[RFC] Transactional CGroup task attachment

Serge E. Hallyn serue at us.ibm.com
Fri Jul 11 08:34:21 PDT 2008


Quoting Paul Menage (menage at google.com):
> On Fri, Jul 11, 2008 at 7:27 AM, Serge E. Hallyn <serue at us.ibm.com> wrote:
> >
> > It does feel like it may be too much designed for one particular user
> > (i.e. is there a reason not to expect a future cgroup to need a check
> > under a spinlock before a check under a mutex - say an i_sem - in the
> > can_attach sequence?),
> 
> It would be fine as long as the code didn't want to *keep* holding the
> spinlock after the first check, while taking the mutex - and since
> that style of code is invalid under the existing locking rules, I
> don't see that as a problem. There's nothing to stop a
> prepare_attach_sleep() method from taking a spinlock as long as it
> releases it before it returns.
> 
> Paul

Good point.  For some stupid reason i was thinking don't take a spinlock
at all.

Have you started an implementation?

thanks,
-serge


More information about the Containers mailing list