rcu: Overlap wakeups with next expedited grace period

The current expedited grace-period implementation makes subsequent grace
periods wait on wakeups for the prior grace period.  This does not fit
the dictionary definition of "expedited", so this commit allows these two
phases to overlap.  Doing this requires four waitqueues rather than two
because tasks can now be waiting on the previous, current, and next grace
periods.  The fourth waitqueue makes the bit masking work out nicely.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index e8fff14..1df100c 100644 (file)
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -103,6 +103,7 @@ struct rcu_state sname##_state = { \
        .name = RCU_STATE_NAME(sname), \
        .abbr = sabbr, \
        .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \
+       .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \
 }
 
 RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
@@ -3637,7 +3638,7 @@ static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
                        trace_rcu_exp_funnel_lock(rsp->name, rnp->level,
                                                  rnp->grplo, rnp->grphi,
                                                  TPS("wait"));
-                       wait_event(rnp->exp_wq[(s >> 1) & 0x1],
+                       wait_event(rnp->exp_wq[(s >> 1) & 0x3],
                                   sync_exp_work_done(rsp,
                                                      &rdp->exp_workdone2, s));
                        return true;
@@ -3857,6 +3858,14 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
        synchronize_sched_expedited_wait(rsp);
        rcu_exp_gp_seq_end(rsp);
        trace_rcu_exp_grace_period(rsp->name, s, TPS("end"));
+
+       /*
+        * Switch over to wakeup mode, allowing the next GP, but -only- the
+        * next GP, to proceed.
+        */
+       mutex_lock(&rsp->exp_wake_mutex);
+       mutex_unlock(&rsp->exp_mutex);
+
        rcu_for_each_node_breadth_first(rsp, rnp) {
                if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s)) {
                        spin_lock(&rnp->exp_lock);
@@ -3865,10 +3874,10 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
                                rnp->exp_seq_rq = s;
                        spin_unlock(&rnp->exp_lock);
                }
-               wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x1]);
+               wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]);
        }
        trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));
-       mutex_unlock(&rsp->exp_mutex);
+       mutex_unlock(&rsp->exp_wake_mutex);
 }
 
 /**
@@ -4530,6 +4539,8 @@ static void __init rcu_init_one(struct rcu_state *rsp)
                        rcu_init_one_nocb(rnp);
                        init_waitqueue_head(&rnp->exp_wq[0]);
                        init_waitqueue_head(&rnp->exp_wq[1]);
+                       init_waitqueue_head(&rnp->exp_wq[2]);
+                       init_waitqueue_head(&rnp->exp_wq[3]);
                        spin_lock_init(&rnp->exp_lock);
                }
        }

diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
index f9d4fbb..1194ab0 100644 (file)
--- a/kernel/rcu/tree.h
+++ b/kernel/rcu/tree.h
@@ -250,7 +250,7 @@ struct rcu_node {
 
        spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
        unsigned long exp_seq_rq;
-       wait_queue_head_t exp_wq[2];
+       wait_queue_head_t exp_wq[4];
 } ____cacheline_internodealigned_in_smp;
 
 /*
@@ -502,6 +502,7 @@ struct rcu_state {
        /* End of fields guarded by barrier_mutex. */
 
        struct mutex exp_mutex;                 /* Serialize expedited GP. */
+       struct mutex exp_wake_mutex;            /* Serialize wakeup. */
        unsigned long expedited_sequence;       /* Take a ticket. */
        atomic_long_t expedited_normal;         /* # fallbacks to normal. */
        atomic_t expedited_need_qs;             /* # CPUs left to check in. */