Adding OpenWrt support for Xiaomi AX3600

To compile it I needed to at least get this patch and it's companion. Even so my wifi upload performance is still bad.

And how are you testing the WLAN performance?

iperf3 -s on device and iperf3 -c and iperf -R -c on my laptop and phone

Can you try multiple parallel streams?
For me its also bad with one stream, but with multiple streams I can easily get really good speeds.
Even speedstest to my ISP over WLAN achieves similar to those syntethic ones.

for some reason i notice very bad perf when doing speedtest from one device to another... if the router is the iperf server all is good... don't know if it's only me

Ok, with paralel streams the speed improves dramatically:

[SUM]   0.00-10.00  sec   464 MBytes   389 Mbits/sec  33660             sender

Though I guess the huge number of retries is still quite high.

Are you using NSS offloading?
Running iperf on the router itself without offloading is gonna get you massive number of retries

I guess NSS is enabled, I see some NSS logs in kernel log, but I guess wifi offload is not enabled. So that would explain it.

I dont think it is enabled, have you selected kmod-qca-nss-drv and kmod-qca-nss-ecm?

WIFI is not offloaded, but NAT and PPoE are and that gets rid of almost all retries which happen due to IRQ-s being bound to one core only for the ethernet.

@Ansuel Hauke pushed the updated backports,they have moved to 5.10.42 so it allows us to get rid of the thermal temperature patch and brings in some minor improvements.

Both are built and loaded

lsmod | grep -i nss
qca_nss_dp             45056  1 qca_nss_drv
qca_nss_drv           499712  1 ecm
qca_ssdk             1658880  2 qca_nss_drv,qca_nss_dp

Can you give us the whole bootlog?

@Ansuel Found that we were missing pyhsdev iptables extension, ECM was depending on the kernel
module for that twice instead of kernel module and the extension itself.
Pushed a fix to that.

Of course, here it is

Hmm, it looks correct.

I can confirm the high number of retries over WLAN while over ethernet there are no retries.
Speed via WLAN is good, 550+Mbits on my T490 which is 2x2 AC only.

Yep, it's the same for me

The thing is that I have no idea kind of retries is normal using WLAN, there is gotta be a lot more that wired for sure.

Well in the download direction there are no retries, so while I don't expect no retries at all I think it should be quite low.

Btw, what about the build error? It doesn't look to me that it can compile without those 2 patches, and even so I'm not sure other patches are needed for the functionality.

1 Like

No idea, I cannot reproduce the build error at all.

some time ago i had that error but i was probably working with a dirty compiled kernel... e make target/linux/clean fixed it... No idea either

I even rebuilt everything after doing make distclean to try and reproduce the issue.
With out without NSS stuff I cannot get that error.