[PATCH v2 00/13] i2c: add and start using i2c_adapter-specific printk helpers

[Bartosz Golaszewski]

On Fri, Feb 27, 2026 at 11:06 AM Johan Hovold <johan at kernel.org> wrote:
>
> On Fri, Feb 27, 2026 at 10:08:34AM +0100, Wolfram Sang wrote:
>
> > > Wolfram, I noticed you merged these last night. Please think again and
> > > let's discuss the end result here. There's no question that there are
> > > lifetime issues in i2c, but this is not the way to solve it.
> >
> > I did think again and do not see a way how the life cycle problems can
> > be solved while drivers happily access the device struct of the adapter.
>
> There's nothing special about the struct device. What matters is that
> drivers don't free memory that's still in use by the core.
>
> > Whatever the solution to the core problem is (revocable, custom SRCU,
> > something else), I still think this step is needed in any case. If I am
> > wrong with this opinion, please enlighten me. Pointer to some existing
> > thread is OK, too. I didn't have the bandwidth to read the revocable
> > mail threads.
>
> It's not even about revocable or SRCU, that's just an implementation
> detail.
>
> It seems all that is needed is to decouple the struct i2c_adapter from
> the driver data and have core manage the lifetime of the former using
> the reference count of the embedded struct device.
>

I feel like we've discussed it already under v1 or elsewhere.

This is a weird pattern you sometimes see where a driver allocates
something and passes the ownership to the subsystem.  This often
causes confusion among driver authors, who logically assume that if
you allocate something, you are responsible for freeing it. Since this
is C and not Rust (where such things are tracked by the compiler), I
strongly believe we should strive to keep ownership consistent: the
driver should free resources it allocated within the bounds of the
lifetime of the device it controls. The subsystem should manage the
data it allocated - in this case the i2c adapter struct device.

I know there are a lot of places where this is done in the kernel but
let's not introduce new ones. This is a bad pattern.

But even if you decided this is the way to go, I fail to see how it
would be easier than what I'm trying to do. You would have to modify
*all* I2C bus drivers as opposed to only modifying those that access
the underlying struct device. Or am I missing something?

Bartosz

> Then you can use an rwsem, SRCU, revocable or something else to handle
> devices going away while they are in use.
>