Re: [PATCH 4/5] staging: vchiq_arm: Create keep-alive thread during probe

[Stefan Wahren <wahrenst@gmx.net>]

Hi Micha,

Am 24.08.25 um 20:37 schrieb MichaIng:
> > Okay, after disabling all VCHIQ users including VCHIQ_CDEV and 
> enabling DETECT_HUNG_TASK, i was able to reproduce it.
>
> > I will try to fix that.
>
> Hi Stefan,
>
> is there any progress in this regards, or a way we can help? Would it 
> make sense to share the issue to the RPi Ltd. downstream repository?
> https://github.com/raspberrypi/linux
this issue has been fixed in mainline/stable by reverting the commit in 
July. So downstream should have received the fix, too.

Best regards
>
> We ship images with downstream RPi Ltd. kernel packages, but with 
> non-GUI/server setups and low memory usage in mind, we 
> disable/blacklist camera and GPU related kernel modules by default, 
> and reduce the dedicated GPU memory to the minimal 16 MiB, so that our 
> users report as well these errors every few minutes on their main 
> console, until GUI, video codec, or camera features are enabled.
>
> I am thinking about letting one kernel module load by default which 
> releases the hanging task. But the majority of systems are headless, 
> and really don't require any of those. Reducing the console log level 
> or muting hung task timeout messages would be another workaround, but 
> that could mute important messages as well.
>
> Just in case someone else is in the same situation: I found the VC CMA 
> driver "vc_sm_cma" module to be the lightest way of releasing the 
> waiting task. Other than the camera/video modules, it does not pull in 
> a large stack of dependency modules. With less than 32 MiB GPU memory 
> (where the cut-down firmware is used automatically), it fails to 
> initialise, but releases the vchiq-keep thread regardless. Its size is 
> also tiny compared to the Broadcom camera/video drivers.
>