[RFC][PATCH] spl: Turn spl_board_init() into weak symbol

[Simon Glass]

Hi Tom,

On Wed, 21 Sept 2022 at 15:49, Tom Rini <trini at konsulko.com> wrote:
>
> On Wed, Sep 21, 2022 at 02:56:29AM +0200, Marek Vasut wrote:
> > On 9/20/22 21:04, Tom Rini wrote:
> > > On Tue, Sep 20, 2022 at 05:56:50PM +0200, Marek Vasut wrote:
> > > > On 9/20/22 17:43, Simon Glass wrote:
> > > > > On Mon, 19 Sept 2022 at 21:52, Marek Vasut <marex at denx.de> wrote:
> > > > > >
> > > > > > Make spl_board_init() a weak symbol and get rid of Kconfig symbols
> > > > > > and ifdeffery guarding this function. Since the spl_board_init() is
> > > > > > now a weak symbol, boards can either use the default implementation
> > > > > > which is empty and gets inlined with zero text increase, or override
> > > > > > the implementation with their own as needed.
> > > > > >
> > > > > > Signed-off-by: Marek Vasut <marex at denx.de>
> > > > > > ---
> > > > > > Cc: Simon Glass <sjg at chromium.org>
> > > > > > Cc: Tom Rini <trini at konsulko.com>
> > > > > > ---
> > > > >
> > > > > Reviewed-by: Simon Glass <sjg at chromium.org>
> > > > >
> > > > > but please add a nice comment to spl_board_init() indicating what it
> > > > > is for and when it is called.
> > > >
> > > > Actually, I wonder, should we start turning all the other symbols which are
> > > > now protected by Kconfig symbol (the #ifdef CONFIG_FOO) into weak symbols
> > > > without any need for Kconfig symbol guard instead as well?
> > >
> > > I'm not sure this is a good idea. An #ifdef in and of itself is not a
> > > bad thing, and now we add a dummy function on another set of platforms.
> >
> > The empty dummy function is inlined, so that itself does not increase text
> > size.
>
> No, it's 4/8 bytes, weak functions don't get inlined. It's not a big
> deal nor a deal breaker, but it's not free.
>
> > > What is true is that a lot of SPL_foo symbols should be select'd rather
> > > than asked because it's not an option. You enable it, it works, you
> > > disable it, your platform doesn't work. If it's just the #if, we could
> > > rewrite the line as
> > > if (CONFIG_IS_ENABLED(BOARD_INIT))
> > >    spl_board_init();
> > >
> > > as the prototype shouldn't be guarded anyhow.
> >
> > The third option is what is implemented in this patch -- the board port
> > developer implements the weak function override and it always works, because
> > it is no longer selectable.
>
> It's similar, yes. I think the biggest hang-up for me is that while the
> Kconfig help text isn't the best documentation for what is needed when
> adding SPL to a board, it's better than code-only comments. I know Simon
> asked for a comment on the weak function, but how about starting
> doc/develop/spl.rst and make at least this new content kernel-doc format
> so it can be included there?

Yes, well you know my feelings on weak functions and how  hard they
make it to figure out what code is actually running. I actually like
you if(...) idea along with better docs, since the Kconfig option is
at least a positive signal that the function is used.

Regards,
Simon