[Letux-kernel] gta04a5 rcu_preempt self-detected stall on CPU with latest kernel

[Andreas Kemnade]

Am Sun, 2 Feb 2025 21:21:12 +0100
schrieb "Ing. Josua Mayer" <josua.mayer at jm0.eu>:

> Am 02.02.25 um 21:05 schrieb Andreas Kemnade:
> > Am Sun, 2 Feb 2025 20:55:07 +0100
> > schrieb "Ing. Josua Mayer" <josua.mayer at jm0.eu>:
> >   
> >> Am 02.02.25 um 20:40 schrieb Andreas Kemnade:  
> >>> Am Sun, 2 Feb 2025 19:35:55 +0100
> >>> schrieb "Ing. Josua Mayer" <josua.mayer at jm0.eu>:
> >>>      
> >>>> Hello everybody,
> >>>>
> >>>> I attempted to boot latest letux-kernel (6.12) from microSD today and
> >>>> encountered rcu stall (see below  [1]).
> >>>>
> >>>> microSD created by:
> >>>> ./makesd -vk letux-6.12.9 gta04one
> >>>>
> >>>> also attempted "stable" (5.4) with similar error:
> >>>> ./makesd -vk stable gta04one
> >>>>
> >>>> Has anyone recently tested a specific version on gta04a5 onenand variant?
> >>>>     
> >>> we need to revert some i2c driver patch.
> >>>
> >>> This one: c770657bd2611b077ec1e7b1fe6aa92f249399bd or disable the
> >>> gyroscope driver.  
> >>
> >> Thanks!
> >> I will look into it.
> >>
> >> In the meantime I have bisected letux-kernel binaries and found the
> >> issue introduced with 5.4.243 (attached logs).  
> > 
> > yes, it soaked into stable kernels. There is one write to the gyroscope
> > causing irq storms and therefore all sort of havoc.   
> 
> I notice Nikolaus put many edge triggered irqs in device-tree.
> That is unusual, mostly because many gpio interrupt controllers
> don't support them properly.
> 
It is the i2c controller interrupt not getting handled properly in all
cases. No GPIO thing. So some NACK/clockstretching, whatever on sending
the reset command to the gyro. 

I will simply send a revert, if TI does not speak up on the LKML to
justify their patch, there is no other choice.

Regards,
Andreas

Regards,
Andreas