Re: [RFC,PATCH] RCUify single-thread case of clock_gettime()

[Ingo Molnar <mingo@elte.hu>]

* Hugh Dickins <hugh@veritas.com> wrote:

> On Mon, 31 Oct 2005, Ingo Molnar wrote:
> > * Hugh Dickins <hugh@veritas.com> wrote:
> > > 
> > > Not my area at all, but this looks really dodgy to me, Paul:
> > > could you explain it further?
> > 
> > the patch below (included in the -rt tree) is the prerequisite. That's 
> > what Paul's "requires RCU on task_struct" comment refers to.
> 
> Thanks, Ingo: Sorry, Paul: I missed that it was an -rt patch: Ignore 
> me.

btw., since the RCU-task-struct thing is beneficial to upstream SMP 
kernels (even without any preempt option enabled), it should be 
considered for upstream too. The tasklist_lock is one of our last 
remaining monolithic and globally-bouncing locks. The patch i attached 
to the previous mail is against the upstream kernel and implements the 
RCU-task-struct logic, without any PREEMPT_RT dependency. (the -rt tree 
is a collection of various preemption related patches, not just 
PREEMPT_RT)

	Ingo