Re: [PATCH 1/1] i2c: Prevent priority inversion on top of bus lock

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

On Wed, 16 Sep 2009 15:35:24 +0300, Mika Kuoppala wrote:
> Hi,
> 
> On Wed, 2009-09-16 at 13:51 +0200, ext Jean Delvare wrote:
> 
> > >  		if (in_atomic() || irqs_disabled()) {
> > > -			ret = mutex_trylock(&adap->bus_lock);
> > > +			ret = rt_mutex_trylock(&adap->bus_lock);
> > >  			if (!ret)
> > >  				/* I2C activity is ongoing. */
> > >  				return -EAGAIN;
> > >  		} else {
> > > -			mutex_lock_nested(&adap->bus_lock, adap->level);
> > > +			rt_mutex_lock(&adap->bus_lock);
> > 
> > Hmm, rt_mutex doesn't handle locking model validation as regular
> > mutexes have? Looks like a regression to me.
> > 
> 
> There seems to be no rtmutex_lock_nested in include/linux/rtmutex.h.
> 
> CONFIG_DEBUG_RT_MUTEXES help says this:
>   This allows rt mutex semantics violations and rt mutex related
>   deadlocks (lockups) to be detected and reported automatically.
> 
> I don't know if this includes the same stuff as lockdep does.
> 
> But as the context is stripped away (adap-level) it's safe to assume
> that atleast some functionality is also stripped. Thus it is a
> regression.

OK. This is the only place where the adapter's nesting level is used,
so if we apply your patch, we might as well get rid of it.

-- 
Jean Delvare