From: Charles Keepax Subject: Re: [PATCH v2 00/22] mfd: demodularization of non-modular drivers Date: Wed, 5 Dec 2018 13:40:16 +0000 Message-ID: <20181205134016.GE16508@imbe.wolfsonmicro.main> References: <1543811009-15112-1-git-send-email-paul.gortmaker@windriver.com> <20181205113535.GX16508@imbe.wolfsonmicro.main> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org To: Linus Walleij Cc: Paul Gortmaker , Lee Jones , "linux-kernel@vger.kernel.org" , Arnd Bergmann , Cory Maccarrone , David Dajun Chen , Dong Aisheng , Eric Miao , Graeme Gregory , Guennadi Liakhovetski , Haojian Zhuang , jinyoungp@nvidia.com, Jorge Eduardo Candelaria , Laxman Dewangan , Mark Brown , Mattias NILSSON , Michael Hennerich , Mike Rapoport List-Id: linux-omap@vger.kernel.org On Wed, Dec 05, 2018 at 12:48:47PM +0100, Linus Walleij wrote: > On Wed, Dec 5, 2018 at 12:36 PM Charles Keepax > wrote: > > On Sun, Dec 02, 2018 at 11:23:07PM -0500, Paul Gortmaker wrote: > > > The solution to #4 is similar - we delete the ".remove" function and > > > the binding into the platform_driver struct. However, since the same > > > ".remove" function could also be triggered by an "unbind" (such as for > > > pass-through of a device to a guest instance) - so we also explicitly > > > disable any unbind for the driver. > > > > > > The unbind mask allows us to ensure we will see if there was some odd > > > corner case out there that was relying on it. Typically it would be a > > > multi-port ethernet card passing a port through to a guest, so a > > > sensible use case in MFD drivers seems highly unlikely. This same > > > solution has already been used in multiple other mainline subsystems. > > > > > > > I guess if this is a general direction thing, but it does seem > > that module unload is not the only reason one might ever unbind a > > driver. So are we sure we want to remove the option to unbind > > these drivers? Certainly for testing it is sometimes useful. > > I personally never understood why these attributes are even > present on non-modular drivers. > > If testing is about exercising unbind/bind to reintialize > the code through a new call to .probe(), why would the developer > not take it all the way through and make it a module? > It just looks like a half-measure. > Well I guess in someways it is a half-measure. I vaguely seem to remember some dependency nightmare that can make it really hard to have the MFD allowed as a module in some cases, I can't remember the exact details but probably why some of these are not modules. I certainly don't strongly object to removing the ability to unbind these drivers, just wanted to make sure everyone is aligned that it's a good thing to do. Does kinda remove a couple of debugging options (debugging things like drivers interfering with each other) and the last chance restart the driver and see if that helps rescue something. Thanks, Charles