[ofa-general] Re: [PATCH 3/4] ipoib: fix a deadlock between ipoib start/stop and child interface create/delete

[Yossi Etigin]

You're right, it should not be atomic.  I wanted to protect the flag
itself but as I was told by Moni Shoua it's useless.
I'm sending v2.

Roland Dreier wrote:
>  > +	atomic_t vlan_task_flag;
> 
> why is this atomic_t?  I only see:
> 
>  > +		atomic_set(&priv->vlan_task_flag, 1);
>  > +		atomic_set(&priv->vlan_task_flag, 0);
>  > +	iffup_value = atomic_read(&priv->vlan_task_flag) ? IFF_UP : 0;
> 
> so as far as I can tell you are not using anything atomic.  So if
> there's a race you're worried about, it's still there...
> 
>  - R.

-- 
--Yossi