[bug report] rcu: Diagnostics for grace-period hangs

[Dan Carpenter <dan.carpenter@oracle.com>]

Hello Paul E. McKenney,

The patch e2ba0f065aec: "rcu: Diagnostics for grace-period hangs"
from Apr 21, 2018, leads to the following static checker warning:

	kernel/rcu/tree.c:2739 rcu_check_gp_start_stall()
	error: double lock 'irqsave:flags'

kernel/rcu/tree.c
  2726  
  2727          raw_spin_lock_irqsave_rcu_node(rnp, flags);
  2728          j = jiffies;
  2729          if (rcu_gp_in_progress(rsp) ||
  2730              ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
  2731              time_before(j, READ_ONCE(rsp->gp_req_activity) + HZ) ||
  2732              time_before(j, READ_ONCE(rsp->gp_activity) + HZ) ||
  2733              atomic_read(&warned)) {
  2734                  raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  2735                  return;
  2736          }
  2737          /* Hold onto the leaf lock to make others see warned=1. */
  2738  
  2739          raw_spin_lock_irqsave_rcu_node(rnp_root, flags);

Saving irqsave is not nestable.  The second save overwrites the first so
that IRQs can't be re-enabled.

  2740          j = jiffies;
  2741          if (rcu_gp_in_progress(rsp) ||
  2742              ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) ||
  2743              time_before(j, rsp->gp_req_activity + HZ) ||
  2744              time_before(j, rsp->gp_activity + HZ) ||
  2745              atomic_xchg(&warned, 1)) {
  2746                  raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags);
  2747                  raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  2748                  return;
  2749          }
  2750          pr_alert("%s: g%ld->%ld gar:%lu ga:%lu f%#x %s->state:%#lx\n",
  2751                   __func__, (long)READ_ONCE(rsp->gp_seq),
  2752                   (long)READ_ONCE(rnp_root->gp_seq_needed),
  2753                   j - rsp->gp_req_activity, j - rsp->gp_activity,
  2754                   rsp->gp_flags, rsp->name,
  2755                   rsp->gp_kthread ? rsp->gp_kthread->state : 0x1ffffL);
  2756          WARN_ON(1);
  2757          raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags);
  2758          raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  2759  }


regards,
dan carpenter