[Banana BPI-R4] Wifi7 status

For Developers
1005

Greetings to all. If somebody would like to test my first attempt to build BPI R4 wifi 7 AP with WO and MLO here is the SD card. This alpha version is not fully functional and is not fully compatible with LuCI wireless.

openwrt-mediatek-filogic-bananapi_bpi-r4-sdcard.img.gz

Here is sample of connected iPhone 16 CZ - (where is not allowed 320Mhz but only 160MHz). Maybe it could be a step ahead maybe not :slightly_smiling_face:

OpenWrt-MLO
OpenWrt-MLO1310×1248 121 KB

4 Likes
1006

can you share your repo or the source for this build ? i am curious on the mlo feature

2 Likes
1007

I've noticed with hardware flow enabled and wpad openssh, the wifi speed is much higher

1008

more to do with HW OFFLOAD ... what do you mean by WPAD OPENSSH ? openssl perhaps

1009

Oh right yeah openssl, the phone auto corrected it

1010

How do you turn it on?

1011

All is on openwrt Wiki

1012

are you sure the 2GB ram limit is correct ? no issues found with 4GB

1013

My build should work with options mt7996e wed_enable=Y sr_scene_detect=Y without memory limitation.

1014

Same here 4gb has no issues

1015

Can you share the GitHub repo

1016

If you are saying that no RAM limitation is needed anymore, I will make a test and edit wiki. Last time (week ago) when I was checking - RAM limitation was needed. Would be surprising for me, that it will start working.

Thanks for pointing me :slight_smile:

1017

Maybe I will do it later. I built it manually step by step with modified mtk git. I have no time to test mlo completely and as I wrote it is the first alpha version that should support
Single Wiphy Architecture
320 MHz bandwidth
4096-QAM MCS12, MCS13
WPA3 key management (AKM24)
MLO Basic Functionality (Advertisement/Discovery/Setup)

and doesn't support
Hardware Peak Performance
MLO Post-Setup Features

1018

Ah well we don't do MTK openwrt here, we do official openwrt, hence why you had MLO

1019

this may be related to the SWIOTLB_DYNAMIC that is not enabled by default on the kernel. I have enabled it on my build.

1 Like
1020

Hm, in my opinion my build is not MTK openwrt. I built it from official openwrt source and some sources from https://git01.mediatek.com/plugins/gitiles/openwrt/feeds/mtk-openwrt-feeds/ so what openwrt is it? :thinking: :slightly_smiling_face:

1021

Will do test today evening and update if all is fine.

1 Like
1022

I'd like to know why we need HW flow offloading when it comes to Wi-Fi performance. Is it purely because what we typically care about and benchmark is WAN<->(w)LAN speed? Or flipping HW flow offloading on also affects some other hardware acceleration features which are more closely related to Wi-Fi?

Edit. Speaking of flow offloading in general, I can confirm that on Rmandrad's latest build it works as it should without the need to set extra iptables/nftables rules. The rules are already there if flow offloading is turned on in /etc/config/firewall.

table inet fw4 {
	flowtable ft {
		hook ingress priority filter
		devices = { eth1, eth2, lan1, lan2, lan3, phy0.0-ap0, phy0.1-ap0, phy0.1-ap1, wan }
		flags offload
		counter
	}
	chain forward {
		type filter hook forward priority filter; policy drop;
		meta l4proto { tcp, udp } flow add @ft
	}

The difference in CPU load is dramatic between offloading and no offloading on a GbE uplink, so it's definitely working out of the box.

1 Like
1023

At the same time, I didn't have any crashes after I enabled options mt7996e wed_enable=Y sr_scene_detect=Y, but Wi-Fi has slowed down 3x (at least on the 5 GHz band). I wonder why. I didn't change anything in config-6.6. Using wpad-openssl.

1024

if you were using my build with wed_enable=Y the SWIOTLB_DYNAMIC kernel setting should take care of the memory so it doesn't crash.

there was someone suggesting on this channel to set sr_scene_detect to N some time ago ... did a quick test and haven't seen any noticeable impact... i will try again once i have some time

1 Like