> > So, finally, here is the second RFC for supporting I2C transfers in atomic
> > contexts (i.e. very late). This will need some text because I tried some things
> > on the way but had to discard them. However, I think it is important to have
> > that documented.
> 
> > Sorry, no TLDR; text here - I think this topic deserves a few words ;)
> 
> Thank you for this work! It was indeed interesting reading.

Thanks, glad to hear that :)

> And since your series is targetting some exiting use cases, I would
> drop as well academic variants of brain-damaged hw design, I think it
> worth to go.

Well, yes and no. I am with you that some complicated muxed setups could
be argued to be way over the top. However, with the panic
fault-injector, I can get my simple "PMIC directly (even exclusively)
attached to root adapter" setup to stall. By simply ignoring the lock,
such setups could work again. But this series does not implement this
because it would need a redesign, i.e. tie into i2c_transfer and not
later in __i2c_transfer.

Yet, before doing this, I was interested in discussion if ignoring the
lock is really desirable.

This series seems like a valid approach to me if we still want to
respect the locking.