lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <2c56410f-2762-4b3c-b37e-e8db75d30560@app.fastmail.com>
Date:   Fri, 10 Nov 2023 15:37:04 +0100
From:   "Arnd Bergmann" <arnd@...db.de>
To:     Martin Tůma <tumic@...see.org>,
        "Arnd Bergmann" <arnd@...nel.org>,
        "Martin Tuma" <martin.tuma@...iteqautomotive.com>,
        "Mauro Carvalho Chehab" <mchehab@...nel.org>,
        "Hans Verkuil" <hverkuil-cisco@...all.nl>
Cc:     linux-media@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/2] media: pci: mgb4: remove bogus 'select' statements

On Wed, Nov 8, 2023, at 19:33, Martin Tůma wrote:
> On 08. 11. 23 17:13, Arnd Bergmann wrote:
>> On Fri, Oct 27, 2023, at 16:17, Martin Tůma wrote:
>>> On 23. 10. 23 18:05, Arnd Bergmann wrote:
>>>> From: Arnd Bergmann <arnd@...db.de>
>
> On SoCs you probably get a kernel configuration that is missing some 
> feature but still boots up when you do not select/depend on the exact 
> controller, but in the case of the mgb4 PCIe card you get a driver that 
> does not work at all (The SPI_XILINX dependency could theoretically be 
> made configurable, but you would lose the ability to flash the correct 
> FW for the current HW module and the access to the card's serial number. 
> I2C and XDMA are crucial.).

My point was that we do this all the time for things that are
essential: if your clock controller or the irqchip have
no driver, then the camera device won't work, but neither
would anything else.

So in a SoC setting, you really just need to enable all
the drivers for devices on that chip through the .config.

>> Since this is a PCI device, it's a bit different, so maybe
>> something like this would work to correctly document which
>> dependencies are required at build time vs run time:
>> 
>> --- a/drivers/media/pci/mgb4/Kconfig
>> +++ b/drivers/media/pci/mgb4/Kconfig
>> @@ -1,15 +1,13 @@
>>   # SPDX-License-Identifier: GPL-2.0-only
>>   config VIDEO_MGB4
>>          tristate "Digiteq Automotive MGB4 support"
>> -       depends on VIDEO_DEV && PCI && I2C && DMADEVICES && SPI && MTD && IIO
>> +       depends on VIDEO_DEV && PCI && I2C && SPI && MTD && IIO
>>          depends on COMMON_CLK
>> +       depends on XILINX_XDMA
>> +       depends on (I2C_XILINX && SPI_XILINX) || COMPILE_TEST
>>          select VIDEOBUF2_DMA_SG
>>          select IIO_BUFFER
>>          select IIO_TRIGGERED_BUFFER
>> -       select I2C_XILINX
>> -       select SPI_XILINX
>> -       select MTD_SPI_NOR
>> -       select XILINX_XDMA
>>          help
>>            This is a video4linux driver for Digiteq Automotive MGB4 grabber
>>            cards.
>> 
>
> My motivation when using "select" was to help people using "make 
> menuconfig" to get the module selected/configured as they will usually 
> not know that there are some Xilinx IP cores used that need separate 
> drivers and the menuconfig GUI simply hides the mgb4 option making it 
> almost impossible just from the menus to find out what has to be selected.
>
> But when there are reasons, why to chose "depends on" (like various 
> configurations, tests or the "readability" of the dependencies) than I'm 
> ok with your patch proposal.

The main reason to use 'depends on' over 'select' here is that
mixing the two is a common source of dependency loops that end
up breaking the build. As a rule of thumb, I would use 'select'
only for symbols that others already select, or that are hidden
from visibility.

      Arnd

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ