On Tue 21 Mar 2023 at 11:40, Marc Gonzalez <marc.w.gonzalez@xxxxxxx> wrote: > Hello everyone, > > I've been benchmarking an Amlogic S905X2 board. > It provides a BCM4359/9 WiFi chip connected through SDIO. > > There's a large performance gap between vendor kernel and mainline. > (Downloading a 1GB file to /dev/null from a device inches away) The title is misleading. You are comparing different sources. This is not a regression. This is merely a difference. If we are talking about mainline, then which board is it ? What is the corresponding DT ? What is the MMC configuration in both case ? Have you checked you are running with the same clock configuration to begin with ? > > # curl -o /dev/null http://192.168.1.254:8095/fixed/1G > % Total % Received % Xferd Average Speed Time Time Time Current > Dload Upload Total Spent Left Speed > 100 1024M 100 1024M 0 0 27.5M 0 0:00:37 0:00:37 --:--:-- 28.6M > vs > 100 1024M 100 1024M 0 0 11.0M 0 0:01:32 0:01:32 --:--:-- 11.0M > > Line 1 = vendor kernel (4.9.180 amlogic android) This cannot help indentify a downstream kernel, and the infinite number of forks and patches associated with it. > Line 2 = mainline kernel (6.2.0-rc8) > > Why is the vendor kernel 2.5 times faster? > > (I'm using the same firmware files, but it seems the vendor kernel reads > an additional configuration file that the mainline vendor seems to ignore.) > > Regards
