Re: [PATCH v7 1/6] sched: Annotate rq->rd with __rcu and update lockless readers

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Aug 27, 2026 at 05:53:58AM -0400, Aaron Tomlin wrote:
> On Thu, Aug 27, 2026 at 11:39:35AM +0200, Peter Zijlstra wrote:
> > On Thu, Aug 27, 2026 at 05:35:44AM -0400, Aaron Tomlin wrote:
> > 
> > > To solve this cleanly, I propose introducing a concise helper, for the root
> > > domain in kernel/sched/sched.h mirroring rcu_dereference_sched_domain(p):
> > > 
> > >     #define rcu_dereference_root_domain(p) \
> > >         rcu_dereference_all_check((p), lockdep_is_held(&sched_domains_mutex))
> > > 
> > > This covers both the update side (under sched_domains_mutex) and read-side
> > > contexts (under rq_lock(), preemption-disabled, or RCU read locks) via
> > > rcu_read_lock_any_held(), while keeping the call sites clean and
> > > eliminating Sparse warnings.
> > > 
> > > Would you prefer rcu_dereference_root_domain(p)?
> > 
> > Yeah, but I'm not sure I see the point of adding it, as it is exactly
> > the same as rcu_dereference_sched_domain(). The root domains are part of
> > the sched domains, so it doesn't seem wrong to me to use that existing
> > one.
> 
> Indeed. However, the motivation is purely code clarity. Defining
> rcu_dereference_root_domain() preserves the natural symmetry of rq->rd and
> rq->sd in struct rq, while keeping grep-ability straightforward.
> 
> That said, should you still prefer reusing rcu_dereference_sched_domain(),
> I am quite happy to convert them over.

*shrug*, either will do I suppose. You can create an alias if you think
it helps.