[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <20120919.134503.2070066674881117065.davem@davemloft.net>
Date: Wed, 19 Sep 2012 13:45:03 -0400 (EDT)
From: David Miller <davem@...emloft.net>
To: eric.dumazet@...il.com
Cc: netdev@...r.kernel.org, maze@...gle.com, therbert@...gle.com,
lorenzo@...gle.com
Subject: Re: [PATCH net-next] net: only run neigh_forced_gc() from one cpu
From: Eric Dumazet <eric.dumazet@...il.com>
Date: Wed, 19 Sep 2012 11:27:07 +0200
> From: Eric Dumazet <edumazet@...gle.com>
>
> With multiqueue NIC or RPS, we can have situation where all cpus are
> spending huge amount of cycles in neigh_forced_gc(), and machine can
> crash.
>
> Since we are under probable attack, its better to let only one cpu
> do the scan, and other cpus immediately return from neigh_forced_gc()
>
> Signed-off-by: Eric Dumazet <edumazet@...gle.com>
I suspect you're only seeing this with ipv6 traffic present.
And you should mention that because it determines how we really
should handle this.
I seriously doubt you can trigger this with a pure-ipv4 workload
because all neigh entries are ref-less and thus trivially reclaimable.
If you can trigger it with ipv4, then reclaim needs to be rewritten so
that for ref-less entries it's more efficient and scalable.
I'm not applying this patch.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists