Re: [RFC PATCH V2 net-next 03/11] net: hibmcge: Add mdio and hardware configuration supported in this module

[Andrew Lunn <andrew@lunn.ch>]

On Fri, Aug 16, 2024 at 02:10:36PM +0800, Jijie Shao wrote:
> 
> on 2024/8/16 10:25, Andrew Lunn wrote:
> > > +struct hbg_mdio_command {
> > > +	union {
> > > +		struct {
> > > +			u32 mdio_devad : 5;
> > > +			u32 mdio_prtad :5;
> > > +			u32 mdio_op : 2;
> > > +			u32 mdio_st : 2;
> > > +			u32 mdio_start : 1;
> > > +			u32 mdio_clk_sel : 1;
> > > +			u32 mdio_auto_scan : 1;
> > > +			u32 mdio_clk_sel_exp : 1;
> > > +			u32 rev : 14;
> > > +		};
> > > +		u32 bits;
> > > +	};
> > > +};
> > This is generally not the way to do this. Please look at the macros in
> > include/linux/bitfield.h. FIELD_PREP, GENMASK, BIT, FIELD_GET
> > etc. These are guaranteed to work for both big and little endian, and
> > you avoid issues where the compiler decides to add padding in your
> > bitfields.
> > 
> > 	Andrew
> 
> Thanks, I already know about macros like FIELD_PREP or FIELD_GET.
> and these macros are already used in parts of this patch set.
> 
> But I think this writing style in here very convenient.
> Although padding needs to be added during definition,
> but you can use command.mdio_start or command->mdio_start
> to access specific bitfields.
> 
> Although FIELD_PREP/FIELD_GET is convenient,
> But it also needs to define the mask using BIT or GENMASK,
> and the mask can only be a static constant,
> which makes it difficult to use sometimes.

Have a look around. How many drivers use this sort of union? How many
use bitfield.h. There is a reason the union is not used. I suspect
there is nothing in the C standard which guarantees it works.

      Andrew