On Fri, Feb 21, 2025 at 04:52:39PM +0100, Frederic Weisbecker wrote: > Le Wed, Feb 19, 2025 at 06:58:36AM -0800, Paul E. McKenney a écrit : > > On Sat, Feb 15, 2025 at 11:23:45PM +0100, Frederic Weisbecker wrote: > > > > Before. There was also some buggy debug code in play. Also, to get the > > > > failure, it was necessary to make TREE03 disable preemption, as stock > > > > TREE03 has an empty sync_sched_exp_online_cleanup() function. > > > > > > > > I am rerunning the test with a WARN_ON_ONCE() after the early exit from > > > > the sync_sched_exp_online_cleanup(). Of course, lack of a failure does > > > > not necessairly indicate > > > > > > Cool, thanks! > > > > No failures. But might it be wise to put this WARN_ON_ONCE() in, > > let things go for a year or two, and complete the removal if it never > > triggers? Or is the lack of forward progress warning enough? > > Hmm, what prevents a WARN_ON_ONCE() after the early exit of > sync_sched_exp_online_cleanup() to hit? > > All it takes is for sync_sched_exp_online_cleanup() to execute between > sync_exp_reset_tree() and __sync_rcu_exp_select_node_cpus() manage > to send an IPI. You are right, that would do it! > But we can warn about the lack of forward progress after a few iterations > of the retry_ipi label in __sync_rcu_exp_select_node_cpus(). Agreed, that would make more sense. > > > > > And if after do we know why? > > > > > > > > Here are some (possibly bogus) possibilities that came to mind: > > > > > > > > 1. There is some coming-online race that deprives the incoming > > > > CPU of an IPI, but nevertheless marks that CPU as blocking the > > > > current grace period. > > > > > > Arguably there is a tiny window between rcutree_report_cpu_starting() > > > and set_cpu_online() that could make ->qsmaskinitnext visible before > > > cpu_online() and therefore delay the IPI a bit. But I don't expect > > > more than a jiffy to fill up the gap. And if that's relevant, note that > > > only !PREEMPT_RCU is then "fixed" by sync_sched_exp_online_cleanup() here. > > > > Agreed. And I vaguely recall that there was some difference due to > > preemptible RCU's ability to clean up at the next rcu_read_unlock(), > > though more recently, possibly deferred. > > Perhaps at the time but today at least I can't find any. And maybe not even back then. ;-) Thanx, Paul