Re: [PATCH 2/2] nvmem: expose NVMEM cells in sysfs

[Greg Kroah-Hartman]

On Mon, Dec 20, 2021 at 09:39:43PM +0100, Rafał Miłecki wrote:
> Hi Greg,
> 
> On 20.12.2021 09:00, Greg Kroah-Hartman wrote:
> > On Mon, Dec 20, 2021 at 07:47:30AM +0100, Rafał Miłecki wrote:
> > >   static void nvmem_cell_entry_add(struct nvmem_cell_entry *cell)
> > >   {
> > > +	struct device *dev = &cell->nvmem->dev;
> > > +	int err;
> > > +
> > >   	mutex_lock(&nvmem_mutex);
> > >   	list_add_tail(&cell->node, &cell->nvmem->cells);
> > >   	mutex_unlock(&nvmem_mutex);
> > > +
> > > +	sysfs_attr_init(&cell->battr.attr);
> > > +	cell->battr.attr.name = cell->name;
> > > +	cell->battr.attr.mode = 0400;
> > > +	cell->battr.read = nvmem_cell_attr_read;
> > > +	err = sysfs_add_bin_file_to_group(&dev->kobj, &cell->battr,
> > > +					  nvmem_cells_group.name);
> > 
> > Why not just use the is_bin_visible attribute instead to determine if
> > the attribute should be shown or not instead of having to add it
> > after-the-fact which will race with userspace and loose?
> 
> I'm sorry I really don't see how you suggest to get it done.
> 
> I can use .is_bin_visible() callback indeed to respect nvmem->root_only.

Great.

> I don't understand addig-after-the-fact part. How is .is_bin_visible()
> related to adding attributes for newly created cells?

You are adding a sysfs attribute to a device that is already registered
in the driver core, and so the creation of that attribute is never seen
by userspace.  The attribute needs to be attached to the device _BEFORE_
it is registered.

Also, huge hint, if a driver has to call as sysfs_*() call, something is
wrong.

> Do you mean I can
> avoid calling sysfs_add_bin_file_to_group()?

Yes.

> Do you recall any existing example of such solution?

Loads.

Just add this attribute group to your driver as a default attribute
group and the driver core will create it for you if needed.

Or if you always need it, no need to mess sith is_bin_visible() at all,
I can't really understand what you are trying to do here at all.

thanks,

greg k-h