Re: [PATCH v2] leds: class: Use firmware nodes for device lookup

[Tommaso Merciai]

On Wed, Jun 10, 2026 at 05:05:41PM +0200, Alban Bedel wrote:
> On Wed, 10 Jun 2026 16:08:49 +0200
> Tommaso Merciai <tommaso.merciai.xr@xxxxxxxxxxxxxx> wrote:
> 
> > Hi Alban,
> > Thanks for your comments.
> > 
> > On Tue, Jun 09, 2026 at 06:54:29PM +0200, Alban Bedel wrote:
> > > On Tue, 9 Jun 2026 16:46:36 +0200
> > > Tommaso Merciai <tommaso.merciai.xr@xxxxxxxxxxxxxx> wrote:
> > >   
> > > > Hi Lee,
> > > > Thanks for your patch.
> > > > 
> > > > On Wed, May 20, 2026 at 04:22:25PM +0100, Lee Jones wrote:
> > > > [...]  
> > > > > What happens if fwnode_property_match_string() returns an
> > > > > error?    
> > > > 
> > > > Agree.
> > > > 
> > > > I think we need to check index:
> > > > 		if (index < 0)
> > > > 			return ERR_PTR(-ENOENT);  
> > > 
> > > I don't think that's the right solution. The documentation of
> > > fwnode_property_get_reference_args() says that it return -ENOENT
> > > when the index is out range. So it looks like the OF implementation
> > > has a bug.

On second thought, given the following call stack:

fwnode_led_get()
    fwnode_property_match_string() -> "led-names" not exist -> index = -EINVAL (-22, int)
    fwnode_find_reference(fwnode, "leds", index)
        /* int(-22) -> unsigned int( index > INT_MAX), fwnode_find_reference takes unsigned int */
        fwnode_property_get_reference_args(fwnode, name, NULL, 0, index, &args)
            of_fwnode_get_reference_args()
                /* index is already unsigned int( index > INT_MAX) */
                of_parse_phandle_with_fixed_args(, index,)
                    /* unsigned int(index > INT_MAX) -> int(-22), takes int */
                    __of_parse_phandle_with_args()
                        if (index < 0)
                            return -EINVAL;

The caller shouldn't pass a signed error as an unsigned int.
I think index should be check into fwnode_led_get().

Thanks, Tommaso

> > > 
> > > Looking at of_fwnode_get_reference_args() it directly pass the
> > > return value of __of_parse_phandle_with_args(), which return
> > > -EINVAL when the index is out of range. We should rather fix the OF
> > > implementation of fwnode_property_get_reference_args() to respect
> > > the documented interface.  
> > 
> > Maybe into of_fwnode_get_reference_args() first thing to do is to
> > check index with:
> > 
> > 	if (index > INT_MAX)
> > 		return -ENOENT;
> 
> Either that or fix it in __of_parse_phandle_with_args() which currently
> return -EINVAL for negative indexes. It is used for all variations of
> of_parse_phandle_with[_fixed|_optional|]_args() but none of these
> documented their error values.
> 
> A quick search showed that the vast majority of users either pass a
> constant index or a value from a loop under their control, so those
> would not be affected by such a change. From the one left all the one I
> checked overwrote the returned value with either -ENODEV or -ENOENT,
> but I haven't checked them all.
> 
> I would tend to do the fix in __of_parse_phandle_with_args() as I don't
> really see the value in differentiating negative index from too large
> ones in this API.
> 
> Alban
>