On Thu, Nov 02, 2017 at 12:49:31AM +0100, Maciej S. Szmigiero wrote:
> Hi Andrew,
>
> On 01.11.2017 13:33, Maciej S. Szmigiero wrote:
> > On 01.11.2017 13:31, Andrew Lunn wrote:
> >>> Yes, I did it the same way as the existing code did for
> >>> phy->phy_led_triggers
> >>> for reasons of both consistency and also to be on the safe side because
> >>> maybe there is some non-obvious reason why it has to be freed
> >>> explicitly (?).
> >>
> >> Hi Maciej
> >>
> >> Occasionally, there is a need to call devm_kfree(). But i don't see
> >> anything here why it is needed. So i would drop your devm_kfree(), and
> >> if you feel like it, add an additional patch removing the existing
> >> one.
> >
> > OK, will do as you suggest.
> >
>
> Upon closer inspection of the code it turned out that these devm_kfree()
> calls actually had some purpose - the PHY core ignores the return value
> of phy_led_triggers_register() and will successfully attach a PHY even
> if this function returns an error.
>
> In that case these LED trigger structures would unnecessary take some
> memory, that's why (probably) the PHY LED code frees them on error
> path.
O.K. Thanks for looking into this.
Andrew
