2024-03-01 20:19:30

by Anand Moon

[permalink] [raw]
Subject: [PATCH v1 2/4] usb: ehci-exynos: Switch from CONFIG_PM guards to pm_ptr()

Use the new PM macros for the suspend and resume functions to be
automatically dropped by the compiler when CONFIG_PM are disabled,
without having to use #ifdef guards. If CONFIG_PM unused,
they will simply be discarded by the compiler.

Use RUNTIME_PM_OPS runtime macro for suspend/resume function.

Signed-off-by: Anand Moon <[email protected]>
---
drivers/usb/host/ehci-exynos.c | 10 ++--------
1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/drivers/usb/host/ehci-exynos.c b/drivers/usb/host/ehci-exynos.c
index 05aa3d9c2a3b..4676f45655cd 100644
--- a/drivers/usb/host/ehci-exynos.c
+++ b/drivers/usb/host/ehci-exynos.c
@@ -234,7 +234,6 @@ static void exynos_ehci_remove(struct platform_device *pdev)
usb_put_hcd(hcd);
}

-#ifdef CONFIG_PM
static int exynos_ehci_suspend(struct device *dev)
{
struct usb_hcd *hcd = dev_get_drvdata(dev);
@@ -268,14 +267,9 @@ static int exynos_ehci_resume(struct device *dev)
ehci_resume(hcd, false);
return 0;
}
-#else
-#define exynos_ehci_suspend NULL
-#define exynos_ehci_resume NULL
-#endif

static const struct dev_pm_ops exynos_ehci_pm_ops = {
- .suspend = exynos_ehci_suspend,
- .resume = exynos_ehci_resume,
+ RUNTIME_PM_OPS(exynos_ehci_suspend, exynos_ehci_resume, NULL)
};

#ifdef CONFIG_OF
@@ -292,7 +286,7 @@ static struct platform_driver exynos_ehci_driver = {
.shutdown = usb_hcd_platform_shutdown,
.driver = {
.name = "exynos-ehci",
- .pm = &exynos_ehci_pm_ops,
+ .pm = pm_ptr(&exynos_ehci_pm_ops),
.of_match_table = of_match_ptr(exynos_ehci_match),
}
};
--
2.43.0



2024-03-01 20:28:54

by Alan Stern

[permalink] [raw]
Subject: Re: [PATCH v1 2/4] usb: ehci-exynos: Switch from CONFIG_PM guards to pm_ptr()

On Sat, Mar 02, 2024 at 01:08:09AM +0530, Anand Moon wrote:
> Use the new PM macros for the suspend and resume functions to be
> automatically dropped by the compiler when CONFIG_PM are disabled,
> without having to use #ifdef guards. If CONFIG_PM unused,
> they will simply be discarded by the compiler.
>
> Use RUNTIME_PM_OPS runtime macro for suspend/resume function.
>
> Signed-off-by: Anand Moon <[email protected]>
> ---
> drivers/usb/host/ehci-exynos.c | 10 ++--------
> 1 file changed, 2 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/usb/host/ehci-exynos.c b/drivers/usb/host/ehci-exynos.c
> index 05aa3d9c2a3b..4676f45655cd 100644
> --- a/drivers/usb/host/ehci-exynos.c
> +++ b/drivers/usb/host/ehci-exynos.c
> @@ -234,7 +234,6 @@ static void exynos_ehci_remove(struct platform_device *pdev)
> usb_put_hcd(hcd);
> }
>
> -#ifdef CONFIG_PM
> static int exynos_ehci_suspend(struct device *dev)
> {
> struct usb_hcd *hcd = dev_get_drvdata(dev);
> @@ -268,14 +267,9 @@ static int exynos_ehci_resume(struct device *dev)
> ehci_resume(hcd, false);
> return 0;
> }
> -#else
> -#define exynos_ehci_suspend NULL
> -#define exynos_ehci_resume NULL
> -#endif

Doesn't this now generate warnings about functions being defined but not
used when you build with CONFIG_PM disabled?

Alan Stern

>
> static const struct dev_pm_ops exynos_ehci_pm_ops = {
> - .suspend = exynos_ehci_suspend,
> - .resume = exynos_ehci_resume,
> + RUNTIME_PM_OPS(exynos_ehci_suspend, exynos_ehci_resume, NULL)
> };
>
> #ifdef CONFIG_OF
> @@ -292,7 +286,7 @@ static struct platform_driver exynos_ehci_driver = {
> .shutdown = usb_hcd_platform_shutdown,
> .driver = {
> .name = "exynos-ehci",
> - .pm = &exynos_ehci_pm_ops,
> + .pm = pm_ptr(&exynos_ehci_pm_ops),
> .of_match_table = of_match_ptr(exynos_ehci_match),
> }
> };
> --
> 2.43.0
>

2024-03-02 15:42:19

by Anand Moon

[permalink] [raw]
Subject: Re: [PATCH v1 2/4] usb: ehci-exynos: Switch from CONFIG_PM guards to pm_ptr()

Hi Alan,

On Sat, 2 Mar 2024 at 01:58, Alan Stern <[email protected]> wrote:
>
> On Sat, Mar 02, 2024 at 01:08:09AM +0530, Anand Moon wrote:
> > Use the new PM macros for the suspend and resume functions to be
> > automatically dropped by the compiler when CONFIG_PM are disabled,
> > without having to use #ifdef guards. If CONFIG_PM unused,
> > they will simply be discarded by the compiler.
> >
> > Use RUNTIME_PM_OPS runtime macro for suspend/resume function.
> >
> > Signed-off-by: Anand Moon <[email protected]>
> > ---
> > drivers/usb/host/ehci-exynos.c | 10 ++--------
> > 1 file changed, 2 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/usb/host/ehci-exynos.c b/drivers/usb/host/ehci-exynos.c
> > index 05aa3d9c2a3b..4676f45655cd 100644
> > --- a/drivers/usb/host/ehci-exynos.c
> > +++ b/drivers/usb/host/ehci-exynos.c
> > @@ -234,7 +234,6 @@ static void exynos_ehci_remove(struct platform_device *pdev)
> > usb_put_hcd(hcd);
> > }
> >
> > -#ifdef CONFIG_PM
> > static int exynos_ehci_suspend(struct device *dev)
> > {
> > struct usb_hcd *hcd = dev_get_drvdata(dev);
> > @@ -268,14 +267,9 @@ static int exynos_ehci_resume(struct device *dev)
> > ehci_resume(hcd, false);
> > return 0;
> > }
> > -#else
> > -#define exynos_ehci_suspend NULL
> > -#define exynos_ehci_resume NULL
> > -#endif
>
> Doesn't this now generate warnings about functions being defined but not
> used when you build with CONFIG_PM disabled?
>
Yes I have tried compile the kernel with disable CONFIG_PM=n and
CONFIG_PM_SLEEP=n
But it's getting selected by default.

Also compiled with W=1 and found no warning with these patches.
To be safe I will add __maybe_unused to suspend / resume functions in
the next version.

diff --git a/arch/arm/configs/exynos_defconfig
b/arch/arm/configs/exynos_defconfig
index c98d5ff8a1ed..e96f5c3bf8c1 100644
--- a/arch/arm/configs/exynos_defconfig
+++ b/arch/arm/configs/exynos_defconfig
@@ -29,8 +29,10 @@ CONFIG_ARM_EXYNOS_CPUIDLE=y
CONFIG_VFP=y
CONFIG_NEON=y
CONFIG_KERNEL_MODE_NEON=y
-CONFIG_PM_DEBUG=y
-CONFIG_PM_ADVANCED_DEBUG=y
+# CONFIG_PM_SLEEP is not set
+# CONFIG_PM is not set
+# CONFIG_PM_DEBUG is not set
+# CONFIG_PM_ADVANCED_DEBUG is not set
CONFIG_ENERGY_MODEL=y
CONFIG_KALLSYMS_ALL=y
CONFIG_MODULES=y

> Alan Stern

Thanks

-Anand

2024-03-04 09:21:57

by Johan Hovold

[permalink] [raw]
Subject: Re: [PATCH v1 2/4] usb: ehci-exynos: Switch from CONFIG_PM guards to pm_ptr()

On Sat, Mar 02, 2024 at 01:08:09AM +0530, Anand Moon wrote:
> Use the new PM macros for the suspend and resume functions to be
> automatically dropped by the compiler when CONFIG_PM are disabled,
> without having to use #ifdef guards. If CONFIG_PM unused,
> they will simply be discarded by the compiler.
>
> Use RUNTIME_PM_OPS runtime macro for suspend/resume function.
>
> Signed-off-by: Anand Moon <[email protected]>
> ---
> drivers/usb/host/ehci-exynos.c | 10 ++--------
> 1 file changed, 2 insertions(+), 8 deletions(-)

> static const struct dev_pm_ops exynos_ehci_pm_ops = {
> - .suspend = exynos_ehci_suspend,
> - .resume = exynos_ehci_resume,
> + RUNTIME_PM_OPS(exynos_ehci_suspend, exynos_ehci_resume, NULL)
> };

This is also broken and clearly not tested. See the definition of
RUNTIME_PM_OPS() which sets the runtime pm callbacks, not the suspend
ones:

#define RUNTIME_PM_OPS(suspend_fn, resume_fn, idle_fn) \
.runtime_suspend = suspend_fn, \
.runtime_resume = resume_fn, \
.runtime_idle = idle_fn,

Johan

2024-03-04 10:26:56

by Anand Moon

[permalink] [raw]
Subject: Re: [PATCH v1 2/4] usb: ehci-exynos: Switch from CONFIG_PM guards to pm_ptr()

Hi Johan,

On Mon, 4 Mar 2024 at 14:51, Johan Hovold <[email protected]> wrote:
>
> On Sat, Mar 02, 2024 at 01:08:09AM +0530, Anand Moon wrote:
> > Use the new PM macros for the suspend and resume functions to be
> > automatically dropped by the compiler when CONFIG_PM are disabled,
> > without having to use #ifdef guards. If CONFIG_PM unused,
> > they will simply be discarded by the compiler.
> >
> > Use RUNTIME_PM_OPS runtime macro for suspend/resume function.
> >
> > Signed-off-by: Anand Moon <[email protected]>
> > ---
> > drivers/usb/host/ehci-exynos.c | 10 ++--------
> > 1 file changed, 2 insertions(+), 8 deletions(-)
>
> > static const struct dev_pm_ops exynos_ehci_pm_ops = {
> > - .suspend = exynos_ehci_suspend,
> > - .resume = exynos_ehci_resume,
> > + RUNTIME_PM_OPS(exynos_ehci_suspend, exynos_ehci_resume, NULL)
> > };
>
> This is also broken and clearly not tested. See the definition of
> RUNTIME_PM_OPS() which sets the runtime pm callbacks, not the suspend
> ones:
>
> #define RUNTIME_PM_OPS(suspend_fn, resume_fn, idle_fn) \
> .runtime_suspend = suspend_fn, \
> .runtime_resume = resume_fn, \
> .runtime_idle = idle_fn,
>
> Johan

Ok, I will drop these changes.

Thanks.
-Anand