From: Tri Vo <trong@android.com>
To: Stephen Boyd <swboyd@chromium.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>,
LKML <linux-kernel@vger.kernel.org>,
Linux PM <linux-pm@vger.kernel.org>, Qian Cai <cai@lca.pw>
Subject: Re: [PATCH] PM / wakeup: Register wakeup class kobj after device is added
Date: Fri, 16 Aug 2019 15:38:07 -0700 [thread overview]
Message-ID: <CANA+-vDZqXS-yac9Q_K54oVrqZLtJojBSKa=ToJ9ZkfYHRcqbQ@mail.gmail.com> (raw)
In-Reply-To: <5d57242c.1c69fb81.bba86.14f6@mx.google.com>
On Fri, Aug 16, 2019 at 2:46 PM Stephen Boyd <swboyd@chromium.org> wrote:
>
> Quoting Tri Vo (2019-08-16 14:27:35)
> > On Fri, Aug 16, 2019 at 7:56 AM Stephen Boyd <swboyd@chromium.org> wrote:
> > > diff --git a/drivers/base/power/sysfs.c b/drivers/base/power/sysfs.c
> > > index 1b9c281cbe41..27ee00f50bd7 100644
> > > --- a/drivers/base/power/sysfs.c
> > > +++ b/drivers/base/power/sysfs.c
> > > @@ -5,6 +5,7 @@
> > > #include <linux/export.h>
> > > #include <linux/pm_qos.h>
> > > #include <linux/pm_runtime.h>
> > > +#include <linux/pm_wakeup.h>
> > > #include <linux/atomic.h>
> > > #include <linux/jiffies.h>
> > > #include "power.h"
> > > @@ -661,14 +662,21 @@ int dpm_sysfs_add(struct device *dev)
> > > if (rc)
> > > goto err_runtime;
> > > }
> > > + if (dev->power.wakeup) {
> >
> > This conditional checks for the situation when wakeup source
> > registration have been previously attempted, but failed at
> > wakeup_source_sysfs_add(). My concern is that it's not easy to
> > understand what this check does without knowing exactly what
> > device_wakeup_enable() does to dev->power.wakeup before we reach this
> > point.
>
> Oh, actually this is wrong. It should be a check for
> dev->power.wakeup->dev being non-NULL. That's the variable that's set by
> wakeup_source_sysfs_add() upon success. So I should make it:
>
> if (dev->power.wakeup && !dev->power.wakeup->dev)
Oh ok, this makes more sense now :)
>
> And there's the problem that CONFIG_PM_SLEEP could be unset. Let me fix
> it up with a new inline function like device_has_wakeup_dev().
>
> >
> > > + rc = wakeup_source_sysfs_add(dev, dev->power.wakeup);
> > > + if (rc)
> > > + goto err_wakeup;
> > > + }
> > > if (dev->power.set_latency_tolerance) {
> > > rc = sysfs_merge_group(&dev->kobj,
> > > &pm_qos_latency_tolerance_attr_group);
> > > if (rc)
> > > - goto err_wakeup;
> > > + goto err_wakeup_source;
> > > }
> > > return 0;
> > >
> > > + err_wakeup_source:
> > > + wakeup_source_sysfs_remove(dev->power.wakeup);
> > > err_wakeup:
> > > sysfs_unmerge_group(&dev->kobj, &pm_wakeup_attr_group);
> > > err_runtime:
> > > diff --git a/drivers/base/power/wakeup.c b/drivers/base/power/wakeup.c
> > > index f7925820b5ca..5817b51d2b15 100644
> > > --- a/drivers/base/power/wakeup.c
> > > +++ b/drivers/base/power/wakeup.c
> > > @@ -220,10 +220,12 @@ struct wakeup_source *wakeup_source_register(struct device *dev,
> > >
> > > ws = wakeup_source_create(name);
> > > if (ws) {
> > > - ret = wakeup_source_sysfs_add(dev, ws);
> > > - if (ret) {
> > > - wakeup_source_free(ws);
> > > - return NULL;
> > > + if (!dev || device_is_registered(dev)) {
> >
> > Is there a possible race condition here? If dev->power.wakeup check in
> > dpm_sysfs_add() is done at the same time as device_is_registered(dev)
> > check here, then wakeup_source_sysfs_add() won't ever be called?
>
> The same race exists for device_set_wakeup_capable() so I didn't bother
> to try to avoid it. I suppose wakeup_source_sysfs_add() could run
> completely, allocate the device and set the name, etc., but not call
> device_add() and then we can set ws->dev and call device_add() under a
> mutex so that we keep a very small window where the wakeup class is
> published to sysfs. Or just throw a big mutex around the whole wakeup
> class creation path so that there isn't a chance of a race. But really,
> is anyone going to call device_set_wakeup_*() on a device that is also
> being added to the system? Seems unlikely.
True. I don't have a strong opinion.
>
> >
> > > + ret = wakeup_source_sysfs_add(dev, ws);
> > > + if (ret) {
> > > + wakeup_source_free(ws);
next prev parent reply other threads:[~2019-08-16 22:38 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-16 14:56 [PATCH] PM / wakeup: Register wakeup class kobj after device is added Stephen Boyd
2019-08-16 21:27 ` Tri Vo
2019-08-16 21:46 ` Stephen Boyd
2019-08-16 22:38 ` Tri Vo [this message]
2019-08-19 9:36 ` Rafael J. Wysocki
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CANA+-vDZqXS-yac9Q_K54oVrqZLtJojBSKa=ToJ9ZkfYHRcqbQ@mail.gmail.com' \
--to=trong@android.com \
--cc=cai@lca.pw \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pm@vger.kernel.org \
--cc=rjw@rjwysocki.net \
--cc=swboyd@chromium.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).