On Tue, Jun 06, 2017 at 07:23:42PM +0200, Peter Zijlstra wrote: > On Mon, Jun 05, 2017 at 03:09:50PM -0700, Paul E. McKenney wrote: > > diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c > > index 3ae8474557df..157654fa436a 100644 > > --- a/kernel/rcu/srcutree.c > > +++ b/kernel/rcu/srcutree.c > > @@ -357,7 +357,7 @@ EXPORT_SYMBOL_GPL(cleanup_srcu_struct); > > > > /* > > * Counts the new reader in the appropriate per-CPU element of the > > - * srcu_struct. Must be called from process context. > > + * srcu_struct. > > * Returns an index that must be passed to the matching srcu_read_unlock(). > > */ > > int __srcu_read_lock(struct srcu_struct *sp) > > @@ -365,7 +365,7 @@ int __srcu_read_lock(struct srcu_struct *sp) > > int idx; > > > > idx = READ_ONCE(sp->srcu_idx) & 0x1; > > - __this_cpu_inc(sp->sda->srcu_lock_count[idx]); > > + this_cpu_inc(sp->sda->srcu_lock_count[idx]); > > smp_mb(); /* B */ /* Avoid leaking the critical section. */ > > return idx; > > } > > So again, the change is to make this an IRQ safe operation, however if > we have this balance requirement, the IRQ will not visibly change the > value and load-store should be good again, no? > > Or am I missing some other detail with this implementation? Unlike Tiny SRCU, Classic and Tree SRCU increment one counter (->srcu_lock_count[]) and decrement another (->srcu_unlock_count[]). So balanced srcu_read_lock() and srcu_read_unlock() within an irq handler would increment both counters, with no decrements. Therefore, __srcu_read_lock()'s counter manipulation needs to be irq-safe. Thanx, Paul
