Re: Calling device_init_wakeup() on driver removal

[Guenter Roeck]

On 01/16/2017 01:50 PM, Rafael J. Wysocki wrote:
> On Sun, Jan 15, 2017 at 4:23 PM, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > On 01/15/2017 06:49 AM, Rafael J. Wysocki wrote:
> > > On Saturday, January 14, 2017 08:46:05 PM Guenter Roeck wrote:
> > > > Hi folks,
> > >
> > >
> > > Hi,
> > >
> > > > while looking through driver initialization and removal functions, I
> > > > noticed that many drivers
> > > > call device_init_wakeup(dev, false) in the removal function. Given that
> > > > the driver is about
> > > > to be removed, that doesn't make much sense to me, especially since
> > > > device_wakeup_disable()
> > > > is called from device_pm_remove() anyway.
> > > >
> > > > Is it safe to assume that all those calls can be removed, or is there a
> > > > possible reason for
> > > > keeping them around ?
> > >
> > >
> > > Removing them automatically might break things, because
> > > device_init_wakeup(dev, false)
> > > also clears the power.can_wakeup flag and removes the "wakeup" attribute
> > > from sysfs.
> >
> > I had the same concern, but I concluded that the wakeup attribute should be
> > removed
> > automatically, since it is added with sysfs_merge_group(), and the matching
> > unmerge call
> > is also made in dpm_sysfs_remove(). power.can_wakeup is part of the device
> > structure,
> > which is in the process of being removed, so I am not sure I understand how
> > that can be
> > problematic.
>
> That's when the device goes away, but ->remove() is invoked when the
> driver goes away too and that's the more common case (by far IMO).
Confused ... ah, no, I am using sloppy terminology again. Sorry for
that. When I talked about "driver remove function, I meant the remove
function defined in 'struct platform_driver' or 'struct i2c_driver',
which is the device remove function, not the driver remove function.
I am not talking about functions like platform_driver_unregister()
or i2c_del_driver().

> > > I guess they could be removed safely in the majority of cases, though.
> >
> >
> > How would one decide if it is needed ? I see some drivers call it on remove,
> > but others
> > don't. I don't see a clear pattern; unless I am missing something, it seems
> > to be
> > more or less random.
>
> My guess would be that drivers calling it don't want the wakeup
> capability to be exposed after they have gone away.  Most likely these
> are the ones that expose the wakeup attribute on probe.
>
> IMO the rule should be that whoever exposes the wakeup attribute one
> init, should also hide it on cleanup.
The probe function is a device probe function. Are you saying that calling
device_init_wakeup(dev, false) in the device remove function is necessary
if the probe function called device_init_wakeup(dev, true) ?

That is exactly what I am wondering. Because, if so, several drivers violate
that. Also, I don't really see the point because the device structure is in
the process of being removed (and I don't see anything in the code that isn't
already cleaned up).

Thanks,
Guenter