Prev: ucb1400_core: avoid calling platform_device_put() twice in ucb1400_core_probe error path
Next: [PATCH 0/2] perf: symbol offset breakage with separated debug
From: Valdis.Kletnieks on 3 Aug 2010 07:30
On Mon, 02 Aug 2010 21:26:28 PDT, "Justin P. Mattock" said:
> diff --git a/drivers/usb/core/sysfs.c b/drivers/usb/core/sysfs.c

> if (alt->string)
> - retval = device_create_file(&intf->dev, &dev_attr_interface);
> + device_create_file(&intf->dev, &dev_attr_interface);
> intf->sysfs_files_created = 1;
> return 0;

What should the code do if device_create_file() manages to fail? Yes, ignoring
the return value is one option, but is it the best one? 'return ret;' might be
another one. Somebody who understands this code and has more caffeine than me
should look this over.

(Nothing personal Justin - it's just my opinion that *anytime* we have a patch
that remove a check for a return code, it needs to justify that ignoring the
return code is appropriate).

From: Valdis.Kletnieks on 3 Aug 2010 11:20
On Tue, 03 Aug 2010 10:29:52 EDT, Alan Stern said:
> On Tue, 3 Aug 2010 Valdis.Kletnieks(a)vt.edu wrote:
>
> > On Mon, 02 Aug 2010 21:26:28 PDT, "Justin P. Mattock" said:
> > > diff --git a/drivers/usb/core/sysfs.c b/drivers/usb/core/sysfs.c
> >
> > > if (alt->string)
> > > - retval = device_create_file(&intf->dev, &dev_attr_interface);
> > > + device_create_file(&intf->dev, &dev_attr_interface);
> > > intf->sysfs_files_created = 1;
> > > return 0;
>
> Justin, did you try compiling your new code? Those unused values are
> there because device_create_file is declared as __must_check.
>
> > What should the code do if device_create_file() manages to fail? Yes, ignoring
> > the return value is one option, but is it the best one? 'return ret;' might be
> > another one. Somebody who understands this code and has more caffeine than me
> > should look this over.
>
> Failure to create a file in sysfs is almost never fatal and usually not
> even dangerous. Ignoring the error is generally better than failing
> the entire operation.

Then why the __must_check attribute if it's usually ignorable? I thought
that was reserved for functions that you damned sight better well check
for errors because bad things are afoot otherwise?
 | 
Pages: 1
Prev: ucb1400_core: avoid calling platform_device_put() twice in ucb1400_core_probe error path
Next: [PATCH 0/2] perf: symbol offset breakage with separated debug