On Sunday, July 08, 2012, Guenter Roeck wrote:
> On Sun, Jul 08, 2012 at 09:48:15PM +0200, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <rjw@xxxxxxx>
> >
> Hi Rafael,
>
> > Make the Exynos4 TMU driver define its PM callbacks through
> > a struct dev_pm_ops object rather than by using legacy PM hooks
> > in struct platform_driver.
> >
> > Signed-off-by: Rafael J. Wysocki <rjw@xxxxxxx>
> > ---
> > drivers/hwmon/exynos4_tmu.c | 18 +++++++++++-------
> > 1 file changed, 11 insertions(+), 7 deletions(-)
> >
> > Index: linux/drivers/hwmon/exynos4_tmu.c
> > ===================================================================
> > --- linux.orig/drivers/hwmon/exynos4_tmu.c
> > +++ linux/drivers/hwmon/exynos4_tmu.c
> > @@ -476,34 +476,38 @@ static int __devexit exynos4_tmu_remove(
> > }
> >
> > #ifdef CONFIG_PM
> > -static int exynos4_tmu_suspend(struct platform_device *pdev, pm_message_t state)
> > +static int exynos4_tmu_suspend(struct device *dev)
> > {
> > - exynos4_tmu_control(pdev, false);
> > + exynos4_tmu_control(to_platform_device(dev), false);
> >
> > return 0;
> > }
> >
> > -static int exynos4_tmu_resume(struct platform_device *pdev)
> > +static int exynos4_tmu_resume(struct device *dev)
> > {
> > + struct platform_device *pdev = to_platform_device(dev);
> > +
> > exynos4_tmu_initialize(pdev);
> > exynos4_tmu_control(pdev, true);
> >
> > return 0;
> > }
> > +
> > +static SIMPLE_DEV_PM_OPS(exynos4_tmu_pm,
> > + exynos4_tmu_suspend, exynos4_tmu_resume);
> > +#define EXYNOS4_TMU_PM (&exynos4_tmu_pm)
> > #else
> > -#define exynos4_tmu_suspend NULL
> > -#define exynos4_tmu_resume NULL
> > +#define EXYNOS4_TMU_PM NULL
> > #endif
>
> Actually, looking into other drivers, the common approach seems to be to declare
>
> static SIMPLE_DEV_PM_OPS(exynos4_tmu_pm,
> exynos4_tmu_suspend, exynos4_tmu_resume);
>
> outside the #ifdef code and then just assign
>
> .pm = &exynos4_tmu_pm;
>
> unconditionally.
>
> That seems to be a much simpler solution. Any special reason for not
> implementing it this way ? Same question applies to the other patches in the
> series.
In fact, there are two ways. One of them is what you said, the other is to
put SIMPLE_DEV_PM_OPS() under the #ifdef like I did. I'm not sure which one
is prevalent, but what I did has the advatnage that the kernel data will
be slightly smaller if CONFIG_PM is unset.
Thanks,
Rafael
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
