[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Thu, 02 Feb 2017 09:12:24 -0800
From: Eric Dumazet <eric.dumazet@...il.com>
To: Anoob Soman <anoob.soman@...rix.com>
Cc: David Miller <davem@...emloft.net>, netdev@...r.kernel.org
Subject: Re: [PATCH net] packet: call fanout_release, while UNREGISTERING a
netdev
On Thu, 2017-02-02 at 16:44 +0000, Anoob Soman wrote:
> On 02/02/17 15:53, Eric Dumazet wrote:
> > On Thu, 2017-02-02 at 14:42 +0000, Anoob Soman wrote:
> >
> >> I have tested both the approaches and LOCKDEP doesn't seem to catch any
> >> problem with the test I was doing.
> > Yeah, I think I will cleanup this mess, we probably can remove rcu
> > locking in control path, and stick to a single mutex to reduce all this
> > lock complexity.
> >
> > But that will be for net-next, while we need your fix for net tree.
> >
> > Thanks.
> >
> >
> Sorry, I didn't get it. You want me to post a patch for net tree,
> changing the way we do dev_{add,remove}_pack(fanout_prot_hook) and you
> will fix up the lock mess in net-next.
Yes. Please send your patch.
In the future (~1 month when net-next is re-open) I will submit a patch
to remove all these complexity out of af_packet.
Once fast path uses RCU, control path does not need to use a galaxy of
spinlocks and mutexes, and rcu.
This is too complex to maintain. A single mutex should be more than
enough.
Powered by blists - more mailing lists