On Sat, Jul 22, 2023 at 10:23:26AM +0100, Yun Levi wrote: > Hi, Paul. > > Thanks for looking into this :) > > > > Except that rcu_report_qs_rdp() is invoked with interrupts enabled, > > which means that there is some possibility of state changes up to the > > raw_spin_lock_irqsave_rcu_node(rnp, flags) statement. > > > > So, did you check whether RCU's interrupt paths change this state? > > In my narrow view, > only a new gp started, cpu_no_qs.b.norm changes as true in the path of > rcu_sched_clock_irq. > But in that case, rcu_report_qs_rdp isn't called. > > Did I understand your question well and are there any missed paths I didn't see? Suppose that the scheduler-clock interrupt invoking rcu_sched_clock_irq() happened just before the lock was acquired in rcu_report_qs_rdp(). Suppose further that the RCU grace-period kthread started a new grace period just before that interrupt occurred. Then mightn't that interrupt notice the new grace period and set ->cpu_no_qs.b.norm to true before fully returning? Thanx, Paul > > Why not start with something like this? > > > > if (!WARN_ON_ONCE(!rdp->cpu_no_qs.b.norm) || > > rdp->gp_seq != rnp->gp_seq || rdp->gpwrap) { > > > > Yes. but with different message