Performance lower than expected with BCM4359/9 on S905X2

[Marc Gonzalez]

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