On 07/23/2015 07:13 PM, Stephen Hemminger wrote:

On Thu, 23 Jul 2015 19:05:56 +0200
Nikolay Aleksandrov [off-list ref] wrote:

quoted

On 07/23/2015 06:59 PM, Stephen Hemminger wrote:

quoted

On Thu, 23 Jul 2015 09:07:37 -0700
Nikolay Aleksandrov [off-list ref] wrote:
  

quoted

+		/* Stop hello and hold timer */
+		spin_lock_bh(&br->lock);
+		del_timer(&br->hello_timer);
+		list_for_each_entry(p, &br->port_list, list)
+			del_timer(&p->hold_timer);
+		spin_unlock_bh(&br->lock);  

Wouldn't it be easier to use del_timer_sync here?
  

I think it should work. Also I have an error in the commit message
about the kernel BPDU sending which I need to correct. I'll prepare
a v2 with your suggestion and fixed commit message.

The one thing to watch out for with del_timer_sync is that
the timer routine and the caller can't be using the same lock
otherwise timer will be spinning waiting to get lock that
is held by caller who is waiting for timer.

Actually I just noticed the locking was wrong in the patch, also we should use
del_timer() only because a spin_lock is acquired before that and cannot
use the blocking del_timer_sync() inside.
I'll fix it all up in the second version.