Re: [PATCH rcu 04/12] rcu: Switch polled grace-period APIs to ->gp_seq_polled

[Paul E. McKenney]

On Wed, Jul 20, 2022 at 05:53:38PM -0700, Boqun Feng wrote:
> Hi Paul,
> 
> On Mon, Jun 20, 2022 at 03:51:20PM -0700, Paul E. McKenney wrote:
> > This commit switches the existing polled grace-period APIs to use a
> > new ->gp_seq_polled counter in the rcu_state structure.  An additional
> > ->gp_seq_polled_snap counter in that same structure allows the normal
> > grace period kthread to interact properly with the !SMP !PREEMPT fastpath
> > through synchronize_rcu().  The first of the two to note the end of a
> > given grace period will make knowledge of this transition available to
> > the polled API.
> > 
> > This commit is in preparation for polled expedited grace periods.
> > 
> > Link: https://lore.kernel.org/all/20220121142454.1994916-1-bfoster@xxxxxxxxxx/
> > Link: https://docs.google.com/document/d/1RNKWW9jQyfjxw2E8dsXVTdvZYh0HnYeSHDKog9jhdN8/edit?usp=sharing
> > Cc: Brian Foster <bfoster@xxxxxxxxxx>
> > Cc: Dave Chinner <david@xxxxxxxxxxxxx>
> > Cc: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
> > Cc: Ian Kent <raven@xxxxxxxxxx>
> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > ---
> >  kernel/rcu/tree.c | 90 +++++++++++++++++++++++++++++++++++++++++++++--
> >  kernel/rcu/tree.h |  2 ++
> >  2 files changed, 89 insertions(+), 3 deletions(-)
> > 
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index 46cfceea87847..637e8f9454573 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -1775,6 +1775,78 @@ static void rcu_strict_gp_boundary(void *unused)
> >  	invoke_rcu_core();
> >  }
> >  
> > +// Has rcu_init() been invoked?  This is used (for example) to determine
> > +// whether spinlocks may be acquired safely.
> > +static bool rcu_init_invoked(void)
> > +{
> > +	return !!rcu_state.n_online_cpus;
> > +}
> > +
> > +// Make the polled API aware of the beginning of a grace period.
> > +static void rcu_poll_gp_seq_start(unsigned long *snap)
> > +{
> > +	struct rcu_node *rnp = rcu_get_root();
> > +
> > +	if (rcu_init_invoked())
> > +		raw_lockdep_assert_held_rcu_node(rnp);
> > +
> > +	// If RCU was idle, note beginning of GP.
> > +	if (!rcu_seq_state(rcu_state.gp_seq_polled))
> > +		rcu_seq_start(&rcu_state.gp_seq_polled);
> > +
> > +	// Either way, record current state.
> > +	*snap = rcu_state.gp_seq_polled;
> > +}
> > +
> > +// Make the polled API aware of the end of a grace period.
> > +static void rcu_poll_gp_seq_end(unsigned long *snap)
> > +{
> > +	struct rcu_node *rnp = rcu_get_root();
> > +
> > +	if (rcu_init_invoked())
> > +		raw_lockdep_assert_held_rcu_node(rnp);
> > +
> > +	// If the the previously noted GP is still in effect, record the
> > +	// end of that GP.  Either way, zero counter to avoid counter-wrap
> > +	// problems.
> > +	if (*snap && *snap == rcu_state.gp_seq_polled) {
> > +		rcu_seq_end(&rcu_state.gp_seq_polled);
> > +		rcu_state.gp_seq_polled_snap = 0;
> > +	} else {
> > +		*snap = 0;
> > +	}
> > +}
> > +
> > +// Make the polled API aware of the beginning of a grace period, but
> > +// where caller does not hold the root rcu_node structure's lock.
> > +static void rcu_poll_gp_seq_start_unlocked(unsigned long *snap)
> > +{
> > +	struct rcu_node *rnp = rcu_get_root();
> > +
> > +	if (rcu_init_invoked()) {
> > +		lockdep_assert_irqs_enabled();
> > +		raw_spin_lock_irq_rcu_node(rnp);
> > +	}
> > +	rcu_poll_gp_seq_start(snap);
> > +	if (rcu_init_invoked())
> > +		raw_spin_unlock_irq_rcu_node(rnp);
> > +}
> > +
> > +// Make the polled API aware of the end of a grace period, but where
> > +// caller does not hold the root rcu_node structure's lock.
> > +static void rcu_poll_gp_seq_end_unlocked(unsigned long *snap)
> > +{
> > +	struct rcu_node *rnp = rcu_get_root();
> > +
> > +	if (rcu_init_invoked()) {
> > +		lockdep_assert_irqs_enabled();
> > +		raw_spin_lock_irq_rcu_node(rnp);
> > +	}
> > +	rcu_poll_gp_seq_end(snap);
> > +	if (rcu_init_invoked())
> > +		raw_spin_unlock_irq_rcu_node(rnp);
> > +}
> > +
> >  /*
> >   * Initialize a new grace period.  Return false if no grace period required.
> >   */
> > @@ -1810,6 +1882,7 @@ static noinline_for_stack bool rcu_gp_init(void)
> >  	rcu_seq_start(&rcu_state.gp_seq);
> >  	ASSERT_EXCLUSIVE_WRITER(rcu_state.gp_seq);
> >  	trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start"));
> > +	rcu_poll_gp_seq_start(&rcu_state.gp_seq_polled_snap);
> >  	raw_spin_unlock_irq_rcu_node(rnp);
> >  
> >  	/*
> > @@ -2069,6 +2142,7 @@ static noinline void rcu_gp_cleanup(void)
> >  	 * safe for us to drop the lock in order to mark the grace
> >  	 * period as completed in all of the rcu_node structures.
> >  	 */
> > +	rcu_poll_gp_seq_end(&rcu_state.gp_seq_polled_snap);
> >  	raw_spin_unlock_irq_rcu_node(rnp);
> >  
> >  	/*
> > @@ -3837,8 +3911,18 @@ void synchronize_rcu(void)
> >  			 lock_is_held(&rcu_lock_map) ||
> >  			 lock_is_held(&rcu_sched_lock_map),
> >  			 "Illegal synchronize_rcu() in RCU read-side critical section");
> > -	if (rcu_blocking_is_gp())
> > +	if (rcu_blocking_is_gp()) {
> > +		// Note well that this code runs with !PREEMPT && !SMP.
> > +		// In addition, all code that advances grace periods runs
> > +		// at process level.  Therefore, this GP overlaps with other
> > +		// GPs only by being fully nested within them, which allows
> > +		// reuse of ->gp_seq_polled_snap.
> > +		rcu_poll_gp_seq_start_unlocked(&rcu_state.gp_seq_polled_snap);
> > +		rcu_poll_gp_seq_end_unlocked(&rcu_state.gp_seq_polled_snap);
> > +		if (rcu_init_invoked())
> > +			cond_resched_tasks_rcu_qs();
> >  		return;  // Context allows vacuous grace periods.
> > +	}
> >  	if (rcu_gp_is_expedited())
> >  		synchronize_rcu_expedited();
> >  	else
> > @@ -3860,7 +3944,7 @@ unsigned long get_state_synchronize_rcu(void)
> >  	 * before the load from ->gp_seq.
> >  	 */
> >  	smp_mb();  /* ^^^ */
> > -	return rcu_seq_snap(&rcu_state.gp_seq);
> > +	return rcu_seq_snap(&rcu_state.gp_seq_polled);
> 
> I happened to run into this. There is one usage of
> get_state_synchronize_rcu() in start_poll_synchronize_rcu(), in which
> the return value of get_state_synchronize_rcu() ("gp_seq") will be used
> for rcu_start_this_gp(). I don't think this is quite right, because
> after this change, rcu_state.gp_seq and rcu_state.gp_seq_polled are
> different values, in fact ->gp_seq_polled is greater than ->gp_seq
> by how many synchronize_rcu() is called in early boot.
> 
> Am I missing something here?

It does not appear that your are missing anything, sad to say!

Does the following make it work better?

							Thanx, Paul

------------------------------------------------------------------------

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 2122359f0c862..cf2fd58a93a41 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -3571,7 +3571,7 @@ EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
 unsigned long start_poll_synchronize_rcu(void)
 {
 	unsigned long flags;
-	unsigned long gp_seq = get_state_synchronize_rcu();
+	unsigned long gp_seq = rcu_seq_snap(&rcu_state.gp_seq);
 	bool needwake;
 	struct rcu_data *rdp;
 	struct rcu_node *rnp;