[ofa-general] [PATCH] ipoib: defer skb_orphan() until irqs enabled

[akepner at sgi.com]

On Wed, Sep 10, 2008 at 11:26:16AM -0700, Roland Dreier wrote:
> ....
> I don't think this closes the race completely: at the point skb_orphan
> is called (after Arthur's patch, by design), we have no locks held.  And
> so the timer-driven send completion handling could already have run and
> freed the skb between when we drop tx_lock and when we call skb_orphan.
> 

Suppose we could just remove the skb_orphan() call from ipoib_send() 
entirely, and wait for net_tx_action() to do it for us. But I imagine 
there must be a (performance-related) reason why it's done the way 
it is.

-- 
Arthur