Re: [PATCH net-next v2] rocker: move netevent neigh update to processes context

[David Miller <davem@davemloft.net>]

From: sfeldma@gmail.com
Date: Tue,  2 Jun 2015 20:43:28 -0700

> From: Scott Feldman <sfeldma@gmail.com>
> 
> v2:
> 
> Changes based on review:
> 
> - David Miller <davem@davemloft.net> raise problem with system_wq not
>   preserving queue order to execution order.  To fix, use driver-private
>   ordered workqueue to preserve ordering of queued work.
> 
> - Jiri Pirko <jiri@resnulli.us> small change on kfree of work queue item.
> 
> v1:
> 
> In review of Simon's patchset "rocker: transaction fixes". it was noted
> that rocker->neigh_tbl_next_index was unprotected in the call path below
> and could race with other contexts calling rocker_port_ipv4_neigh():

How it rocker->neigh_tbl_next_index not protected?

rocker->neigh_tbl_lock is _always_ held when it is accessed.

This patch, therefore, looks like completely unnecessary complexity
to me.  Furthermore, I would completely prefer if the operation stays
completely synchronous to the call path where the neigh operation
occurs rather than throwing it out to a workqueue.

Thanks.