On Thu, 2015-10-08 at 13:53 -0700, Tom Herbert wrote:
> If the incoming CPU is set for a connected UDP via
> sk_incoming_cpu_update, wouldn't this check subsequently _only_ allow
> packets for that socket to come from the same CPU?
>
Hmm, I thought the SO_REUSEPORT path would be taken only for non
connected UDP sockets (like TCP listeners.).
But you might be right !
> Also, the check seems a little austere. Why not do something like:
>
> if (sk->sk_incoming_cpu != -1) {
> if (sk->sk_incoming_cpu != raw_smp_processor_id())
> score += 4;
> }
>
> My worry is that the packet steering configuration may change without
> the application's knowledge, so it's possible packets may come in on
> CPUs that the are unexpected to the application and then they would be
> dropped without matching a socket. I suppose that this could work with
> the original patch if a socket is bound to every CPU or there is at
> least one listener socket that is not bound to any CPU.
This is what I initially wrote, then I attempted a short cut, (abort
full list scan), then forgot to re-instate the first try, when I decided
to let this for future patch (Ying patch)
if (sk->sk_incoming_cpu == raw_smp_processor_id())
score++;
(Note we do not even have to test for sk_incoming_cpu == -1 in this
variant)
I'll include this in v2.
Thanks.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
