From mboxrd@z Thu Jan 1 00:00:00 1970 From: Roland Dreier Subject: Re: [ofa-general] InfiniBand/RDMA merge plans for 2.6.24 Date: Fri, 14 Sep 2007 09:18:01 -0700 Message-ID: References: <46E97BB0.9030106@opengridcomputing.com> <1189724358.9540.113.camel@dell> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "Steve Wise" , "netdev" , linux-kernel@vger.kernel.org, general@lists.openfabrics.org, anilgv@broadcom.com, uri@broadcom.com To: "Michael Chan" Return-path: Received: from sj-iport-3-in.cisco.com ([171.71.176.72]:6601 "EHLO sj-iport-3.cisco.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1764614AbXINQSO (ORCPT ); Fri, 14 Sep 2007 12:18:14 -0400 In-Reply-To: <1189724358.9540.113.camel@dell> (Michael Chan's message of "Thu, 13 Sep 2007 15:59:18 -0700") Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org > > I've been meaning to track down the bnx2 iscsi offload patch to look > > and see if this issue is addressed, since the same problem seems to > > exist: it seems an iscsi connection and a main stack tcp connection > > might share the same 4-tuple unless something is done to avoid that > > happening. > iSCSI does not do passive listens, only active connections to the > target. But you're right, the port space is still shared between iSCSI > and the main stack. We currently rely on user apps binding to the main > stack to reserve certain ephemeral ports, and telling the iSCSI driver > which ports to use. Got it... I wasn't thinking that clearly, but it is clear that a full 4-tuple collision with only active connections is quite unlikely. I guess you would have to make both an offloaded and a non-offloaded iSCSI connection to the same target and get really unlucky with ephemeral port allocation. So in practice I guess it's not an issue at all with your driver yet. However, do you have any plans to support iSCSI offload for targets? Also, looking at the first CNIC patch, I can't help but notice that you seem to have at least some support for iWARP there. How does the CNIC look? Does it share the same interface/addresses as the non-offload NIC, or does it create a completely separate netdevice? I want to make sure that whatever solution we come up with for cxgb3 doesn't cause problems for you. And of course if you have a better idea than what Steve has come up with, that would be great :) - R.