Current status and possible improvements in CONFIG_MODULE_FORCE_UNLOAD

Current status and possible improvements in CONFIG_MODULE_FORCE_UNLOAD

[Aditya Garg]

Hi

I am Aditya Garg. I often require using out of tree drivers to support various hardwares on Linux. Sometimes the provider doesn't write good drivers, and often they have to be force unloaded. It's a common thing in proprietary drivers. I know the author of the driver should take note of the issues, but still the force unloading of the modules does come in handy many times.

Unfortunately if CONFIG_MODULE_FORCE_UNLOAD is not enabled in your kernel, which most probably is not enabled if you are using a Distribution pre compiled kernel, you have to recompile the whole kernel again.

I want wondering if instead of a kernel config option, we could use a kernel parameter to enable/disable this feature, I believe it should act as a better alternative. After all there must be people like me who are forced to recompile the whole linux kernel just for the sake of getting a functionality.

I hope for a reply and suggestions

Regards
Aditya

Re: Current status and possible improvements in CONFIG_MODULE_FORCE_UNLOAD

[Christoph Hellwig]

On Thu, Jun 06, 2024 at 06:49:59AM +0000, Aditya Garg wrote:
> Hi
> 
> I am Aditya Garg. I often require using out of tree drivers to support various hardwares on Linux.

Just stop buying hardwarew that requires this, or improve and upstream
the drivers to make your life easier instead of making other peoples
life worse.

Re: Current status and possible improvements in CONFIG_MODULE_FORCE_UNLOAD

[Lucas De Marchi]

On Thu, Jun 06, 2024 at 06:49:59AM GMT, Aditya Garg wrote:
>Hi
>
>I am Aditya Garg. I often require using out of tree drivers to support various hardwares on Linux. Sometimes the provider doesn't write good drivers, and often they have to be force unloaded. It's a common thing in proprietary drivers. I know the author of the driver should take note of the issues, but still the force unloading of the modules does come in handy many times.
>
>Unfortunately if CONFIG_MODULE_FORCE_UNLOAD is not enabled in your kernel, which most probably is not enabled if you are using a Distribution pre compiled kernel, you have to recompile the whole kernel again.

CONFIG_MODULE_FORCE_UNLOAD only ever makes sense on a developer
environment loading/unloading multiple times his own .ko module. Then
the developer knows better the state of the driver and hw to judge if
it's safe to ignore krefs.

>
>I want wondering if instead of a kernel config option, we could use a kernel parameter to enable/disable this feature, I believe it should act as a better alternative. After all there must be people like me who are forced to recompile the whole linux kernel just for the sake of getting a functionality.

Just allowing it like this is not a good thing. You may have a all kind
of issues with use after free, dangling pointers etc.  That would just
make life harder for people not involved with proprietary modules.


>I hope for a reply and suggestions

I´d ask them to upstream their driver and start sending the issues to
their side.

Lucas De Marchi

>
>Regards
>Aditya

Re: Current status and possible improvements in CONFIG_MODULE_FORCE_UNLOAD

[Aditya Garg]

Thanks for the reply Lucas. It makes sense now!

> On 15 Jun 2024, at 12:18 AM, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
> 
> On Thu, Jun 06, 2024 at 06:49:59AM GMT, Aditya Garg wrote:
>> Hi
>> 
>> I am Aditya Garg. I often require using out of tree drivers to support various hardwares on Linux. Sometimes the provider doesn't write good drivers, and often they have to be force unloaded. It's a common thing in proprietary drivers. I know the author of the driver should take note of the issues, but still the force unloading of the modules does come in handy many times.
>> 
>> Unfortunately if CONFIG_MODULE_FORCE_UNLOAD is not enabled in your kernel, which most probably is not enabled if you are using a Distribution pre compiled kernel, you have to recompile the whole kernel again.
> 
> CONFIG_MODULE_FORCE_UNLOAD only ever makes sense on a developer
> environment loading/unloading multiple times his own .ko module. Then
> the developer knows better the state of the driver and hw to judge if
> it's safe to ignore krefs.
> 
>> 
>> I want wondering if instead of a kernel config option, we could use a kernel parameter to enable/disable this feature, I believe it should act as a better alternative. After all there must be people like me who are forced to recompile the whole linux kernel just for the sake of getting a functionality.
> 
> Just allowing it like this is not a good thing. You may have a all kind
> of issues with use after free, dangling pointers etc.  That would just
> make life harder for people not involved with proprietary modules.
> 
> 
>> I hope for a reply and suggestions
> 
> I´d ask them to upstream their driver and start sending the issues to
> their side.
> 
> Lucas De Marchi
> 
>> 
>> Regards
>> Aditya