From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mc.com (iris-63.mc.com [63.96.239.141]) by ozlabs.org (Postfix) with ESMTP id 905D0DDEE3 for ; Thu, 7 Jun 2007 18:10:18 +1000 (EST) From: Jean-Christophe Dubois To: michael@ellerman.id.au Subject: Re: [RFC/PATCH 4/4] Add support for MSI on Axon-based Cell systems Date: Thu, 7 Jun 2007 10:10:04 +0200 References: <200706050900.54022.jdubois@mc.com> <1181203139.4166.26.camel@concordia.ozlabs.ibm.com> In-Reply-To: <1181203139.4166.26.camel@concordia.ozlabs.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Message-Id: <200706071010.05172.jdubois@mc.com> Cc: linuxppc-dev@ozlabs.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Thursday 07 June 2007 09:58:59 Michael Ellerman wrote: > On Tue, 2007-06-05 at 09:00 +0200, Jean-Christophe Dubois wrote: > > On Monday 04 June 2007 15:00:05 Michael Ellerman wrote: > > > + pr_debug("axon_msi: initialising ...\n"); > > > + > > > + for_each_compatible_node(node, NULL, "ibm,axon-msic") { > > > + if (axon_msi_setup_one(of_node_get(node)) == 0) > > > + found++; > > > + } > > > + of_node_put(node); > > > > 2 comments: > > > > 1) There is no "ibm,axon-msic" compatible property in the SLOF tree > > provided by IBM on the CAB (or other such platforms). Therefore this code > > will not work on these platforms. > > It's not clear to me if MSIs will work on CAB, with Axon in end-point > mode. But it'd be good to try it at some point. The CAB (or other derived products) can be set in RC mode (via a jumper). In this case the MSI could be supported. > The compatible property is new, it's only just been added to SLOF > recently by Heiko, not sure when it will be in released firmware. > > > 2) you should somehow check for the Axon version. Axon 1.1 (and prior) > > are know to be loosing MSI/MBX interrupts (but not the data part > > associated to them). Therefore this should be enabled only on Axon 2.1 > > (or later). > > Yep. It works some of the time on 1.1, but not enough to enable it. > > I guess I'd say firmware that runs on 1.1 should just not export the msi > nodes, or not put the compatible property on them - rather than having > the kernel explicitly checking for the version. > > cheers