[PATCH 2.6.19-rc6] powerpc: fix of_device __must_check warning

[PATCH 2.6.19-rc6] powerpc: fix of_device __must_check warning

[Ryusuke Konishi]

This fixes the following compile warning:

arch/powerpc/kernel/of_device.c: In function 'of_device_register':
arch/powerpc/kernel/of_device.c:199: warning: ignoring return value of 'device_create_file', declared with attribute warn_unused_result

Signed-off-by: Ryusuke Konishi <ryusuke@osrg.net>
---
Hi,

This patch becomes useless when Benjamin's "powerpc: of_platform 
improvements" work will be done.  But, transiently, it helps to 
reduce gcc warning arising on the latest Linus's tree.

Regards,

Ryusuke

diff --git a/arch/powerpc/kernel/of_device.c b/arch/powerpc/kernel/of_device.c
index 397c83e..8831f00 100644
--- a/arch/powerpc/kernel/of_device.c
+++ b/arch/powerpc/kernel/of_device.c
@@ -196,9 +196,11 @@ int of_device_register(struct of_device
 	if (rc)
 		return rc;
 
-	device_create_file(&ofdev->dev, &dev_attr_devspec);
+	rc = device_create_file(&ofdev->dev, &dev_attr_devspec);
+	if (rc)
+		device_unregister(&ofdev->dev);
 
-	return 0;
+	return rc;
 }
 
 void of_device_unregister(struct of_device *ofdev)

Re: [PATCH 2.6.19-rc6] powerpc: fix of_device __must_check warning

[Paul Mackerras]

Ryusuke Konishi writes:

> This fixes the following compile warning:
> 
> arch/powerpc/kernel/of_device.c: In function 'of_device_register':
> arch/powerpc/kernel/of_device.c:199: warning: ignoring return value of 'device_create_file', declared with attribute warn_unused_result

Your patch also makes a policy change, though, in that without your
patch, a failure in creating the "devspec" file in sysfs doesn't
prevent of_device_register from performing its main function.  With
your patch, of_device_register will fail if we can't create the
"devspec" file.

On the whole, I am inclined to think that it is useful for
of_device_register to succeed even if the devspec file can't be
created, since nothing in the kernel relies on the devspec file being
present.

Paul.

Re: [PATCH 2.6.19-rc6] powerpc: fix of_device __must_check warning

[Ryusuke Konishi]

Hi,

Paul Mackerras wrote:
> Your patch also makes a policy change, though, in that without your
> patch, a failure in creating the "devspec" file in sysfs doesn't
> prevent of_device_register from performing its main function.  With
> your patch, of_device_register will fail if we can't create the
> "devspec" file.
>
> On the whole, I am inclined to think that it is useful for
> of_device_register to succeed even if the devspec file can't be
> created, since nothing in the kernel relies on the devspec file being
> present.
> 
> Paul.

Yes, you have a point there.

For the purpose of reference, I looked into other architectures.
Then, I found that sparc (and sparc64) had same of_device_register()
functions, and both are checking the result of device_create_file()
as follows.

in function arch/sparc/kernel/of_device.c:
>int of_device_register(struct of_device *ofdev)
>{
>        int rc;
>
>        BUG_ON(ofdev->node == NULL);
>
>        rc = device_register(&ofdev->dev);
>        if (rc)
>                return rc;
>
>        rc = device_create_file(&ofdev->dev, &dev_attr_devspec);
>        if (rc)
>                device_unregister(&ofdev->dev);
>
>        return rc;
>}

But then, I'd like to respect your point of view,
and I won't stick to this case example.

Ryusuke.