Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 7 Apr 2023 21:37:56 +0300
From:   Sakari Ailus <sakari.ailus@iki.fi>
To:     Martin Kepplinger <martin.kepplinger@puri.sm>
Cc:     mchehab@kernel.org, laurent.pinchart@ideasonboard.com,
        kernel@puri.sm, linux-media@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH v1 1/2] media: hi846: fix usage of
 pm_runtime_get_if_in_use()
Message-ID: <ZDBjBNu6UXksVSwQ@valkosipuli.retiisi.eu>
References: <20230405092904.1129395-1-martin.kepplinger@puri.sm>
 <20230405092904.1129395-2-martin.kepplinger@puri.sm>
 <ZC1vJNIGeEu+xi+i@valkosipuli.retiisi.eu>
 <cb0c1ad7e00523a43ce6ad73a79e7e396c4b52aa.camel@puri.sm>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <cb0c1ad7e00523a43ce6ad73a79e7e396c4b52aa.camel@puri.sm>
Precedence: bulk

Hi Martin,

On Fri, Apr 07, 2023 at 03:31:02PM +0200, Martin Kepplinger wrote:
> Am Mittwoch, dem 05.04.2023 um 15:52 +0300 schrieb Sakari Ailus:
> > Hi Martin,
> > 
> > On Wed, Apr 05, 2023 at 11:29:03AM +0200, Martin Kepplinger wrote:
> > > pm_runtime_get_if_in_use() does not only return nonzero values when
> > > the device is in use, it can return a negative errno too.
> > > 
> > > And especially during resuming from system suspend, when runtime pm
> > > is not yet up again, this can very well happen. And in such a case
> > > the subsequent pm_runtime_put() call would result in a refcount
> > > underflow!
> > 
> > I think this issue should have a more generic solution, it's very
> > difficult
> > to address this in drivers only with the current APIs.
> > 
> > pm_runtime_get_if_in_use() will also return an error if runtime PM is
> > disabled, so this patch will break the driver for that configuration.
> 
> ok but the driver is currently broken for any *other* error returned by
> pm_runtime_get_if_in_use() (than the runtime-PM disabled error).
> 
> The execution-path during system-resume I'm interested in gets -EAGAIN
> here. Would it be ok for you if I'd return early only for that one
> error only here?

I guess... but I think to address this in a way that's reasonable to
drivers, we'll need improvements to runtime PM API. A largish number of
drivers need changes and before doing that we should figure out exactly
what should be done.

I thought you could effectively trigger this issue by calling runtime PM
resume/suspend functions before enabling runtime PM, but this seems to be a
different case.

-- 
Kind regards,

Sakari Ailus