References: <b9bd16cb-f1fa-34b7-d599-8637cbe5032b@free.fr>
User-agent: mu4e 1.8.13; emacs 28.2
From:   Jerome Brunet <jbrunet@baylibre.com>
To:     Marc Gonzalez <marc.w.gonzalez@free.fr>,
        linux-wireless@vger.kernel.org, MMC <linux-mmc@vger.kernel.org>,
        AML <linux-amlogic@lists.infradead.org>
Cc:     Neil Armstrong <neil.armstrong@linaro.org>,
        Arend van Spriel <aspriel@gmail.com>,
        Franky Lin <franky.lin@broadcom.com>,
        Hante Meuleman <hante.meuleman@broadcom.com>,
        Kalle Valo <kvalo@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        Ulf Hansson <ulf.hansson@linaro.org>,
        Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>,
        Alexander Prutskov <alep@cypress.com>,
        Joseph chuang <jiac@cypress.com>,
        Kevin Hilman <khilman@baylibre.com>,
        Martin Blumenstingl <martin.blumenstingl@gmail.com>,
        Angus Ainslie <angus@akkea.ca>,
        Heiner Kallweit <hkallweit1@gmail.com>,
        Pierre-Hugues Husson <phh@phh.me>
Subject: Re: [Performance regression] BCM4359/9 on S905X2
Date:   Tue, 21 Mar 2023 11:46:57 +0100
In-reply-to: <b9bd16cb-f1fa-34b7-d599-8637cbe5032b@free.fr>
Message-ID: <1jh6ue74x9.fsf@starbuckisacylon.baylibre.com>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk


On Tue 21 Mar 2023 at 11:40, Marc Gonzalez <marc.w.gonzalez@free.fr> 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