Re: [RCU] rcu_tasks_trace_qs(): trc_reader_special.b.need_qs value incorrect likely()?

[Paul E. McKenney]

On Tue, Dec 05, 2023 at 03:59:27PM -0500, Steven Rostedt wrote:
> On Sat, 2 Dec 2023 18:45:07 -0500
> Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> 
> > On Sat, 2 Dec 2023 14:24:26 -0800
> > "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> > 
> > > > Note, the unlikely tracing is running on my production server v6.6.3.
> > > > 
> > > > The above trace is from my test box with latest Linus's tree.    
> > > 
> > > Nice tool!!!  
> > 
> > Thanks! It's only been in the kernel since 2008 ;-)
> > 
> >   1f0d69a9fc815 ("tracing: profile likely and unlikely annotations")
> > 
> > > 
> > > My kneejerk reaction is that that condition is suboptimal.  Does the 
> > > (untested) patch below help things?  
> > 
> > I'll give it a try on Monday.
> > 
> 
> This looks to have caused a difference. Although there's other RCU
> functions that need dealing with, but that's for when I have time to
> analyze all the places that have bad annotations.
> 
> 
> Anyway:
> 
>  correct incorrect  %        Function                  File              Line
>  ------- ---------  -        --------                  ----              ----
>  [..]
>    17924        0   0 rcu_softirq_qs                 tree.c               247
> 
> Tested-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>

Thank you very much, and I will apply this on my next rebase.

One of the disadvantages of userspace-free rcutorture testing...

							Thanx, Paul