amba: tegra-ahb: mark PM functions as __maybe_unused

clang warns about an unused variable when CONFIG_PM is disabled,
since it is only referenced from an #ifdef:

drivers/amba/tegra-ahb.c:97:18: error: variable 'tegra_ahb_gizmo' is not needed and will not be emitted [-Werror,-Wunneeded-internal-declaration]

Rather than trying to get the #ifdef right, remove it and
use __maybe_unused here, which is less error prone.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

---
 drivers/amba/tegra-ahb.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

-- 
2.20.0

On Fri, Apr 26, 2019 at 04:56:03PM +0200, Arnd Bergmann wrote:
> clang warns about an unused variable when CONFIG_PM is disabled,

> since it is only referenced from an #ifdef:

> 

> drivers/amba/tegra-ahb.c:97:18: error: variable 'tegra_ahb_gizmo' is not needed and will not be emitted [-Werror,-Wunneeded-internal-declaration]

> 

> Rather than trying to get the #ifdef right, remove it and

> use __maybe_unused here, which is less error prone.

> 

> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

> ---

>  drivers/amba/tegra-ahb.c | 6 ++----

>  1 file changed, 2 insertions(+), 4 deletions(-)


Shouldn't tegra_ahb_gizmo have the same annotation, then? I do see that
it's "used" in tegra_ahb_probe() as part of an ARRAY_SIZE() expression,
but that technically doesn't mean that it would need to be emitted, but
it might be enough to shut up clang?

Or does the __maybe_unused not get propagated?

Looking at the code, I guess we could save a bit of memory if we didn't
allocate memory for the zero-length "ctx" array for !PM since that's no
longer needed. But then again, we've recently changed 32-bit Tegra to
forcefully enable PM, just like we did for 64-bit Tegra, so that's moot
anyway.

Do you want me to pick this up, or would you rather stash it into ARM
SoC directly? If the latter:

Acked-by: Thierry Reding <treding@nvidia.com>


> diff --git a/drivers/amba/tegra-ahb.c b/drivers/amba/tegra-ahb.c

> index b0b688c481e8..3751d811be39 100644

> --- a/drivers/amba/tegra-ahb.c

> +++ b/drivers/amba/tegra-ahb.c

> @@ -170,8 +170,7 @@ int tegra_ahb_enable_smmu(struct device_node *dn)

>  EXPORT_SYMBOL(tegra_ahb_enable_smmu);

>  #endif

>  

> -#ifdef CONFIG_PM

> -static int tegra_ahb_suspend(struct device *dev)

> +static int __maybe_unused tegra_ahb_suspend(struct device *dev)

>  {

>  	int i;

>  	struct tegra_ahb *ahb = dev_get_drvdata(dev);

> @@ -181,7 +180,7 @@ static int tegra_ahb_suspend(struct device *dev)

>  	return 0;

>  }

>  

> -static int tegra_ahb_resume(struct device *dev)

> +static int __maybe_unused tegra_ahb_resume(struct device *dev)

>  {

>  	int i;

>  	struct tegra_ahb *ahb = dev_get_drvdata(dev);

> @@ -190,7 +189,6 @@ static int tegra_ahb_resume(struct device *dev)

>  		gizmo_writel(ahb, ahb->ctx[i], tegra_ahb_gizmo[i]);

>  	return 0;

>  }

> -#endif

>  

>  static UNIVERSAL_DEV_PM_OPS(tegra_ahb_pm,

>  			    tegra_ahb_suspend,

> -- 

> 2.20.0

>

On Fri, Apr 26, 2019 at 5:20 PM Thierry Reding <thierry.reding@gmail.com> wrote:
>

> On Fri, Apr 26, 2019 at 04:56:03PM +0200, Arnd Bergmann wrote:

> > clang warns about an unused variable when CONFIG_PM is disabled,

> > since it is only referenced from an #ifdef:

> >

> > drivers/amba/tegra-ahb.c:97:18: error: variable 'tegra_ahb_gizmo' is not needed and will not be emitted [-Werror,-Wunneeded-internal-declaration]

> >

> > Rather than trying to get the #ifdef right, remove it and

> > use __maybe_unused here, which is less error prone.

> >

> > Signed-off-by: Arnd Bergmann <arnd@arndb.de>

> > ---

> >  drivers/amba/tegra-ahb.c | 6 ++----

> >  1 file changed, 2 insertions(+), 4 deletions(-)

>

> Shouldn't tegra_ahb_gizmo have the same annotation, then? I do see that

> it's "used" in tegra_ahb_probe() as part of an ARRAY_SIZE() expression,

> but that technically doesn't mean that it would need to be emitted, but

> it might be enough to shut up clang?


It would be enough to annotated tegra_ahb_gizmo as __maybe_unused,
but I don't like the idea of using both __maybe_unused and #ifdef to
deal with the CONFIG_PM issue, so I was try to make this more consistent.

> Or does the __maybe_unused not get propagated?

>

> Looking at the code, I guess we could save a bit of memory if we didn't

> allocate memory for the zero-length "ctx" array for !PM since that's no

> longer needed. But then again, we've recently changed 32-bit Tegra to

> forcefully enable PM, just like we did for 64-bit Tegra, so that's moot

> anyway.


Right

> Do you want me to pick this up, or would you rather stash it into ARM

> SoC directly? If the latter:


Olof is currently doing the merges and I forgot to Cc him, so I'd prefer
if you could forward the patch to him or include it in a later pull
request.

        Arnd