Re: [PATCH 1/1] rcu/tree: Drop the lock before entering to page allocator

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

On Wed, Jul 15, 2020 at 09:32:50PM +0200, Sebastian Andrzej Siewior wrote:
> On 2020-07-15 21:02:43 [+0200], Uladzislau Rezki wrote:
> > 
> > <snip>
> >     spin_lock();
> >     __get_free_page(GFP_NOWAIT | __GFP_NOWARN);
> >     spin_unlock();
> > <snip>
> > 
> > Also, please note we do it for regular kernel.
> 
> ach right okay then.
> 
> > >
> > > What happened to the part where I asked for a spinlock_t?
> > > 
> > What do you mean?
> 
> Please drop that raw_spinlock_t for the kfree_rcu() based locking and
> use just a plain spinlock_t for the locking. Then you can keep the same
> code flow for RT and !RT without any special cases and everything.

My concern is that some critical bug will show up at some point
that requires double-argument kfree_rcu() be invoked while holding
a raw spinlock.  (Single-argument kfree_rcu() must sometimes invoke
synchronize_rcu(), so it can never be invoked in any state forbidding
invoking schedule().)

Yes, dropping to a plain spinlock would be simple in the here and now,
but experience indicates that it is only a matter of time, and that when
that time comes it will come as an emergency.

One approach would be to replace the "IS_ENABLED(CONFIG_PREEMPT_RT)"
with some sort of check for being in a context where spinlock acquisition
is not legal.  What could be done along those lines?

							Thanx, Paul