tty: serial: qcom_geni: avoid duplicate struct member init

From: Arnd Bergmann <arnd@arndb.de>

When -Woverride-init is enabled in a build, gcc points out that
qcom_geni_serial_pm_ops contains conflicting initializers:

drivers/tty/serial/qcom_geni_serial.c:1586:20: error: initialized field overwritten [-Werror=override-init]
 1586 |         .restore = qcom_geni_serial_sys_hib_resume,
      |                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/tty/serial/qcom_geni_serial.c:1586:20: note: (near initialization for 'qcom_geni_serial_pm_ops.restore')
drivers/tty/serial/qcom_geni_serial.c:1587:17: error: initialized field overwritten [-Werror=override-init]
 1587 |         .thaw = qcom_geni_serial_sys_hib_resume,
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Open-code the initializers with the version that was already used,
and use the pm_sleep_ptr() method to deal with unused ones,
in place of the __maybe_unused annotation.

Fixes: 35781d8356a2 ("tty: serial: qcom-geni-serial: Add support for Hibernation feature")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/tty/serial/qcom_geni_serial.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

On Thu, Dec 15, 2022, at 21:46, Doug Anderson wrote:
> On Thu, Dec 15, 2022 at 8:55 AM Arnd Bergmann <arnd@kernel.org> wrote:

>> index b487823f0e61..03dda47184d9 100644
>> --- a/drivers/tty/serial/qcom_geni_serial.c
>> +++ b/drivers/tty/serial/qcom_geni_serial.c
>> @@ -1516,7 +1516,7 @@ static int qcom_geni_serial_remove(struct platform_device *pdev)
>>         return 0;
>>  }
>>
>> -static int __maybe_unused qcom_geni_serial_sys_suspend(struct device *dev)
>> +static int qcom_geni_serial_sys_suspend(struct device *dev)
>
> Officially the removal of "__maybe_unused" could be a totally
> different patch, right? SET_SYSTEM_SLEEP_PM_OPS() already eventually
> used pm_sleep_ptr() even without your change, so the removal of these
> tags is unrelated to the rest of your change, right?

It's a little more complicated: SYSTEM_SLEEP_PM_OPS() uses pm_sleep_ptr()
to avoid the need for a __maybe_unused(). The depreacated
SET_SYSTEM_SLEEP_PM_OPS() is based on SYSTEM_SLEEP_PM_OPS() these days,
but still retains the old semantics of using an empty definition
without CONFIG_PM_SLEEP, so it still leaves the function unused as
far as gcc is concerned.

There could be an intermediate step of open-coding the
SET_SYSTEM_SLEEP_PM_OPS(), but that would result in the rather
silly

 static const struct dev_pm_ops qcom_geni_serial_pm_ops = {
#ifdef CONFIG_PM_SLEEP
       .suspend = qcom_geni_serial_sys_suspend,
       .resume = qcom_geni_serial_sys_resume,
       .freeze = qcom_geni_serial_sys_suspend,
       .poweroff = qcom_geni_serial_sys_suspend,
#endif
       .restore = qcom_geni_serial_sys_hib_resume,
       .thaw = qcom_geni_serial_sys_hib_resume,
}

which makes no sense to me, as I think you either want
all the members or none of them.

>>  static const struct dev_pm_ops qcom_geni_serial_pm_ops = {
>> -       SET_SYSTEM_SLEEP_PM_OPS(qcom_geni_serial_sys_suspend,
>> -                                       qcom_geni_serial_sys_resume)
>> -       .restore = qcom_geni_serial_sys_hib_resume,
>> -       .thaw = qcom_geni_serial_sys_hib_resume,
>> +       .suspend = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
>> +       .resume = pm_sleep_ptr(qcom_geni_serial_sys_resume),
>> +       .freeze = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
>> +       .poweroff = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
>> +       .restore = pm_sleep_ptr(qcom_geni_serial_sys_hib_resume),
>> +       .thaw = pm_sleep_ptr(qcom_geni_serial_sys_hib_resume),
>
> Personally, the order you listed them is less intuitive than the order
> that SET_SYSTEM_SLEEP_PM_OPS() lists functions. IMO it's better to
> consistently alternate matching suspend/resume functions. ;-)

Makes sense. I kept the order that we already had here, but
I could redo this patch if anyone cares.

> Both of those are nits, so I'm also fine with:
>
> Reviewed-by: Douglas Anderson <dianders@chromium.org>

Thanks,

     Arnd

Hi,

On Fri, Dec 16, 2022 at 2:41 AM Arnd Bergmann <arnd@arndb.de> wrote:
>
> On Thu, Dec 15, 2022, at 21:46, Doug Anderson wrote:
> > On Thu, Dec 15, 2022 at 8:55 AM Arnd Bergmann <arnd@kernel.org> wrote:
>
> >> index b487823f0e61..03dda47184d9 100644
> >> --- a/drivers/tty/serial/qcom_geni_serial.c
> >> +++ b/drivers/tty/serial/qcom_geni_serial.c
> >> @@ -1516,7 +1516,7 @@ static int qcom_geni_serial_remove(struct platform_device *pdev)
> >>         return 0;
> >>  }
> >>
> >> -static int __maybe_unused qcom_geni_serial_sys_suspend(struct device *dev)
> >> +static int qcom_geni_serial_sys_suspend(struct device *dev)
> >
> > Officially the removal of "__maybe_unused" could be a totally
> > different patch, right? SET_SYSTEM_SLEEP_PM_OPS() already eventually
> > used pm_sleep_ptr() even without your change, so the removal of these
> > tags is unrelated to the rest of your change, right?
>
> It's a little more complicated: SYSTEM_SLEEP_PM_OPS() uses pm_sleep_ptr()
> to avoid the need for a __maybe_unused(). The depreacated
> SET_SYSTEM_SLEEP_PM_OPS() is based on SYSTEM_SLEEP_PM_OPS() these days,
> but still retains the old semantics of using an empty definition
> without CONFIG_PM_SLEEP, so it still leaves the function unused as
> far as gcc is concerned.
>
> There could be an intermediate step of open-coding the
> SET_SYSTEM_SLEEP_PM_OPS(), but that would result in the rather
> silly
>
>  static const struct dev_pm_ops qcom_geni_serial_pm_ops = {
> #ifdef CONFIG_PM_SLEEP
>        .suspend = qcom_geni_serial_sys_suspend,
>        .resume = qcom_geni_serial_sys_resume,
>        .freeze = qcom_geni_serial_sys_suspend,
>        .poweroff = qcom_geni_serial_sys_suspend,
> #endif
>        .restore = qcom_geni_serial_sys_hib_resume,
>        .thaw = qcom_geni_serial_sys_hib_resume,
> }
>
> which makes no sense to me, as I think you either want
> all the members or none of them.

Ah, I guess I didn't trace through all the similarly named macros
quite correctly. ;-) Thanks for explaining.


> >>  static const struct dev_pm_ops qcom_geni_serial_pm_ops = {
> >> -       SET_SYSTEM_SLEEP_PM_OPS(qcom_geni_serial_sys_suspend,
> >> -                                       qcom_geni_serial_sys_resume)
> >> -       .restore = qcom_geni_serial_sys_hib_resume,
> >> -       .thaw = qcom_geni_serial_sys_hib_resume,
> >> +       .suspend = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
> >> +       .resume = pm_sleep_ptr(qcom_geni_serial_sys_resume),
> >> +       .freeze = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
> >> +       .poweroff = pm_sleep_ptr(qcom_geni_serial_sys_suspend),
> >> +       .restore = pm_sleep_ptr(qcom_geni_serial_sys_hib_resume),
> >> +       .thaw = pm_sleep_ptr(qcom_geni_serial_sys_hib_resume),
> >
> > Personally, the order you listed them is less intuitive than the order
> > that SET_SYSTEM_SLEEP_PM_OPS() lists functions. IMO it's better to
> > consistently alternate matching suspend/resume functions. ;-)
>
> Makes sense. I kept the order that we already had here, but
> I could redo this patch if anyone cares.

I wouldn't worry about it.

-Doug