rcu: Break up rcu_gp_kthread() into subfunctions
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Fri, 22 Jun 2012 18:08:41 +0000 (11:08 -0700)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Sun, 23 Sep 2012 14:41:53 +0000 (07:41 -0700)
Then rcu_gp_kthread() function is too large and furthermore needs to
have the force_quiescent_state() code pulled in.  This commit therefore
breaks up rcu_gp_kthread() into rcu_gp_init() and rcu_gp_cleanup().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
kernel/rcutree.c

index ddc6acc85d269ce425e4bbac08bc5681e756a502..f0c0c1b4b6d4e02f1cacb3ce62ce19906d887515 100644 (file)
@@ -1028,95 +1028,159 @@ rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_dat
 }
 
 /*
- * Body of kthread that handles grace periods.
+ * Initialize a new grace period.
  */
-static int __noreturn rcu_gp_kthread(void *arg)
+static int rcu_gp_init(struct rcu_state *rsp)
 {
-       unsigned long gp_duration;
        struct rcu_data *rdp;
-       struct rcu_node *rnp;
-       struct rcu_state *rsp = arg;
+       struct rcu_node *rnp = rcu_get_root(rsp);
 
-       for (;;) {
+       raw_spin_lock_irq(&rnp->lock);
+       rsp->gp_flags = 0;
 
-               /* Handle grace-period start. */
-               rnp = rcu_get_root(rsp);
-               for (;;) {
-                       wait_event_interruptible(rsp->gp_wq, rsp->gp_flags);
-                       if (rsp->gp_flags)
-                               break;
-                       flush_signals(current);
-               }
+       if (rcu_gp_in_progress(rsp)) {
+               /* Grace period already in progress, don't start another.  */
+               raw_spin_unlock_irq(&rnp->lock);
+               return 0;
+       }
+
+       if (rsp->fqs_active) {
+               /*
+                * We need a grace period, but force_quiescent_state()
+                * is running.  Tell it to start one on our behalf.
+                */
+               rsp->fqs_need_gp = 1;
+               raw_spin_unlock_irq(&rnp->lock);
+               return 0;
+       }
+
+       /* Advance to a new grace period and initialize state. */
+       rsp->gpnum++;
+       trace_rcu_grace_period(rsp->name, rsp->gpnum, "start");
+       WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT);
+       rsp->fqs_state = RCU_GP_INIT; /* Stop force_quiescent_state. */
+       rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
+       record_gp_stall_check_time(rsp);
+       raw_spin_unlock_irq(&rnp->lock);
+
+       /* Exclude any concurrent CPU-hotplug operations. */
+       get_online_cpus();
+
+       /*
+        * Set the quiescent-state-needed bits in all the rcu_node
+        * structures for all currently online CPUs in breadth-first order,
+        * starting from the root rcu_node structure, relying on the layout
+        * of the tree within the rsp->node[] array.  Note that other CPUs
+        * will access only the leaves of the hierarchy, thus seeing that no
+        * grace period is in progress, at least until the corresponding
+        * leaf node has been initialized.  In addition, we have excluded
+        * CPU-hotplug operations.
+        *
+        * The grace period cannot complete until the initialization
+        * process finishes, because this kthread handles both.
+        */
+       rcu_for_each_node_breadth_first(rsp, rnp) {
                raw_spin_lock_irq(&rnp->lock);
-               rsp->gp_flags = 0;
                rdp = this_cpu_ptr(rsp->rda);
+               rcu_preempt_check_blocked_tasks(rnp);
+               rnp->qsmask = rnp->qsmaskinit;
+               rnp->gpnum = rsp->gpnum;
+               rnp->completed = rsp->completed;
+               if (rnp == rdp->mynode)
+                       rcu_start_gp_per_cpu(rsp, rnp, rdp);
+               rcu_preempt_boost_start_gp(rnp);
+               trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
+                                           rnp->level, rnp->grplo,
+                                           rnp->grphi, rnp->qsmask);
+               raw_spin_unlock_irq(&rnp->lock);
+               cond_resched();
+       }
 
-               if (rcu_gp_in_progress(rsp)) {
-                       /*
-                        * A grace period is already in progress, so
-                        * don't start another one.
-                        */
-                       raw_spin_unlock_irq(&rnp->lock);
-                       cond_resched();
-                       continue;
-               }
+       rnp = rcu_get_root(rsp);
+       raw_spin_lock_irq(&rnp->lock);
+       /* force_quiescent_state() now OK. */
+       rsp->fqs_state = RCU_SIGNAL_INIT;
+       raw_spin_unlock_irq(&rnp->lock);
+       put_online_cpus();
+       return 1;
+}
 
-               if (rsp->fqs_active) {
-                       /*
-                        * We need a grace period, but force_quiescent_state()
-                        * is running.  Tell it to start one on our behalf.
-                        */
-                       rsp->fqs_need_gp = 1;
-                       raw_spin_unlock_irq(&rnp->lock);
-                       cond_resched();
-                       continue;
-               }
+/*
+ * Clean up after the old grace period.
+ */
+static int rcu_gp_cleanup(struct rcu_state *rsp)
+{
+       unsigned long gp_duration;
+       struct rcu_data *rdp;
+       struct rcu_node *rnp = rcu_get_root(rsp);
 
-               /* Advance to a new grace period and initialize state. */
-               rsp->gpnum++;
-               trace_rcu_grace_period(rsp->name, rsp->gpnum, "start");
-               WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT);
-               rsp->fqs_state = RCU_GP_INIT; /* Stop force_quiescent_state. */
-               rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
-               record_gp_stall_check_time(rsp);
-               raw_spin_unlock_irq(&rnp->lock);
+       raw_spin_lock_irq(&rnp->lock);
+       gp_duration = jiffies - rsp->gp_start;
+       if (gp_duration > rsp->gp_max)
+               rsp->gp_max = gp_duration;
 
-               /* Exclude any concurrent CPU-hotplug operations. */
-               get_online_cpus();
+       /*
+        * We know the grace period is complete, but to everyone else
+        * it appears to still be ongoing.  But it is also the case
+        * that to everyone else it looks like there is nothing that
+        * they can do to advance the grace period.  It is therefore
+        * safe for us to drop the lock in order to mark the grace
+        * period as completed in all of the rcu_node structures.
+        *
+        * But if this CPU needs another grace period, it will take
+        * care of this while initializing the next grace period.
+        * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL
+        * because the callbacks have not yet been advanced: Those
+        * callbacks are waiting on the grace period that just now
+        * completed.
+        */
+       rdp = this_cpu_ptr(rsp->rda);
+       if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) {
+               raw_spin_unlock_irq(&rnp->lock);
 
                /*
-                * Set the quiescent-state-needed bits in all the rcu_node
-                * structures for all currently online CPUs in breadth-first
-                * order, starting from the root rcu_node structure.
-                * This operation relies on the layout of the hierarchy
-                * within the rsp->node[] array.  Note that other CPUs will
-                * access only the leaves of the hierarchy, which still
-                * indicate that no grace period is in progress, at least
-                * until the corresponding leaf node has been initialized.
-                * In addition, we have excluded CPU-hotplug operations.
+                * Propagate new ->completed value to rcu_node
+                * structures so that other CPUs don't have to
+                * wait until the start of the next grace period
+                * to process their callbacks.
                 */
                rcu_for_each_node_breadth_first(rsp, rnp) {
                        raw_spin_lock_irq(&rnp->lock);
-                       rcu_preempt_check_blocked_tasks(rnp);
-                       rnp->qsmask = rnp->qsmaskinit;
-                       rnp->gpnum = rsp->gpnum;
-                       rnp->completed = rsp->completed;
-                       if (rnp == rdp->mynode)
-                               rcu_start_gp_per_cpu(rsp, rnp, rdp);
-                       rcu_preempt_boost_start_gp(rnp);
-                       trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
-                                                   rnp->level, rnp->grplo,
-                                                   rnp->grphi, rnp->qsmask);
+                       rnp->completed = rsp->gpnum;
                        raw_spin_unlock_irq(&rnp->lock);
                        cond_resched();
                }
-
                rnp = rcu_get_root(rsp);
                raw_spin_lock_irq(&rnp->lock);
-               /* force_quiescent_state() now OK. */
-               rsp->fqs_state = RCU_SIGNAL_INIT;
-               raw_spin_unlock_irq(&rnp->lock);
-               put_online_cpus();
+       }
+
+       rsp->completed = rsp->gpnum; /* Declare grace period done. */
+       trace_rcu_grace_period(rsp->name, rsp->completed, "end");
+       rsp->fqs_state = RCU_GP_IDLE;
+       if (cpu_needs_another_gp(rsp, rdp))
+               rsp->gp_flags = 1;
+       raw_spin_unlock_irq(&rnp->lock);
+       return 1;
+}
+
+/*
+ * Body of kthread that handles grace periods.
+ */
+static int __noreturn rcu_gp_kthread(void *arg)
+{
+       struct rcu_state *rsp = arg;
+       struct rcu_node *rnp = rcu_get_root(rsp);
+
+       for (;;) {
+
+               /* Handle grace-period start. */
+               for (;;) {
+                       wait_event_interruptible(rsp->gp_wq, rsp->gp_flags);
+                       if (rsp->gp_flags && rcu_gp_init(rsp))
+                               break;
+                       cond_resched();
+                       flush_signals(current);
+               }
 
                /* Handle grace-period end. */
                rnp = rcu_get_root(rsp);
@@ -1125,56 +1189,12 @@ static int __noreturn rcu_gp_kthread(void *arg)
                                                 !ACCESS_ONCE(rnp->qsmask) &&
                                                 !rcu_preempt_blocked_readers_cgp(rnp));
                        if (!ACCESS_ONCE(rnp->qsmask) &&
-                           !rcu_preempt_blocked_readers_cgp(rnp))
+                           !rcu_preempt_blocked_readers_cgp(rnp) &&
+                           rcu_gp_cleanup(rsp))
                                break;
+                       cond_resched();
                        flush_signals(current);
                }
-
-               raw_spin_lock_irq(&rnp->lock);
-               gp_duration = jiffies - rsp->gp_start;
-               if (gp_duration > rsp->gp_max)
-                       rsp->gp_max = gp_duration;
-
-               /*
-                * We know the grace period is complete, but to everyone else
-                * it appears to still be ongoing.  But it is also the case
-                * that to everyone else it looks like there is nothing that
-                * they can do to advance the grace period.  It is therefore
-                * safe for us to drop the lock in order to mark the grace
-                * period as completed in all of the rcu_node structures.
-                *
-                * But if this CPU needs another grace period, it will take
-                * care of this while initializing the next grace period.
-                * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL
-                * because the callbacks have not yet been advanced: Those
-                * callbacks are waiting on the grace period that just now
-                * completed.
-                */
-               if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) {
-                       raw_spin_unlock_irq(&rnp->lock);
-
-                       /*
-                        * Propagate new ->completed value to rcu_node
-                        * structures so that other CPUs don't have to
-                        * wait until the start of the next grace period
-                        * to process their callbacks.
-                        */
-                       rcu_for_each_node_breadth_first(rsp, rnp) {
-                               raw_spin_lock_irq(&rnp->lock);
-                               rnp->completed = rsp->gpnum;
-                               raw_spin_unlock_irq(&rnp->lock);
-                               cond_resched();
-                       }
-                       rnp = rcu_get_root(rsp);
-                       raw_spin_lock_irq(&rnp->lock);
-               }
-
-               rsp->completed = rsp->gpnum; /* Declare grace period done. */
-               trace_rcu_grace_period(rsp->name, rsp->completed, "end");
-               rsp->fqs_state = RCU_GP_IDLE;
-               if (cpu_needs_another_gp(rsp, rdp))
-                       rsp->gp_flags = 1;
-               raw_spin_unlock_irq(&rnp->lock);
        }
 }