Re: Robust futexes

[Rusty Russell <rusty@rustcorp.com.au>]

On Fri, 2006-02-17 at 08:23 -0800, Darren Hart wrote:
> Rusty Russell wrote:
> telling the kernel that the lock is the tid allows the
> > kernel to do prio inheritence etc. in future.
> 
> Priority Inheritance has come up a couple of times in relation to Ingo's new 
> LightWeight Robust Futexes.  Ingo has said that PI is orthogonal to LWRF, but I 
> don't think we've heard if there are plans already in the works (or in his head 
> :-) for PI.  Rusty's comment above reads as "the current LWRF implementation 
> cannot support PI" - is there something about it that makes PI impractical to 
> implement?

Hi Darren!

	Ingo's approach is indeed orthogonal.  But the obvious approach to PI
etc is to tell the kernel who is holding the lock, by making the lock
value == TID of the holder.  If we are heading towards this anyway, the
kernel could use this to implement robust mutexes, too, although not
with a 100% guarantee (due to tid wrap).  Ingo doesn't like that,
though.

Hope that clarifies!
Rusty.
-- 
 ccontrol: http://ozlabs.org/~rusty/ccontrol