On Thu, Sep 27, 2018 at 3:00 PM, Eric Dumazet <eric.duma...@gmail.com> wrote: > > > On 09/27/2018 01:10 AM, Dmitry Vyukov wrote: > >> >> Would a stack trace for call_rcu be helpful here? I have this idea for >> a long time, but never get around to implementing it: >> https://bugzilla.kernel.org/show_bug.cgi?id=198437 >> >> Also FWIW I recently used the following hack for another net bug. It >> made that other bug involving call_rcu way more likely to fire. Maybe >> it will be helpful here too. >> >> diff --git a/net/core/dst.c b/net/core/dst.c >> index 81ccf20e28265..591a8d0aca545 100644 >> --- a/net/core/dst.c >> +++ b/net/core/dst.c >> @@ -187,8 +187,16 @@ void dst_release(struct dst_entry *dst) >> if (unlikely(newrefcnt < 0)) >> net_warn_ratelimited("%s: dst:%p refcnt:%d\n", >> __func__, dst, newrefcnt); >> - if (!newrefcnt) >> - call_rcu(&dst->rcu_head, dst_destroy_rcu); >> + if (!newrefcnt) { >> + if (lock_is_held(&rcu_bh_lock_map) || >> + lock_is_held(&rcu_lock_map) || >> + lock_is_held(&rcu_sched_lock_map)) { >> + call_rcu(&dst->rcu_head, dst_destroy_rcu); >> + } else { >> + synchronize_rcu(); > > dst_release() can be called in context we hold a spinlock, this would be bad > to reschedule here.
I am not suggesting to commit this. This is just a hack for debugging. It in fact lead to some warnings, but still allowed me to reproduce the bug reliably. >> + dst_destroy_rcu(&dst->rcu_head); >> + } >> + } >> } >> }
