Re: [PATCH 3/3] srcu: Fix broken node geometry after early ssp init

["Paul E. McKenney" <paulmck@kernel.org>]

On Fri, Apr 02, 2021 at 12:02:21PM +0200, Frederic Weisbecker wrote:
> On Thu, Apr 01, 2021 at 06:12:41PM -0700, Paul E. McKenney wrote:
> > On Fri, Apr 02, 2021 at 01:47:04AM +0200, Frederic Weisbecker wrote:
> gg> >  void __init srcu_init(void)
> > >  {
> > > -	struct srcu_struct *ssp;
> > > +	static struct srcu_data __initdata old_sdp;
> > > +	static struct srcu_node __initdata old_mynode;
> > > +	struct srcu_struct *ssp, *tmp;
> > >  
> > >  	srcu_init_done = true;
> > > -	while (!list_empty(&srcu_boot_list)) {
> > > -		ssp = list_first_entry(&srcu_boot_list, struct srcu_struct,
> > > -				      work.work.entry);
> > > +	/*
> > > +	 * ssp's that got initialized before rcu_init_geometry() have a stale
> > > +	 * node hierarchy. Rebuild their node trees and propagate states from
> > > +	 * pruned leaf nodes.
> > > +	 */
> > > +	list_for_each_entry_safe(ssp, tmp, &srcu_early_init_list, early_init) {
> > > +		struct srcu_data *sdp = this_cpu_ptr(ssp->sda);
> > > +
> > > +		list_del(&ssp->early_init);
> > > +		old_sdp = *sdp;
> > > +		old_mynode = *sdp->mynode;
> > > +		init_srcu_struct_nodes(ssp);
> > > +		restore_srcu_sdp(sdp, &old_sdp, &old_mynode);
> > 
> > Why not just pull all of the pre-initialization callbacks onto a local
> > list (re-initialization the rcu_segcblist structures if necessary),
> > then iterate through that list re-invoking call_srcu() on each?
> > There shouldn't be that many pre-initialization call_srcu() invocations,
> > and if someday there are, it would not be hard to make __call_srcu()
> > take lists of callbacks and a count instead of a single callback, correct?
> > 
> > Or am I once again missing something basic?
> 
> So that would work for callbacks based started grace periods but not for the
> others. So if anybody calls start_poll_synchronize_rcu() early, simply requeuing
> the callbacks won't help. Even worse, a blind reinitialization of the ssp would
> lose the started grace period and a future poll_state_synchronize_rcu() might
> never succeed.

Good point...

> Arguably that's a quite a corner case and I don't expect anyone to call
> start_poll_synchronize_srcu() so early but who knows. An alternative is to
> forbid it and warn if used before srcu is initialized.

Another approach would be to have start_poll_synchronize_rcu() check to
see if initialization has happened, and if not, simply queue a callback.

Any other ways to make this work?

							Thanx, Paul