Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <1952449.TVsm6CJCTy@aspire.rjw.lan> <CAJZ5v0jOVLZDoW5BAy9obp7AptTYzKikFo7Qd15kNt5Y9=vXRQ@mail.gmail.com>
 <CAPDyKFrq-r-rFDDB3yU=W4mOF2-A4xGGomqFqUba6MVWZugHmQ@mail.gmail.com>
 <24600564.pMaimVIQbW@aspire.rjw.lan> <CAJZ5v0gBYyoumVkLTuEcHsFE8QKEXdYuhDRGwmvoKPhx7gPvgQ@mail.gmail.com>
In-Reply-To: <CAJZ5v0gBYyoumVkLTuEcHsFE8QKEXdYuhDRGwmvoKPhx7gPvgQ@mail.gmail.com>
From:   Ulf Hansson <ulf.hansson@linaro.org>
Date:   Wed, 6 Feb 2019 12:23:26 +0100
Message-ID: <CAPDyKFoUfM8kopWfkorDhVKfk35wVaKVPzJeZ55RPRdc2Y1MYw@mail.gmail.com>
Subject: Re: [PATCH v2 0/9] driver core: Fix some device links issues and add
 "consumer autoprobe" flag
To:     "Rafael J. Wysocki" <rafael@kernel.org>
Cc:     "Rafael J. Wysocki" <rjw@rjwysocki.net>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux PM <linux-pm@vger.kernel.org>,
        Daniel Vetter <daniel@ffwll.ch>,
        Lukas Wunner <lukas@wunner.de>,
        Andrzej Hajda <a.hajda@samsung.com>,
        Russell King - ARM Linux <linux@armlinux.org.uk>,
        Lucas Stach <l.stach@pengutronix.de>,
        Linus Walleij <linus.walleij@linaro.org>,
        Thierry Reding <thierry.reding@gmail.com>,
        Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
        Marek Szyprowski <m.szyprowski@samsung.com>,
        Joerg Roedel <jroedel@suse.de>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, 6 Feb 2019 at 10:56, Rafael J. Wysocki <rafael@kernel.org> wrote:
>
> On Tue, Feb 5, 2019 at 12:27 PM Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> >
> > On Tuesday, February 5, 2019 9:15:49 AM CET Ulf Hansson wrote:
> > > On Mon, 4 Feb 2019 at 12:45, Rafael J. Wysocki <rafael@kernel.org> wrote:
> > > >
> > > > On Mon, Feb 4, 2019 at 12:40 PM Rafael J. Wysocki <rafael@kernel.org> wrote:
> > > > >
> > > > > On Fri, Feb 1, 2019 at 4:18 PM Ulf Hansson <ulf.hansson@linaro.org> wrote:
>
> [cut]
>
> > > >
> > > > For example, if the consumer device is suspended after the
> > > > device_link_add() that incremented the supplier's PM-runtime count and
> > > > then resumed again, the rpm_active refcount will be greater than one
> > > > because of the last resume and not because of the initial link
> > > > creation.  In that case, dropping the supplier's PM-runtime count on
> > > > link deletion may not work as expected.
> > >
> > > I see what your are saying and I must admit, by looking at the code,
> > > that it has turned into being rather complicated. Assuming of good
> > > reasons, of course.
> > >
> > > Anyway, I will play a little bit more with my tests to see what I can find out.
> > >
> > > >
> > > > > Arguably, device_link_del() could be made automatically drop the
> > > > > supplier's PM-runtime count by one if the link's rpm_active refcount
> > > > > is not one, but there will be failing scenarios in that case too
> > > > > AFAICS.
> > >
> > > Let's see.
> >
> > So for the record, below is the (untested) patch I'm thinking about.
> >
> > Having considered this for some time, I think that it would be better to
> > try to drop the supplier's PM-runtime usage counter on link removal even if
> > the link doesn't go away then.  That would be more consistent at least IMO.
>
> So I can't convince myself that this is the case.
>
> Either way, if there are two callers of device_link_add() for one
> consumer-supplier pair trying to add a stateless link between them and
> one of these callers passes DL_FLAG_RPM_ACTIVE set in the flags to it,
> there may be issues regardless of what device_link_del() and
> device_link_remove() do.  However, if they decrement the link's
> rpm_active refcount (and possibly the supplier's PM-runtime usage
> counter too), the supplier may be suspended prematurely, whereas in
> the other case (no decrementation of rpm_active, which how the code
> works after this series) it may just be prevented from suspending.  To
> me, the former is worse than the latter.

Well, I would say it sucks in both cases. :-)

>
> Moreover, there is a workaround for the latter issue that seems to be
> easy enough: it is sufficient to let the consumer runtime suspend
> after calling device_link_add() to create the link (with
> DL_FLAG_RPM_ACTIVE set) and before trying to remove it.

I get your point, but unfortunate I don't think it's that simple.

For example, someone (like a child) may prevent runtime suspend for
the consumer. Hence, also the supplier is prevented from being runtime
suspended.

So, if you want to push this responsibility to the driver, then I
think we need make __pm_runtime_set_status() to respect device links,
similar to how it already deals with child/parents.

In that way, the driver could call pm_runtime_set_suspended(), before
dropping the device link in ->probe(), which would allow the supplier
to also become runtime suspended.

I did a quick research of users of device links, unless I am mistaken,
this seems like an okay approach.

What do you think?

>
> Because of the above, I'm just going to post a patch to document the
> current behavior of the code as a known limitation.

Let's not give up, yet, please. I am sure we can figure something out.

Kind regards
Uffe