On Wed, May 06, 2020 at 04:19:45PM +0200, Arnd Bergmann wrote:
> On Wed, May 6, 2020 at 4:44 AM Nathan Chancellor
> <natechancellor@xxxxxxxxx> wrote:
> >
> > On Tue, May 05, 2020 at 04:19:17PM +0200, Arnd Bergmann wrote:
> > > clang points out that doing arithmetic between diffent enums is usually
> > ^ different
> > > a mistake:
> > >
> > > drivers/leds/leds-lm355x.c:167:28: warning: bitwise operation between different enumeration types ('enum lm355x_tx2' and 'enum lm355x_ntc') [-Wenum-enum-conversion]
> > > reg_val = pdata->pin_tx2 | pdata->ntc_pin;
> > > ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
> > > drivers/leds/leds-lm355x.c:178:28: warning: bitwise operation between different enumeration types ('enum lm355x_tx2' and 'enum lm355x_ntc') [-Wenum-enum-conversion]
> > > reg_val = pdata->pin_tx2 | pdata->ntc_pin | pdata->pass_mode;
> > > ~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
> > >
> > > In this driver, it is intentional, so add a cast to hide the false-positive
> >
> > Not sure that I would call this a false positive. The warning is correct
> > that there is a bitwise operation between different enumeration types
> > but we do not care since we are just using the enumerated type for its
> > integer value in lieu of a #define VAR value.
>
> Right, I meant that the code works as intended and said "false positive"
> to avoid claiming the driver is broken when this was a deliberate
> design point.
Ack.
> We do want clang to warn about this though as you say, so I can
> rephrase it to explain that both the driver and the compiler work
> as intended but they clash in their views of how to do it ;-)
Yes, that would be good if we don't go a different direction based on
your commends below.
> > > - reg_val = pdata->pass_mode;
> > > + reg_val = (u32)pdata->pass_mode;
> >
> > Is this cast needed? I don't think there should be warning from going
> > from an enumerated type to unsigned int.
>
> This cast is not needed for warnings, I added it for consistency because
> it seemed odd to cast only four of the five enums. I can remove if though
> if you think it's clearer without the cast.
I don't really have a strong opinion but I do think that not having the
cast makes the patch a little more specific/precise.
> There may also be a different solution in completely removing the
> lm355x_chip_init() function, as it was added at a time when we
> were converting the last board files into devicetree, and there has
> never been a board file defining lm355x_platform_data.
>
> There is unfortunately no DT support either in it, so I assume we
> could just remove the driver completely, or change it to use a
> DT binding similar to
> Documentation/devicetree/bindings/leds/leds-lm36*.txt
>
> LED maintainers, any opinions on this?
>
> Arnd
Cheers,
Nathan
