From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andy Shevchenko Subject: Re: [PATCH v1 06/15] i2c: acpi: Assign fwnode for devices created via i2c_acpi_new_device() Date: Wed, 21 Nov 2018 15:35:49 +0200 Message-ID: <20181121133549.GL10650@smile.fi.intel.com> References: <20181120155924.10773-1-andriy.shevchenko@linux.intel.com> <20181120155924.10773-7-andriy.shevchenko@linux.intel.com> 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: Hans de Goede Cc: Darren Hart , platform-driver-x86@vger.kernel.org, "Rafael J. Wysocki" , linux-acpi@vger.kernel.org, Jonathan Cameron , Wolfram Sang , Mika Westerberg , linux-i2c@vger.kernel.org, Heikki Krogerus , linux-kernel@vger.kernel.org List-Id: platform-driver-x86.vger.kernel.org On Wed, Nov 21, 2018 at 12:44:27PM +0100, Hans de Goede wrote: > On 20-11-18 16:59, Andy Shevchenko wrote: > > i2c_acpi_new_device() doesn't assign fwnode like it's done, for example, > > in i2c_acpi_register_devices() path. > > > > Assign fwnode in i2c_acpi_new_device() as it's done elsewhere. > As already mentioned in my reply to the cover letter, a NACK from me for > this one. > > The problem is that this causes the fwnode to be shared by all the > "struct device"-s (embedded in the i2c-clients created). > > This in itself is not necessarily a problem, but it does become a > problem when combined with using device_add_properties as > intel_cht_int33fe.c does. > > Since the info in the ACPI tables for devices which use this > driver is pretty terrible, we add a bunch of properties to notify > the drivers about how the different parts which make up the Type-C > functionality are tied together. > > device_add_properties creates a new fwnode and then calls > set_secondary_fwnode. For devices which already have a fwnode > assigned (which the i2c-clients will have after this patch) > set_secondary_fwnode sets dev->fwnode->secondary to the fwnode > which has been newly created to hold the added properties. > > Since all i2c-clients instantiated now share dev->fwnode, > the result of this is that all i2c-clients created now will > have the properties of the last i2c-client instantiated for > an acpi-node which describes multiple clients in a single > node. > > Also the fwnodes created for the properties of earlier > instantiated i2c-clients will be leaked. Thanks for this very detailed problem report. > > I see 2 possible solutions here: > > 1) Changing this patch so that i2c_acpi_new_device() does: > > /* > * The fwnode can not be shared between instantiated clients when > * adding device-properties to the instantiated client. > */ > if (!info->properties) > info->fwnode = acpi_fwnode_handle(adev); > > 2) Leave whether info->fwnode should be set or not up to the > caller of i2c_acpi_new_device() (info is already passed in as > a param). E.g. i2c-multi-instantiate.c could do this (for now, > maybe in the future we will also want to add properties for > some HIDs). > > I think that 1. is the best solution and this is all kernel internal, > so we can always revisit this if necessary. Heikki told me he has some ideas, but for now I will drop this patch. -- With Best Regards, Andy Shevchenko