Re: [PATCH v2 1/3] i3c: Add support for mastership request to I3C subsystem

[Przemyslaw Gaj <pgaj@cadence.com>]

The 01/15/2019 22:09, Boris Brezillon wrote:
> EXTERNAL MAIL
> 
> 
> On Fri, 11 Jan 2019 17:43:35 +0000
> Przemyslaw Gaj <pgaj@cadence.com> wrote:
> 
> > +		i3c_bus_normaluse_unlock(i3cbus);
> > +	}
> > +
> >  	i3c_bus_normaluse_lock(i3cbus);
> >  	ret = sprintf(buf, "%d-%llx\n", i3cbus->id,
> >  		      i3cbus->cur_master->info.pid);
> > @@ -663,6 +728,12 @@ static int i3c_master_send_ccc_cmd_locked(struct i3c_master_controller *master,
> >  		    !rwsem_is_locked(&master->bus.lock)))
> >  		return -EINVAL;
> >  
> > +	if (!i3c_master_owns_bus(master)) {
> > +		ret = i3c_master_request_mastership(master);
> > +		if (ret)
> > +			return ret;
> > +	}
> 
> As I said above, I think bus ownership should be requested in
> maintenance mode, which means it has to be done before entering this
> function. You can probably start acquiring the lock in write (AKA
> maintenance) mode and downgrade it to read (AKA normal) mode before we
> start sending the frame (CCC, SDR, I2C or HDR). 
> 

So, I'll have to call all the _locked functions with maintenance lock.
Inside i3c_master_send_ccc_cmd_locked() I will downgrade the lock to read
(normal use). Isn't it inconsistent that after _locked functions I will
unlock the bus from normal use, even if I locked it for maintenance?

Also, I will have to downgrade the lock in case of failure in all _locked
functions to unlock the bus properly using i3c_bus_normaluse_lock(). DO you
think this looks good?

-- 
-- 
Przemyslaw Gaj

_______________________________________________
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c