Re: [PATCH 2/3] locking: ww_mutex: Allow to use rt_mutex instead of mutex for the baselock

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Mar 09, 2015 at 02:21:53PM +0100, Sebastian Andrzej Siewior wrote:
> On 03/09/2015 12:29 PM, Mike Galbraith wrote:
> > On Mon, 2015-03-09 at 12:07 +0100, Sebastian Andrzej Siewior wrote:
> >> On 03/09/2015 11:51 AM, Mike Galbraith wrote:
> >>> Why do both mutex and rtmutex then exist one might ask? ;-)  No big deal
> >>> either way though, it's not like it becomes immutable once applied.
> >>
> >> You don't choose rtmutex afaik. rtmutex is used by futex (only?)
> > 
> > Almost only, but not quite.
> > 
> > drivers/media/usb/em28xx/em28xx.h:      struct rt_mutex i2c_bus_lock;
> > include/linux/i2c.h:    struct rt_mutex bus_lock;
> > kernel/rcu/tree.h:      struct rt_mutex boost_mtx;
> 
> So you have two users here: RCU and i2c-bus. The RCU thingy came from
> -RT (I think).

RCU uses rt_mutex rather than mutex because it needs rt_mutex's
priority-inheritance features for RCU priority boosting.  This did
indeed originally come from -rt, but is now in mainline.

If mutex gains rt_mutex's priority-boosting capability, then RCU
could switch to mutex.

But you both probably knew all that already...

							Thanx, Paul

> and I2C is actually one user. I am not sure Mauro used the rt-mutex in
> em28xx for a reason or just blindly copied the i2c code. The i2c-core
> holds a bus_lock so I don't think he needs to do the same thing.
> And i2c is the other user. It does a try_lock() in "irq_disabled()"
> context which records the wrong owner for PI-boosting if it is used
> from IRQ context.
> 
> I'm not against it but from ww-mutex point it makes most likely sense
> to switch them all and not decide which one to switch.
> 
> Sebastian
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>