Re: [PATCH v9 4/7] rcu/trace: Add tracing for how segcb list changes

[Frederic Weisbecker <frederic@kernel.org>]

On Wed, Nov 04, 2020 at 06:08:07AM -0800, Paul E. McKenney wrote:
> On Tue, Nov 03, 2020 at 04:17:31PM +0100, Frederic Weisbecker wrote:
> > On Tue, Nov 03, 2020 at 09:26:00AM -0500, Joel Fernandes (Google) wrote:
> > > +/*
> > > + * Return how many CBs each segment along with their gp_seq values.
> > > + *
> > > + * This function is O(N) where N is the number of segments. Only used from
> > > + * tracing code which is usually disabled in production.
> > > + */
> > > +#ifdef CONFIG_RCU_TRACE
> > > +static void rcu_segcblist_countseq(struct rcu_segcblist *rsclp,
> > > +			 int cbcount[RCU_CBLIST_NSEGS],
> > > +			 unsigned long gpseq[RCU_CBLIST_NSEGS])
> > > +{
> > > +	int i;
> > > +
> > > +	for (i = 0; i < RCU_CBLIST_NSEGS; i++) {
> > > +		cbcount[i] = rcu_segcblist_get_seglen(rsclp, i);
> > > +		gpseq[i] = rsclp->gp_seq[i];
> > > +	}
> > > +}
> > > +
> > > +void __trace_rcu_segcb_stats(struct rcu_segcblist *rsclp, const char *context)
> > > +{
> > > +	int cbs[RCU_CBLIST_NSEGS];
> > > +	unsigned long gps[RCU_CBLIST_NSEGS];
> > > +
> > > +	if (!trace_rcu_segcb_stats_enabled())
> > > +		return;
> > 
> > Yes, very good!
> > 
> > Paul just told me that RCU_TRACE can be used in production so that confirms that we
> > wanted to avoid this loop of 8 iterations when tracing is disabled.
> 
> RCU's "don't try this in production" Kconfig option is PROVE_RCU.
> 
> I would be looking for checks that the sum of the segment lengths
> match the overall ->len or checks that all of the segment lengths
> are zero when ->cblist is empty to be guarded by something like
> IS_ENABLED(CONFIG_PROVE_RCU).  Of course, checks of this sort need to
> be confined to those portions of rcu_do_batch() that are excluding other
> accesses to ->cblist.

Right.

> 
> But if rcu_segcblist_countseq() is invoked only when a specific trace
> event is enabled, it should be OK to have it guarded only by RCU_TRACE.

Indeed, so I think we are good.

Thanks.