From: James Bottomley Subject: Re: [PATCH] lpfc : add module parameter that allows adapter instances to avoid attachment Date: Fri, 27 Mar 2009 11:41:02 -0500 Message-ID: <1238172062.3292.14.camel@localhost.localdomain> References: <1238097300.27023.2.camel@ogier> <1238098747.3342.54.camel@localhost.localdomain> <49CBE772.80304@emulex.com> <1238100849.3342.58.camel@localhost.localdomain> <20090327150940.GK8014@parisc-linux.org> <1238168037.3292.8.camel@localhost.localdomain> <20090327161448.GM8014@parisc-linux.org> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: Received: from accolon.hansenpartnership.com ([76.243.235.52]:53038 "EHLO accolon.hansenpartnership.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753961AbZC0QlH (ORCPT ); Fri, 27 Mar 2009 12:41:07 -0400 In-Reply-To: <20090327161448.GM8014@parisc-linux.org> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Matthew Wilcox Cc: James Smart , "linux-scsi@vger.kernel.org" On Fri, 2009-03-27 at 10:14 -0600, Matthew Wilcox wrote: > On Fri, Mar 27, 2009 at 03:33:57PM +0000, James Bottomley wrote: > > Actually they better be ... we'll get into real trouble if they're not > > because of the way we flatten the space for multiple bus binding > > drivers. > > Er ... where do we do that? As far as I'm aware, to bind to multiple > busses, you register multiple foo_driver. They're separate namespaces. Yes, you're right. I was sure there was some way of flattening the namespace so we mixed bus_ids but perhaps Kay fixed it all. > > Even if I accepted your argument, I still can't see why we'd only > > implement this for PCI, and thus why it shouldn't be in the generic > > device part (except possibly with a bus type name qualifier). > > Adding the bus name qualifier would work. It seems like much more typing, > in order to get what advantage over just implementing it for PCI? The advantage is it's done generically and so works for all buses instead of just PCI. The disadvantage, I suppose, is that the matching must be on only things the generic model knows about, however, the proposal was PCI IDs, that doesn't seem to be a problem ... if you were to make it PCIe DSN then yes, it would be harder to do it generically. James