX WDS as AP-only repeaters. Defensive sysupgrades worked first time on all three: "wifi down; killall -9 wpad; sysupgrade blah.bin"

I had a bunch of kmod-nss-* selected in my diffconfig from the 5.10 build - naively trying offload before - but dropped back to just nss-dp due to the compile errors mentioned above.

There's no drastic change in memory usage vs. 5.10. My low-activity AX3600 still increases from 200 to 300Mb used over the course of an hour, and then immediately drops back to 200Mb when a cron-scheduled iperf3 burst is run.

I'm not seeing the DHCP issues reported above. Wireless clients of the slave APs can get DHCP leases from the master router/AP OK. Perhaps as I've got a WDS rather than cabled link?

Gingernut · 2022-01-26T11:06:27.594Z

Not sure.

Do the wireless clients that connect directly to your main ax3600 that is the DHCP server have any issues?

It seems to be problem in the switch driver.

dspalu32 · 2022-01-26T11:56:23.041Z

I don't know much, but can imagine an intelligent/accelerated switch interfering with IP broadcast and affecting DHCP.

All situations seem to have working DHCP for me: wired and wireless (only ath11k 2.4GHz tested) direct clients of the AX3600 acting as DHCP server, as well as wired/wireless (only ath11k 2.4GHz/5GHz tested) clients of the downstream AX3600 WDS repeaters.

The downstream WDS AX3600s have static IPs, but all clients are dynamic.

Unfortunately I can't easily test with a slave AX3600 cable-connected to that master.

I had vaguely similar symptoms in the past with bad inter-LAN comms and DHCP, but that was 100% fixed by an ARP patch specific to ath11k WDS (https://github.com/robimarko/openwrt/pull/16, ath11k-send-multicast-in-4addr.patch, later incorporated into AX3600-5.10-restart and is still present in ipq807x-5.15). I had to try a few permutations to get to the point of blaming ath11k: cabled, ath10k/ath11k master/slave/symmetric.

sqrwv · 2022-01-26T21:46:36.501Z

I've made some more tests today using lokisp 5.15 image.
I have the same problem I reported before with 5.15 wifi clients.
No problem with 5.10 builds.

My test setup:

ISP - [WAN](tplink archer c7 router)[LAN] - (Zyxel GS-1200-8 switch) -  [LAN](dumbAP_AX3600#1-Kernel5.15)

a) DNS and DHCP servers are in tplink router, but I also tried without using DHCP server, putting a static IP configuration in the phone.
b) zyxel switch is configured as a dumb switch, I can't remove it today for this test.
b) My wireless devices have fixed MAC, because I use static leases in DHCP server.

The issue:
Just after I connect to tplink wifi (5G or 2.4G), I can never get IP from or connect to ax3600(5G or 2.4G). If I connect with static IP, I can't see my LAN, I can only see ax3600 LAN IP.
After I manage to connect to ax3600 one time, I can change connection to/from 5G and 2.4G and it always connects.

A workaround that (looks like) fixed the issue:
I configured my android mobile phone to have a random MAC address.
This way it always connects (it gets a dynamic IP from the dhcp).

So it has something to do with the way the LAN switches learn my phone MAC, depending where it was connected before. I don't know more to explain it better.

The same configuration with ax3600 5.10, with fixed MAC in phone never had this or any other issue. So this workaround is not a solution, but a way to try to understand the problem.

robimarko · 2022-01-26T21:58:45.499Z

@dchard Finally decided to stop trying to figure out the clock issues and focused on some general cleanup of the patches as well as adding the 512M profile.

Finally, my 5.15 kernel git tree and OpenWrt patches actually match, previously it was a big mess.
Also, moved to the new SMEM and TCSR bindings that match the patches that were sent upstream.

So, there should be more memory available now.

Also, I still have no idea WTF does AX3600 have all of the clocks registered properly while it doesn't work for others

dchard · 2022-01-26T22:36:27.912Z

Great news (not for the clock issue). Let me take a look.

Gingernut · 2022-01-27T06:19:15.684Z

Has anyone with this problem tested this small script that was posted on the x3600 roaming issue thread?
It goes in /etc/rc.local (local startup tab in LuCI)

#!/bin/ash
function flush() {
	ssdk_sh fdb entry flush 1
}

IFS=$'\n'
logread -f | while read line; do
	if [[ "$line" == *"AP-STA-"* ]]; then
		flush </dev/null >/dev/null 2>&1
	fi
done

dchard · 2022-01-27T09:34:19.005Z

After 10 hours, everything looks fine, extra memory is available again (407MB total), memory usage is hovering around 50%. PPPoE can get 900+Mbits with SW offload. By the way I can see a clear diff between wifi fw 1192 and 1201. The later has less uplink speed, but better DL speeds and less overall retransmissions.

robimarko · 2022-01-27T09:54:14.170Z

Great, glad to hear that