]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
rcu: Eliminate BUG_ON() for sync.c
authorPaul E. McKenney <paulmck@linux.ibm.com>
Mon, 22 Oct 2018 14:43:22 +0000 (07:43 -0700)
committerPaul E. McKenney <paulmck@linux.ibm.com>
Fri, 9 Nov 2018 05:41:57 +0000 (21:41 -0800)
The sync.c file has a number of calls to BUG_ON(), which panics the
kernel, which is not a good strategy for devices (like embedded) that
don't have a way to capture console output.  This commit therefore
changes these BUG_ON() calls to WARN_ON_ONCE(), but does so quite naively.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
kernel/rcu/sync.c

index 3f943efcf61c1e3c31373799f475c80967ae3fd0..a6ba446a9693bba192a647d1488eb93cc56bf3d1 100644 (file)
@@ -125,8 +125,7 @@ void rcu_sync_enter(struct rcu_sync *rsp)
                rsp->gp_state = GP_PENDING;
        spin_unlock_irq(&rsp->rss_lock);
 
-       BUG_ON(need_wait && need_sync);
-
+       WARN_ON_ONCE(need_wait && need_sync);
        if (need_sync) {
                gp_ops[rsp->gp_type].sync();
                rsp->gp_state = GP_PASSED;
@@ -139,7 +138,7 @@ void rcu_sync_enter(struct rcu_sync *rsp)
                 * Nobody has yet been allowed the 'fast' path and thus we can
                 * avoid doing any sync(). The callback will get 'dropped'.
                 */
-               BUG_ON(rsp->gp_state != GP_PASSED);
+               WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
        }
 }
 
@@ -166,8 +165,8 @@ static void rcu_sync_func(struct rcu_head *rhp)
        struct rcu_sync *rsp = container_of(rhp, struct rcu_sync, cb_head);
        unsigned long flags;
 
-       BUG_ON(rsp->gp_state != GP_PASSED);
-       BUG_ON(rsp->cb_state == CB_IDLE);
+       WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
+       WARN_ON_ONCE(rsp->cb_state == CB_IDLE);
 
        spin_lock_irqsave(&rsp->rss_lock, flags);
        if (rsp->gp_count) {
@@ -225,7 +224,7 @@ void rcu_sync_dtor(struct rcu_sync *rsp)
 {
        int cb_state;
 
-       BUG_ON(rsp->gp_count);
+       WARN_ON_ONCE(rsp->gp_count);
 
        spin_lock_irq(&rsp->rss_lock);
        if (rsp->cb_state == CB_REPLAY)
@@ -235,6 +234,6 @@ void rcu_sync_dtor(struct rcu_sync *rsp)
 
        if (cb_state != CB_IDLE) {
                gp_ops[rsp->gp_type].wait();
-               BUG_ON(rsp->cb_state != CB_IDLE);
+               WARN_ON_ONCE(rsp->cb_state != CB_IDLE);
        }
 }