Re: [PATCH net-next v3 08/12] net: stmmac: Pass stmmac_priv in some callbacks

[Simon Horman]

On Mon, Apr 10, 2023 at 04:24:22PM -0500, Andrew Halaney wrote:
> On Fri, Apr 07, 2023 at 12:34:53PM -0500, Andrew Halaney wrote:
> > On Sat, Apr 01, 2023 at 05:06:21PM +0200, Simon Horman wrote:
> > > On Fri, Mar 31, 2023 at 04:45:45PM -0500, Andrew Halaney wrote:
> > > > Passing stmmac_priv to some of the callbacks allows hwif implementations
> > > > to grab some data that platforms can customize. Adjust the callbacks
> > > > accordingly in preparation of such a platform customization.
> > > > 
> > > > Signed-off-by: Andrew Halaney <ahalaney@xxxxxxxxxx>
> > > 
> > > ...
> > > 
> > > >  #define stmmac_reset(__priv, __args...) \
> > > > @@ -223,59 +240,59 @@ struct stmmac_dma_ops {
> > > >  #define stmmac_dma_init(__priv, __args...) \
> > > >  	stmmac_do_void_callback(__priv, dma, init, __args)
> > > >  #define stmmac_init_chan(__priv, __args...) \
> > > > -	stmmac_do_void_callback(__priv, dma, init_chan, __args)
> > > > +	stmmac_do_void_callback(__priv, dma, init_chan, __priv, __args)
> > > 
> > > Hi Andrew,
> > > 
> > > Rather than maintaining these macros can we just get rid of them?
> > > I'd be surprised if things aren't nicer with functions in their place [1].
> > > 
> > > f.e., we now have (__priv, ..., __priv, ...) due to a generalisation
> > >       that seems to take a lot more than it gives.
> > > 
> > > [1] https://lore.kernel.org/linux-arm-kernel/ZBst1SzcIS4j+t46@xxxxxxxxxxxx/
> > > 
> > 
> > Thanks for the pointer. I think that makes sense, I'll take that
> > approach for these functions (and maybe in a follow-up series I'll
> > tackle all of them just because the lack of consistency will eat me up).
> > 
> 
> I tried taking this approach for a spin, and I'm not so sure about it
> now!
> 
> 1. Implementing the functions as static inline requires us to know
>    about stmmac_priv, but that's getting into circular dependency land
> 2. You could define them in hwif.c, but then they're not inlined
> 3. There's still a good bit of boilerplate that's repeated all over
>    with the approach. Ignoring 1 above, you get something like this:
> 
> static inline int stmmac_init_chan(struct stmmac_priv *priv,
> 				   void __iomem *ioaddr,
> 				   struct stmmac_dma_cfg *dma_cfg, u32 chan)
> {
> 	if (priv->hw->dma && priv->hw->dma->init_chan) {
> 		priv->hw->dma->init_chan(priv, ioaddr, dma_cfg, chan);
> 		return 0;
> 	}
> 	return -EINVAL;
> }
> 
> that is then repeated for every function... which is making me actually
> appreciate the macros some for reducing boilerplate.
> 
> Am I suffering from a case of holiday brain, and 1-3 above are silly
> points with obvious answers, or do they make you reconsider continuing
> with the current approach in hwif.h?

I'm about to embark to the holiday brain zone.

But before I do I wanted to acknowledge your concerns and that, yes,
it may be easier said than done.