From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?iso-8859-1?Q?Ga=EBtan?= Rivet Subject: Re: [PATCH v4 04/12] net/failsafe: add fail-safe PMD Date: Mon, 5 Jun 2017 01:04:38 +0200 Message-ID: <20170604230438.GG18840@bidouze.vm.6wind.com> References: <22f681327bdff42db58d880fd2238d1fbfb2730d.1496065002.git.gaetan.rivet@6wind.com> <20170531081353.2103a676@xeon-e3> <20170601140137.GB18840@bidouze.vm.6wind.com> <20170601105722.6e4e2101@xeon-e3> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Cc: dev@dpdk.org To: Stephen Hemminger Return-path: Received: from mail-wm0-f47.google.com (mail-wm0-f47.google.com [74.125.82.47]) by dpdk.org (Postfix) with ESMTP id 39A7F7CC9 for ; Mon, 5 Jun 2017 01:04:45 +0200 (CEST) Received: by mail-wm0-f47.google.com with SMTP id x70so286273wme.0 for ; Sun, 04 Jun 2017 16:04:45 -0700 (PDT) Content-Disposition: inline In-Reply-To: <20170601105722.6e4e2101@xeon-e3> List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" On Thu, Jun 01, 2017 at 10:57:22AM -0700, Stephen Hemminger wrote: > On Thu, 1 Jun 2017 16:01:37 +0200 > Gaëtan Rivet wrote: > > > What is necessary is a common piece of info (it can be MAC address, a > > class Id, anything else) that allows a script to detect that the right > > device has been plugged in. As long as the NFV server allows determinism > > here, the user will be able to use its VF. > > There is a serial number property (GUID) on vmbus that is what Windows uses > to associate VF and SF device. It maybe visible over sysfs. > > But it make senses to put that kind of system dependent matching rule in > some other place (DPDK bus logic? or external daemon). I think it might make sense to have expanded matching feature in rte_buses. For example, being able to match devices using metadatas specific to a bus, while allowing wildcards-like behavior. Such as expressing "Probe all devices having a vendor:device id pair in such ranges", and having an "INTR_ADD" event added to mirror the "INTR_RMV" already existing. Other possible grammar elements for device matching could be thought up, not limited to NFV servers. Once the buses have somewhat stabilized in this release or the next. But the exec stanza allows the fail-safe to have answers for this configuration and possibly many others, at least while the buses are not capable enough. -- Gaëtan Rivet 6WIND