On 21/03/2023 11:46, Jerome Brunet wrote: > On Tue 21 Mar 2023 at 11:40, Marc Gonzalez wrote: > >> 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. I have changed the subject line to a (hopefully) more accurate one. > 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 ? Perhaps my initial request was unclear, my apologies. I was asking whether anyone using any S905X2 board had noticed such a performance discrepancy. (Neil seems to have provided a full explanation for the situation.) For what it's worth, I'm using an sei530 board, which appears to be a minor variant of the sei510 (down to the schematics). >> # 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 identify a downstream kernel, and the infinite number > of forks and patches associated with it. I was not asking anyone to reproduce my exact setup. I find it confusing that some assume there is something different about the board I'm using. The more likely hypothesis is that all S905X2 boards have the same behavior. Regards
