On Sat, Mar 28, 2020 at 03:06:33PM -0500, Bjorn Helgaas wrote:
> On Thu, Mar 26, 2020 at 07:35:24AM +0100, Greg KH wrote:
> > On Wed, Mar 25, 2020 at 05:10:33PM -0500, Bjorn Helgaas wrote:
> > > -static DEVICE_ATTR_WO(dev_rescan);
> > > +static struct device_attribute dev_attr_dev_rescan = __ATTR(rescan, 0200, NULL,
> > > + dev_rescan_store);
> >
> > Oops, this should just be DEVICE_ATTR(), no need for __ATTR() as this
> > isn't a kobject-only file.
> >
> > So how about:
> >
> > static DEVICE_ATTR(rescan, 0200, NULL, dev_rescan_store);
>
> I don' think DEVICE_ATTR() works in this case because it uses the
> first argument ("rescan") to build both the C symbol for the
> device_attribute struct and the sysfs filename.
>
> There are two instances in this file. The two sysfs "rescan" files
> are not a problem, but the two "dev_attr_rescan_name" C symbols *are*.
> We could resolve that by putting the bus attributes in a different
> source file than the dev attributes, but it doesn't seem worth it now.
>
> I tentatively have the patch below on pci/misc. I dropped the
> tested-by and reviewed-by because I didn't want to put words in your
> mouths :)
Ah, yeah, you are right __ATTR() is what you need to use here.
Your patch looks good:
Reviewed-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
