From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755942AbaC0LOJ (ORCPT ); Thu, 27 Mar 2014 07:14:09 -0400 Received: from mail-we0-f180.google.com ([74.125.82.180]:33508 "EHLO mail-we0-f180.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754638AbaC0LOH (ORCPT ); Thu, 27 Mar 2014 07:14:07 -0400 Message-ID: <533407F9.7060509@gmail.com> Date: Thu, 27 Mar 2014 12:14:01 +0100 From: Boris BREZILLON User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.4.0 MIME-Version: 1.0 To: Sylwester Nawrocki , Fabio Estevam CC: "linux-arm-kernel@lists.infradead.org" , Mike Turquette , "devicetree@vger.kernel.org" , Russell King , t.figa@samsung.com, linux-kernel , Kevin Hilman , Shawn Guo , Sascha Hauer , Nicolas Ferre , Jean-Christophe PLAGNIOL-VILLARD , Gregory Clement , Olof Johansson , "'Arnd Bergmann'" Subject: Re: [PATCH] clk: register fixed-clock only if #clock-cells property is present References: <1395858127-18730-1-git-send-email-s.nawrocki@samsung.com> <53333145.4080903@gmail.com> <5333DA13.2080505@gmail.com> <5333F704.7020504@samsung.com> In-Reply-To: <5333F704.7020504@samsung.com> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Sylwester, Le 27/03/2014 11:01, Sylwester Nawrocki a écrit : > Hi Boris, > > On 27/03/14 08:58, Boris BREZILLON wrote: >> This solution solve the problem for this specific case because clks are >> declared in the correct order in imx DTs. >> But, even with your patch I think we could see similar issues by >> reordering DT nodes... >> >> The real problem here is that imx platform does not declare the CCM clocks >> dependencies upon ckil, ckih1 and osc fixed clocks within the DT [1], and >> retrieve these clocks when initializing the CCM clocks ([2] and [3]). >> >> We should try to a add these dependencies in the DT and see if it works. > While presumably all of us agree the dependencies should be correctly > specified in dts I think we should minimize possible regressions by > keeping the clocks registration order as before, i.e. as parsed by the > kernel from DT. Rather than explicitly reversing it, which does not gain > us anything AFAICS. Instead we are seeing regressions where new kernels > stop working with old dtbs. I totally agree with you on this point: my patch is not a replacement of yours. I just wanted to point out that we need to fix DT definitions to avoid these kind of issues in the future. > > I'm going to resend the patch replacing list_add() with list_add_tail(), > with this the mvebu platform would work and there should be no regression > on imx and exynos. > > Please note that specifying dependencies between CCM on imx and the fixed > clocks might not be enough. If the fixed clocks get matched on "fixed-clock" > compatible some clock specifiers (i.e. those using phandle to the CCM) could > get invalid, since the clocks won't get registered by the ccm driver, but by > the regular fixed clock driver. That means a phandle to different node would > need to be used to reference the fixed clock. I'm not sure if this is the case > for imx, but changes may be needed all over various dts files. > In addition, we should make sure the kernel works with current and modified > dtbs. > >> [1] http://lxr.free-electrons.com/source/arch/arm/boot/dts/imx6sl.dtsi#L379 >> [2] http://lxr.free-electrons.com/source/arch/arm/mach-imx/clk-imx6q.c#L151 >> [3] http://lxr.free-electrons.com/source/arch/arm/mach-imx/clk.c#L30