Messages in this thread | | | Date | Wed, 23 Mar 2022 18:04:02 -0700 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH 4/4] rcu: Name internal polling flag |
| |
On Tue, Mar 22, 2022 at 11:32:24AM +0100, Frederic Weisbecker wrote: > On Mon, Mar 21, 2022 at 07:11:07PM -0700, Paul E. McKenney wrote: > > On Wed, Mar 16, 2022 at 03:42:55PM +0100, Frederic Weisbecker wrote: > > > Give a proper self-explanatory name to the expedited grace period > > > internal polling flag. > > > > > > Signed-off-by: Frederic Weisbecker <frederic@kernel.org> > > > Cc: Neeraj Upadhyay <quic_neeraju@quicinc.com> > > > Cc: Boqun Feng <boqun.feng@gmail.com> > > > Cc: Uladzislau Rezki <uladzislau.rezki@sony.com> > > > Cc: Joel Fernandes <joel@joelfernandes.org> > > > --- > > > kernel/rcu/rcu.h | 5 +++++ > > > kernel/rcu/tree.c | 2 +- > > > kernel/rcu/tree_exp.h | 9 +++++---- > > > 3 files changed, 11 insertions(+), 5 deletions(-) > > > > > > diff --git a/kernel/rcu/rcu.h b/kernel/rcu/rcu.h > > > index eccbdbdaa02e..8a62bb416ba4 100644 > > > --- a/kernel/rcu/rcu.h > > > +++ b/kernel/rcu/rcu.h > > > @@ -30,6 +30,11 @@ > > > #define RCU_GET_STATE_USE_NORMAL 0x2 > > > #define RCU_GET_STATE_BAD_FOR_NORMAL (RCU_GET_STATE_FROM_EXPEDITED | RCU_GET_STATE_USE_NORMAL) > > > > > > +/* > > > + * Low-order bit definitions for polled grace-period internals. > > > + */ > > > +#define RCU_EXP_SEQ_POLL_DONE 0x1 > > > + > > > /* > > > * Return the counter portion of a sequence number previously returned > > > * by rcu_seq_snap() or rcu_seq_current(). > > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c > > > index 5da381a3cbe5..b3223b365f9f 100644 > > > --- a/kernel/rcu/tree.c > > > +++ b/kernel/rcu/tree.c > > > @@ -4679,7 +4679,7 @@ static void __init rcu_init_one(void) > > > spin_lock_init(&rnp->exp_lock); > > > mutex_init(&rnp->boost_kthread_mutex); > > > raw_spin_lock_init(&rnp->exp_poll_lock); > > > - rnp->exp_seq_poll_rq = 0x1; > > > + rnp->exp_seq_poll_rq = RCU_EXP_SEQ_POLL_DONE; > > > INIT_WORK(&rnp->exp_poll_wq, sync_rcu_do_polled_gp); > > > } > > > } > > > diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h > > > index c4a19c6a83cf..7ccb909d6355 100644 > > > --- a/kernel/rcu/tree_exp.h > > > +++ b/kernel/rcu/tree_exp.h > > > @@ -910,14 +910,14 @@ static void sync_rcu_do_polled_gp(struct work_struct *wp) > > > unsigned long s; > > > > > > s = READ_ONCE(rnp->exp_seq_poll_rq); > > > - if (s & 0x1) > > > + if (s & RCU_EXP_SEQ_POLL_DONE) > > > return; > > > while (!sync_exp_work_done(s)) > > > __synchronize_rcu_expedited(true); > > > > One additional question. If we re-read rnp->exp_seq_poll_rq on each pass > > through the loop, wouldn't we have less trouble with counter wrap? > > We can indeed do that, though it won't eliminate the possibility of wrapping.
True. But in conjunction with an exact check for expired grace-period sequence number, it reduces the maximum addtional penalty for wrapping to two grace periods.
Thanx, Paul
> > > raw_spin_lock_irqsave(&rnp->exp_poll_lock, flags); > > > s = rnp->exp_seq_poll_rq; > > > - if (!(s & 0x1) && sync_exp_work_done(s)) > > > - WRITE_ONCE(rnp->exp_seq_poll_rq, s | 0x1); > > > + if (!(s & RCU_EXP_SEQ_POLL_DONE) && sync_exp_work_done(s)) > > > + WRITE_ONCE(rnp->exp_seq_poll_rq, s | RCU_EXP_SEQ_POLL_DONE); > > > raw_spin_unlock_irqrestore(&rnp->exp_poll_lock, flags); > > > } > > > > > > @@ -946,7 +946,8 @@ unsigned long start_poll_synchronize_rcu_expedited(void) > > > rnp = rdp->mynode; > > > if (rcu_init_invoked()) > > > raw_spin_lock_irqsave(&rnp->exp_poll_lock, flags); > > > - if ((rnp->exp_seq_poll_rq & 0x1) || ULONG_CMP_LT(rnp->exp_seq_poll_rq, s)) { > > > + if ((rnp->exp_seq_poll_rq & RCU_EXP_SEQ_POLL_DONE) || > > > + ULONG_CMP_LT(rnp->exp_seq_poll_rq, s)) { > > > WRITE_ONCE(rnp->exp_seq_poll_rq, s); > > > if (rcu_init_invoked()) > > > queue_work(rcu_gp_wq, &rnp->exp_poll_wq); > > > -- > > > 2.25.1 > > >
| |