Re: [PATCH net-next] i2c: tegra: runtime PM is not IRQ-safe

[Mikko Perttunen <mperttunen@nvidia.com>]

On Thursday, March 19, 2026 3:17 AM Russell King (Oracle) wrote:
> On Wed, Feb 18, 2026 at 08:30:49AM +0000, Jon Hunter wrote:
> > On 18/02/2026 01:35, Mikko Perttunen wrote:
> > 
> > ...
> > 
> > > > > > Yes we should always follow that rule. However, in this case, I
> > > > > > believe
> > > > > > that the build time dependency on the PINCTRL subsystem was only
> > > > > > exposed
> > > > > > by adding the 'i2c_dev->dev->pins'. Unless I am misunderstanding
> > > > > > ...
> > > > > 
> > > > > Yes, it looks like it.
> > > > > 
> > > > > However, I wonder why the dependency has to be complicated.
> > > > > 
> > > > > ARCH_TEGRA in both arm64 and arm selects PINCTRL, so we can assume
> > > > > that
> > > > > 
> > > > > PINCTRL will be set for ARCH_TEGRA. So:
> > > > >    config I2C_TEGRA
> > > > >    
> > > > >    	tristate "NVIDIA Tegra internal I2C controller"
> > > > >    	depends on ARCH_TEGRA || (COMPILE_TEST && (ARC || ARM || 
ARM64
> > > > >    	||
> > > 
> > > M68K
> > > 
> > > > >    	|| RISCV || SUPERH || SPARC))>
> > > > > 
> > > > > +	depends on PINCTRL
> > > > > 
> > > > > is a shorter way of writing this, and it makes sense - pinctrl isn't
> > > > > required because we're doing a compile test, it's required because
> > > > > the driver itself fundamentally requires it with this change whether
> > > > > or not we're doing a compile test.
> > > > 
> > > > Yes that's true indeed.
> > > > 
> > > > Mikko, do you want to take care of this?
> > > 
> > > My thought was it would be better to keep the PINCTRL dependency grouped
> > > with COMPILE_TEST. That makes it clear it's only needed because of it
> > > -- clearer to the reader that ARCH_TEGRA implies it. Kind of like not
> > > checking for NULL pointers in C code when the contract is that the
> > > pointer is not NULL.> 
> > Russell's point is that regardless of the compile test, the driver has a
> > dependency on pinctrl and so should always be dependent on it. The I2C
> > instances for the DPAUX device on certain devices require this and will
> > not
> > work without it (before your change was added). I guess I should have
> > added
> > this dependency back with commit 718917b9875f ("i2c: tegra: Add pinctrl
> > support").
> > 
> > > I can change it though if you'd like.
> > 
> > I think we should.
> > 
> > Thanks!
> 
> When is this bug going to be fixed? This is a regression that's
> affecting Xavier systems. It's been over a month since I proposed
> a patch to fix this:
> 
> BUG: sleeping function called from invalid context at
> kernel/locking/mutex.c:591 in_atomic(): 0, irqs_disabled(): 1, non_block:
> 0, pid: 59, name: kworker/u24:6
> 
> Please see:
> 
> https://lore.kernel.org/r/E1vsNBv-00000009nfA-27ZK@rmk-PC.armlinux.org.uk
> 
> for my original proposed fix, complete kernel messages and analysis.
> 
> --
> RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
> FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

FWIW I did post v3 of my patch a while back: https://lore.kernel.org/linux-tegra/20260303-i2c-dpaux-irqsafe-v3-1-75ca95b96666@nvidia.com/

Mikko