Re: [PATCH v3 02/10] iio: dac: ad5446: Use DMA safe buffer for transfers

[Jonathan Cameron <jonathan.cameron@huawei.com>]

On Fri, 31 Oct 2025 17:07:33 +0200
Andy Shevchenko <andriy.shevchenko@intel.com> wrote:

> On Fri, Oct 31, 2025 at 03:00:07PM +0000, Nuno Sá wrote:
> > On Fri, 2025-10-31 at 15:36 +0200, Andy Shevchenko wrote:  
> > > On Fri, Oct 31, 2025 at 12:31:23PM +0000, Nuno Sá via B4 Relay wrote:  
> 
> ...
> 
> > > > +	union {
> > > > +		__be16 d16;
> > > > +		u8 d24[3];  
> > > 
> > > Why not __be32 d24; ? Yes, it will require explicit size to be provided, but at
> > > least it will look consistent with the above. OR u8 d16[2]; ? But then it becomes
> > > simply a u8 buf[3] __aligned...;  
> > 
> > Because I'm just keeping put_unaligned_be24() as before. In fact I'm just keeping the
> > same type. Sure we could do __be32 and the cpu_to_be32() with a proper shift but 
> > I'm already doing way too much than I signed up for when sending v1 :)  
> 
> I think no shift would be needed.

Using a __be32 to me implies things we don't want to imply. In particular
if it's read in to &d24 then if interpreted as a __be32 it would be 256x
too big (I'd not do this for le24 either though which doesn't need the shift)
The u8 [3] makes no such implication.

So I'd prefer sticking to storing be24 in a 3 byte location.

(just picking out this one bit to reply to).