Re: No master_xfer_atomic for i2c-mv64xxx.c

[Andrew Lunn]

> > However, I'm not entirely sure how we could implement it without
> > sleeping. The controller is basically a state machine that triggers an
> > interrupt on each state change, so you first set the address, get an
> > interrupt, then set the direction, then you get an interrupt, etc.
> >
> > I guess we could implement it using polling, but I'm not sure if
> > that's wise in an interrupt context either.
> 
> I believe that is actually how some of the other drivers handle it,
> using polling. You can mask or disable the interrupts while in the
> xfer_atomic callback, and the i2c core won't schedule two transfers
> at the same time anyway.

The ocore driver is similar to the Marvell driver, a big state
machine. It implements polling for atomic transfers. It needs polling
support anyway, because some instantiations of the hardware have
broken interrupts :-(

Maybe there is some code which can be copied from the ocore driver?

      Andrew