svn commit: r285051 - head/sys/netinet

[Ermal Luçi]

On Wed, Jul 29, 2015 at 6:48 PM, George Neville-Neil <gnn at neville-neil.com>
wrote:

>
>
> On 29 Jul 2015, at 11:05, Gleb Smirnoff wrote:
>
>  Ermal,
>>
>> On Wed, Jul 29, 2015 at 03:00:59PM +0200, Ermal Luçi wrote:
>> E> > E> @@ -934,6 +950,7 @@ ip_forward(struct mbuf *m, int srcrt)
>> E> > E>       */
>> E> > E>      if (!srcrt && ia == NULL) {
>> E> > E>              icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
>> E> > E> +            RO_RTFREE(&ro);
>> E> > E>              return;
>> E> > E>      }
>> E> >
>> E> > Here the ifa reference is leaked upon return.
>> E> >
>> E> >
>> E> Gleb,
>> E>
>> E> the improvement on the ifa_ref not needed is something to look at but
>> the
>> E> ifa_ref here is not lost since ia == NULL, no?
>> E> Maybe i am missing something else.
>>
>> Sure you are right. Mea culpa.
>>
>> E> Also can we put this on a review?
>>
>> It is possible. Let's just wait for Olivier to return and ask him to
>> do a benchmark :)
>>
>>
> Olivier isnt' the only one that can do a benchmark.  I can chuck this up
> in the
> Sentex lab, that's what it's for.  Give me a brief outline and I'll code
> something
> up in Conductor.
>

The outline is simple.
Just forwarding performance in terms of PPS for normal forwarding with the
patch suggested from Gleb builtin.
pmcstat data would be useful as well during this bench :)


>
> Best,
> George
>



-- 
Ermal