Re: [PATCH 1/2] i2c: add "reset" sysfs entry for adapters.

[Jean Delvare <khali-PUYAD+kWke1g9hUCZPvPmw@public.gmane.org>]

On Mon, 9 Feb 2009 16:02:55 +0100, Rodolfo Giometti wrote:
> On Mon, Feb 09, 2009 at 02:55:46PM +0000, Ben Dooks wrote:
> > On Fri, Feb 06, 2009 at 04:23:17PM +0100, Rodolfo Giometti wrote:
> > > It could happen that an i2c adapter may lock the bus due due
> > > electrical problems, so the user may recover this stale state by using:
> > > 
> > > 	$ echo 1 > /sys/class/i2c-adapter/i2c-0/reset
> > > 
> > > Signed-off-by: Rodolfo Giometti <giometti-k2GhghHVRtY@public.gmane.org>
> > 
> > There doesn't seem to be any locking to stop issuing a reset during
> > an extant transaction... should the controller driver deal with
> > cancelling any outstanding transactions?
> 
> I think this should be done inside each adapter driver since it's
> an adapter specific issue.
> 
> If you see next patch of this patchset you can see that the adapter
> i2c-pxa aborts any transfer currently under way during reset.

I disagree. An i2c bus driver should never get stuck in the middle of a
transaction (that is, with i2c_adapter->bus_lock held.) If a
transaction fails, the driver should timeout and return. Thus I think
Ben is right and it would make a lot of sense to have function
write_adapter_reset take i2c_adapter->bus_lock. This guarantees that
all bus drivers will get locking right and won't allow resetting in the
middle of a (non-stuck) transaction.

-- 
Jean Delvare