From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from phobos.denx.de (phobos.denx.de [85.214.62.61]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 13430C433F5 for ; Sat, 26 Feb 2022 18:42:15 +0000 (UTC) Received: from h2850616.stratoserver.net (localhost [IPv6:::1]) by phobos.denx.de (Postfix) with ESMTP id 0678A83D5B; Sat, 26 Feb 2022 19:40:32 +0100 (CET) Authentication-Results: phobos.denx.de; dmarc=pass (p=none dis=none) header.from=chromium.org Authentication-Results: phobos.denx.de; spf=pass smtp.mailfrom=u-boot-bounces@lists.denx.de Authentication-Results: phobos.denx.de; dkim=pass (1024-bit key; unprotected) header.d=chromium.org header.i=@chromium.org header.b="OR35D/NH"; dkim-atps=neutral Received: by phobos.denx.de (Postfix, from userid 109) id 8174383BCC; Sat, 26 Feb 2022 19:38:09 +0100 (CET) Received: from mail-oo1-xc35.google.com (mail-oo1-xc35.google.com [IPv6:2607:f8b0:4864:20::c35]) (using TLSv1.3 with cipher TLS_AES_128_GCM_SHA256 (128/128 bits)) (No client certificate requested) by phobos.denx.de (Postfix) with ESMTPS id CC4A283C04 for ; Sat, 26 Feb 2022 19:37:49 +0100 (CET) Authentication-Results: phobos.denx.de; dmarc=pass (p=none dis=none) header.from=chromium.org Authentication-Results: phobos.denx.de; spf=pass smtp.mailfrom=sjg@google.com Received: by mail-oo1-xc35.google.com with SMTP id x6-20020a4a4106000000b003193022319cso11933609ooa.4 for ; Sat, 26 Feb 2022 10:37:49 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=chromium.org; s=google; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc; bh=M0jb9iWqaNT5I7Z4hHjhtLJYKsIVfQHDtb0ZCccnQcY=; b=OR35D/NH10CISGutPnUJCfaEuvHqFbspsf/zMxsFw8fjgEn6u0xrOZ24QO867Ecq38 5/xcuuKdqUAjwptio+nP6A31Zbtk8Okk5Jz8JglFlZbl17yA9Qrd5cs593fsyUMJUBBn hjYNw1yQ5YPeHjog9RC8m4iVg3Q4n+QJn/iWs= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=M0jb9iWqaNT5I7Z4hHjhtLJYKsIVfQHDtb0ZCccnQcY=; b=s6zSfeA5guIV/Wmds1w9hepT9zv0zm4eOQr8K30QgthGaXN5Pb878P0dKnYs7gjqt9 XPaGbWCu9yA4VcbmgNomgT12YzO1VzLxwhdSLcWK5xLviJVw3dqV9mPj1A/n/m9eUoOu Pnk13UVg/BYBP3bi5Lm9+rbW9QIxm8N42TkCXesTO2HXpknZWcKpnsNtYCiXIdUkaz7U uY1SicbzTXjJkCNZlyme/RVyMVp/aSd88UbJWWOOFidCTL176Z06vIUwOzMN/U74O0IF uQk1HYcPBBTQEYOHHDcSETdnHMib5ElhAHe2+++nBZmT1CS9iAe0WhPNV/zCV7SXhZwF AR1w== X-Gm-Message-State: AOAM532YeNfg1+/eWr/vfuLaXK9RaHssNzzoWTbvmE5d7ZlcVEYqGp1l VzliOfXUj8VTvsFWJCwIPjRl5Qbku0zzVwzOE83DSg== X-Google-Smtp-Source: ABdhPJyccxnKm8/VmyN7GCCK/3fvchtkKnTdpWRa9cXnWT7lJrK2Se0sJyAiDZR8gCBGZo/9/ui/NqGfInaFZxIMAKE= X-Received: by 2002:a05:6870:a44e:b0:d7:a8d:1afc with SMTP id n14-20020a056870a44e00b000d70a8d1afcmr840864oal.27.1645900668290; Sat, 26 Feb 2022 10:37:48 -0800 (PST) MIME-Version: 1.0 References: <20220115222504.617013-1-seanga2@gmail.com> <20220115222504.617013-2-seanga2@gmail.com> <9cc38f2d-2c1f-e656-56ca-b7888ff0df63@gmail.com> <4bdef436-8cca-6879-f017-20b721f9a8dc@gmail.com> <75aed704-b354-a6c3-892a-9640ecad858f@gmail.com> In-Reply-To: <75aed704-b354-a6c3-892a-9640ecad858f@gmail.com> From: Simon Glass Date: Sat, 26 Feb 2022 11:36:43 -0700 Message-ID: Subject: Re: [PATCH 1/7] clk: Make rfree return void To: Sean Anderson Cc: U-Boot Mailing List , Lukasz Majewski Content-Type: text/plain; charset="UTF-8" X-BeenThere: u-boot@lists.denx.de X-Mailman-Version: 2.1.39 Precedence: list List-Id: U-Boot discussion List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: u-boot-bounces@lists.denx.de Sender: "U-Boot" X-Virus-Scanned: clamav-milter 0.103.5 at phobos.denx.de X-Virus-Status: Clean Hi Sean, On Tue, 1 Feb 2022 at 21:24, Sean Anderson wrote: > > On 2/1/22 10:59 PM, Simon Glass wrote: > > Hi Sean, > > > > On Tue, 1 Feb 2022 at 07:49, Sean Anderson wrote: > >> > >> On 1/27/22 4:35 PM, Simon Glass wrote: > >>> Hi Sean, > >>> > >>> On Thu, 27 Jan 2022 at 08:43, Sean Anderson wrote: > >>>> > >>>> On 1/27/22 10:05 AM, Simon Glass wrote: > >>>>> Hi Sean, > >>>>> > >>>>> On Sat, 15 Jan 2022 at 15:25, Sean Anderson wrote: > >>>>>> > >>>>>> When freeing a clock there is not much we can do if there is an error, and > >>>>>> most callers do not actually check the return value. Even e.g. checking to > >>>>>> make sure that clk->id is valid should have been done in request() in the > >>>>>> first place (unless someone is messing with the driver behind our back). > >>>>>> Just return void and don't bother returning an error. > >>>>>> > >>>>>> Signed-off-by: Sean Anderson > >>>>>> --- > >>>>>> > >>>>>> drivers/clk/clk-uclass.c | 7 +++---- > >>>>>> drivers/clk/clk_sandbox.c | 6 +++--- > >>>>>> include/clk-uclass.h | 8 +++----- > >>>>>> 3 files changed, 9 insertions(+), 12 deletions(-) > >>>>>> > >>>>> > >>>>> We have the same thing in other places too, but I am a little worried > >>>>> about removing error checking. We try to avoid checking arguments too > >>>>> much in U-Boot, due to code-size concerns, so I suppose I agree that > >>>>> an invalid clk should be caught by a debug assertion rather than a > >>>>> full check. But with driver model we have generally added an error > >>>>> return to every uclass method, for consistency and to permit returning > >>>>> error information if needed. > >>>>> > >>>>> Regards, > >>>>> Simon > >>>>> > >>>> > >>>> So there are a few reasons why I don't think a return value is useful > >>>> here. To illustrate this, consider a typical user of the clock API: > >>>> > >>>> struct clk a, b; > >>>> > >>>> ret = clk_get_by_name(dev, "a", &a); > >>>> if (ret) > >>>> return ret; > >>>> > >>>> ret = clk_get_by_name(dev, "b", &b); > >>>> if (ret) > >>>> goto free_a; > >>>> > >>>> ret = clk_set_rate(&a, 5000000); > >>>> if (ret) > >>>> goto free_b; > >>>> > >>>> ret = clk_enable(&b); > >>>> > >>>> free_b: > >>>> clk_free(&b); > >>>> free_a: > >>>> clk_free(&a); > >>>> return ret; > >>>> > >>>> - Because a and b are "thick pointers" they do not need any cleanup to > >>>> free their own resources. The only cleanup might be if the clock > >>>> driver has allocated something in clk_request (more on this below) > >>>> - By the time we call clk_free, the mutable portions of the function > >>>> have already completed. In effect, the function has succeeded, > >>>> regardless of whether clk_free fails. Additionally, we cannot take any > >>>> action if it fails, since we still have to free both clocks. > >>>> - clk_free occurs during the error path of the function. Even if it > >>>> errored, we do not want to override the existing error from one of the > >>>> functions doing "real" work. > >>>> > >>>> The last thing is that no clock driver actually does anything in rfree. > >>>> The only driver with this function is the sandbox driver. I would like > >>>> to remove the function altogether. As I understand it, the existing API > >>>> is inspired by the reset drivers, so I would like to review its usage in > >>>> the reset subsystem before removing it for the clock subsystem. I also > >>>> want to make some changes to how rates and enables/disables are > >>>> calculated which might provide a case for rfree. But once that is > >>>> complete I think there will be no users still. > >>> > >>> What does this all look like in Linux? > >> > >> Their equivalent (clk_put) returns void, and generally so do most other > >> cleanup functions, since .device_remove also returns void. > > > > We really cannot ignore errors from device_remove(). > > Once you are at device_remove, all the users are gone and it's up to the > device to clean up after itself. And often there is nothing we can do > once remove is called. As you yourself say in device_remove, > > /* We can't put the children back */ Well this assumes that device_remove() is actually removing the device, not just disabling DMA, etc. > > Really the only sensible thing is to print an error and continue booting > if possible. > > And of course no clock drivers actually use this function anyway, nor do > (all but 5) users check it. > > > Anyway I think what you say about the 'thick pointer' makes sense. But > > my expectation was that removing a clock might turn off a clock above > > it in the tree, for example. > > No, this just frees resources (as is documented). If you want to turn > off a clock, you have to call clk_disable. In fact, a very common use > case is just like the example above, where the consmer frees the clock > after enabling it. > > (This is also why clk->enable_count/rate are basically useless for > anything other than CCF clocks) How about a clock provided by an audio codec on an I2C bus? Should clk_free() do anything in that case? I assume not. I think the compelling part of your argument is that it is a 'think pointer' and disable does nothing. So can you update clk_rfree() etc. to document what is allowed to be done in that function? Reviewed-by: Simon Glass Regards, Simon