From mboxrd@z Thu Jan 1 00:00:00 1970 From: John Fastabend Subject: Re: [PATCH] net: pktgen: support injecting packets for qdisc testing Date: Sun, 10 Jan 2016 21:21:41 -0800 Message-ID: <56933BE5.8090702@gmail.com> References: <20160108020908.5940.89612.stgit@john-Precision-Tower-5810> <568F1EE2.7020107@gmail.com> <568FC9F8.8040201@mojatatu.com> <568FD498.5030401@gmail.com> <56926615.6040902@mojatatu.com> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, alexei.starovoitov@gmail.com, davem@davemloft.net To: Jamal Hadi Salim Return-path: Received: from mail-pa0-f52.google.com ([209.85.220.52]:33566 "EHLO mail-pa0-f52.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751035AbcAKFV4 (ORCPT ); Mon, 11 Jan 2016 00:21:56 -0500 Received: by mail-pa0-f52.google.com with SMTP id cy9so315643805pac.0 for ; Sun, 10 Jan 2016 21:21:56 -0800 (PST) In-Reply-To: <56926615.6040902@mojatatu.com> Sender: netdev-owner@vger.kernel.org List-ID: On 16-01-10 06:09 AM, Jamal Hadi Salim wrote: > On 16-01-08 10:24 AM, John Fastabend wrote: >> On 16-01-08 06:38 AM, Jamal Hadi Salim wrote: >>> On 16-01-07 09:28 PM, John Fastabend wrote: >>>> >>>> Hi Jamal, >>>> >>>> per your comment about using pktgen to test qdiscs here is the >>>> patch I've been using most the day which has been working well. >>>> I'm guessing this is more or less what you had in mind. >>>> >>> >>> I have attached the one i used over the holidays - see if there are >>> any differences(sorry dont have much time right now). >>> >>> cheers, >>> jamal >> >> Close except I don't back off the qdisc when I get >> xmit_{drop|cn|policed} return codes because I'm trying to >> stress the qdisc. >> > > In that case we need two modes. I think it is important to stress > the qdisc backoff. It doesnt make sense to send packets that will > be dropped. You said you were testing over vlan - how were you > enforcing no backoff? it was aborting so I wasn't able to do this with vlans. It is interesting for example with pfifo_tail_enqueue where we return NET_XMIT_CN but still enqueue the packet. If only to test the qdisc, agreed though "real" users should backoff. > >> Also I don't see the point to doing the skb_get_tx_queue and >> txq_trans_update() this should be called by dev_queue_xmit >> anyways. >> > > Yes, those could go away. They are leftovers of another experiment. > > >> If its all the same to you I would just assume have my version in >> pktgen and later if its useful for you add an option to back-off >> when the qdisc throws an error. > > > Refer to above. If you can fix it so we can do either i will ACK it. I'll send a Jamal approved version when net-next opens again ;) Thanks! John