Re: [PATCH 7/7] x86: Tell about irq stack coverage

[Linus Torvalds]

On Wed, Sep 25, 2013 at 5:40 PM, Joe Perches <joe@xxxxxxxxxxx> wrote:
>
> Huh?  That matches all the ARCH_HAS_<foo> patterns.

Right. And they are all crap. lib/string.c is a prime example of
something that should never have happened.

The ARCH_HAS_xyz pattern is totally retarded. It's wrong.

For big conceptual features, we should use Kconfig symbols.

And for smaller things - like lib/string.c - where we have
compatibility fallback functions but want architectures able to
override them with optimized ones one function at a time, we should
either use weak functions (appropriate for some cases), or the symbol
that protects them should the the SAME SYMBOL WE USE. Rather than some
made-up crap-for-brains new ARCH_HAS_xyz symbol. That way it shows up
in greps, and that way we don't have any question about what random
symbol pattern we use that particular day.

So for *bad* use, see lib/string.c, and the ARCH_AS_xyz horror.

For *good* use, see lib/div64.c or lib/find_next_bit.c.

Notice how div64.c doesn't make up new ARCH_HAS_random_crap names? And
no, you don't have to define those things as macros, you can define
them as functions (inline or not), and then just do

    #define find_next_zero_bit find_next_zero_bit

to tell the rest of the world "Look, I have this defined".

The whole "make up a totally unrelated second name for it" means that
we have things like __HAVE_ARCH_STRLEN but also things like
ARCH_HAS_PREFETCHW. Ugh.

                       Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/