Re: [PATCH net-next v3 1/3] net: phy: mxl-86110: add basic support for led_brightness_set op

[Daniel Golle <daniel@makrotopia.org>]

On Thu, Aug 21, 2025 at 05:36:20PM -0700, Jakub Kicinski wrote:
> On Wed, 20 Aug 2025 14:34:07 +0100 Daniel Golle wrote:
> > > > +	if (index >= MXL86110_MAX_LEDS)
> > > > +		return -EINVAL;
> > > > +
> > > > +	/* force manual control */
> > > > +	set = MXL86110_COM_EXT_LED_GEN_CFG_LFE(index);
> > > > +	/* clear previous force mode */
> > > > +	mask = MXL86110_COM_EXT_LED_GEN_CFG_LFM(index);
> > > > +
> > > > +	/* force LED to be permanently on */
> > > > +	if (value != LED_OFF)
> > > > +		set |= MXL86110_COM_EXT_LED_GEN_CFG_LFME(index);  
> > > 
> > > That is particularly complex. We know index is a u8, so why not
> > > GENMASK_U8(1 + 3 * index, 3 * index)? But set is a u16, so
> > > GENMASK_U16() would also be valid.  
> > 
> > I chose this construct to avoid reusing the macro parameter as gcc would
> > rightously complain about that potentially having unexpected side-effects.
> > 
> > Eg.
> > 
> > #define FOO(a) ((a)+(a))
> > 
> > Now with var=10, when calling FOO(var++) the result will be 21 and
> > var will be equal to 12, which isn't intuitive without seeing the
> > macro definition.
> > 
> > Also using GENMASK_TYPE would not avoid the problem of macro
> > parameter reuse.
> > 
> > However, I agree that the macro itself is also weirdly complex and
> > confusing (but at the same time also very common, a quick grep reveals
> > hundreds of occurances of that pattern in Linux sources), so maybe we
> > should introduce some generic helpers for this (quite common) use-case?
> > I can do that, but I certainly can't take care of migrating all the
> > existing uses of this pattern to switch to the new helper.
> 
> IMHO this is one of the cases where the code would be far clearer
> without he macros..
> 
> #define MXL86110_COM_EXT_LED_GEN_CFG			0xA00B
> # define MXL86110_COM_EXT_LED_GEN_CFG_LFME(x)		(1 << (3 * (x)))
> # define MXL86110_COM_EXT_LED_GEN_CFG_LFE(x)		(2 << (3 * (x)))
> # define MXL86110_COM_EXT_LED_GEN_CFG_MASK(x)		(3 << (3 * (x)))
> 
> Right?

I agree. If everyone is fine with using numbers and bitshift instead of
GENMASK(...) then I'd also prefer it in this case.