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;
Message-ID: <5d56bb77.1c69fb81.58e9d.1f86@mx.google.com>
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <CAJZ5v0hY8=0j=heXuAS-5cBafDSE8ZakLDW4NGCjAbxUAt3j4Q@mail.gmail.com>
References: <1565731976.8572.16.camel@lca.pw> <5d53b238.1c69fb81.d3cd3.cd53@mx.google.com> <20190814084014.GB52127@atomide.com> <CANA+-vDeSAYUNfTQzQPT2N_CUgvYr6i_LP_BdHT_zX+FPt8NHg@mail.gmail.com> <CAJZ5v0hY8=0j=heXuAS-5cBafDSE8ZakLDW4NGCjAbxUAt3j4Q@mail.gmail.com>
Subject: Re: "PM / wakeup: Show wakeup sources stats in sysfs" causes boot warnings
From:   Stephen Boyd <swboyd@chromium.org>
Cc:     Tony Lindgren <tony@atomide.com>, Qian Cai <cai@lca.pw>,
        Dmitry Torokhov <dmitry.torokhov@gmail.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        "Rafael J. Wysocki" <rafael@kernel.org>,
        Linux PM <linux-pm@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
To:     "Rafael J. Wysocki" <rafael@kernel.org>, Tri Vo <trong@android.com>
User-Agent: alot/0.8.1
Date:   Fri, 16 Aug 2019 07:19:35 -0700
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Quoting Rafael J. Wysocki (2019-08-16 05:17:23)
> On Wed, Aug 14, 2019 at 8:37 PM Tri Vo <trong@android.com> wrote:
> >
> > On Wed, Aug 14, 2019 at 1:40 AM Tony Lindgren <tony@atomide.com> wrote:
> > >
> > > * Stephen Boyd <swboyd@chromium.org> [691231 23:00]:
> > > > I also notice that device_set_wakeup_capable() has a check to see i=
f the
> > > > device is registered yet and it skips creating sysfs entries for the
> > > > device if it isn't created in sysfs yet. Why? Just so it can be cal=
led
> > > > before the device is created? I guess the same logic is handled by
> > > > dpm_sysfs_add() if the device is registered after calling
> > > > device_set_wakeup_*().
> > >
> > > Hmm just guessing.. It's maybe because drivers can enable and disable
> > > the wakeup capability at any point for example like driver/net drivers
> > > do based on WOL etc?
> > >
> > > > There's two approaches I see:
> > > >
> > > >       1) Do a similar check for device_set_wakeup_enable() and skip
> > > >       adding the wakeup class until dpm_sysfs_add().
> > > >
> > > >       2) Find each case where this happens and only call wakeup APIs
> > > >       on the device after the device is added.
> > > >
> > > > I guess it's better to let devices have wakeup modified on them bef=
ore
> > > > they're registered with the device core?
> > >
> > > I think we should at least initially handle case #1 above as multiple
> > > places otherwise seem to break. Then maybe we could add a warning to
> > > help fix all the #2 cases if needed?
> >
> > Makes sense. For case#1, we could also just register the wakeup source
> > without specifying the parent device if the latter hasn't been
> > registered yet. Userspace won't be able to associate a wakeup source
> > to the parent device. But I think it's a reasonable fix, assuming we
> > want to fix devices not being added before calling wakeup APIs #2.
>=20
> Well, OK
>=20
> I'm going to drop the entire series from linux-next at this point and
> let's start over.

I was going to send the first patch I floated as a more formal patch to
be applied to the PM tree. I was waiting to see if the semantics of
device_set_wakeup_*() could be clarified because I don't understand if
they're allowed to be called before device_add().

>=20
> Also note that all of this is not an issue until we start to add
> children under the device passed to device_set_wakeup_enable() and
> friends so maybe that is not a good idea after all?

My primary goal is to know what wakeup is associated with a device. If
we delay creation of the sysfs node to the time that device_add() is
called then it will allow device_set_wakeup_enable() to be called before
the device is published to userspace. Is anything wrong with that? This
seems to be the intention of the API based on the way
device_set_wakeup_capable() is written. Furthermore, if we make this
change then we don't need to fix various drivers to reorder calls to
device_set_wakeup_enable() and device_add(), so it looks like the right
approach.

I'll send the patch over the list now and let you decide. I'll also send
a patch for serio to have it operate on the device in a less racy way,
but not necessarily after the device_add() is called.