Message-ID: <b9bd16cb-f1fa-34b7-d599-8637cbe5032b@free.fr>
Date:   Tue, 21 Mar 2023 11:40:40 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Content-Language: en-US
To:     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>,
        Jerome Brunet <jbrunet@baylibre.com>,
        Martin Blumenstingl <martin.blumenstingl@gmail.com>,
        Angus Ainslie <angus@akkea.ca>,
        Heiner Kallweit <hkallweit1@gmail.com>,
        Pierre-Hugues Husson <phh@phh.me>
From:   Marc Gonzalez <marc.w.gonzalez@free.fr>
Subject: [Performance regression] BCM4359/9 on S905X2
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

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)

# 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)
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