On Sat, Apr 18 2020, Alan Stern wrote:

> On Sat, 18 Apr 2020, Matthew Wilcox wrote:
>
>> On Sat, Apr 18, 2020 at 11:41:07AM -0700, Randy Dunlap wrote:
>> > +++ linux-next-20200327/drivers/usb/core/sysfs.c
>> > @@ -1263,7 +1263,7 @@ void usb_create_sysfs_intf_files(struct
>> >  	if (!alt->string && !(udev->quirks & USB_QUIRK_CONFIG_INTF_STRINGS))
>> >  		alt->string = usb_cache_string(udev, alt->desc.iInterface);
>> >  	if (alt->string && device_create_file(&intf->dev, &dev_attr_interface))
>> > -		;	/* We don't actually care if the function fails. */
>> > +		do_empty(); /* We don't actually care if the function fails. */
>> >  	intf->sysfs_files_created = 1;
>> >  }
>> 
>> Why not just?
>> 
>> +	if (alt->string)
>> +		device_create_file(&intf->dev, &dev_attr_interface);
>
> This is another __must_check function call.
>
> The reason we don't care if the call fails is because the file
> being created holds the USB interface string descriptor, something
> which is purely informational and hardly ever gets set (and no doubt
> gets used even less often).
>
> Is this another situation where the comment should be expanded and the 
> code modified to include a useless test and cast-to-void?
>
> Or should device_create_file() not be __must_check after all?

One approach to dealing with __must_check function that you don't want
to check is to cause failure to call
   pr_debug("usb: interface descriptor file not created");
or similar.  It silences the compiler, serves as documentation, and
creates a message that is almost certainly never seen.

This is what I did in drivers/md/md.c...

	if (mddev->kobj.sd &&
	    sysfs_create_group(&mddev->kobj, &md_bitmap_group))
		pr_debug("pointless warning\n");

(I give better warnings elsewhere - I must have run out of patience by
 this point).

NeilBrown