Re: [PATCH v1 1/1] lockdep: Mark chain_hlock_class_idx() with __maybe_unused

[Boqun Feng <boqun.feng@gmail.com>]

On Mon, Jan 20, 2025 at 09:11:09AM +0100, Ingo Molnar wrote:
> 
> * Boqun Feng <boqun.feng@gmail.com> wrote:
> 
> > On Sun, Dec 15, 2024 at 11:21:36PM +0200, Andy Shevchenko wrote:
> > > On Sun, Dec 15, 2024 at 11:55:08AM -0800, Boqun Feng wrote:
> > > > On Mon, Dec 09, 2024 at 07:08:10PM +0200, Andy Shevchenko wrote:
> > > > > When chain_hlock_class_idx() is unused, it prevents kernel builds with clang,
> > > > > `make W=1` and CONFIG_WERROR=y:
> > > > > 
> > > > > kernel/locking/lockdep.c:435:28: error: unused function 'chain_hlock_class_idx' [-Werror,-Wunused-function]
> > > > > 
> > > > > Fix this by marking it with __maybe_unused.
> > > > > 
> > > > > See also commit 6863f5643dd7 ("kbuild: allow Clang to find unused static
> > > > > inline functions for W=1 build").
> > > > > 
> > > > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > > > 
> > > > This looks fine to me, so I'm going to queue it, but could you do me
> > > > favor if you could share the exact configs that would make
> > > > chain_hlock_class_idx() an unused fuction in kernel/locking/lockdep.c ?
> > > 
> > > I might harvest this when I come from vacations (some around mid-January).
> > > 
> > 
> > After some investigation myself, it turns out that
> > chain_hlock_class_idx() is defined outside "#ifdef CONFIG_PROVING_LOCK",
> > but only used under CONFIG_PROVING_LOCK=y, hence the unused warning.
> > 
> > TBH, I feel we should really clean-up/consolidate those "#ifdef" to make
> > internal definition/usage more clear. But nothing blocks this patch, it
> > fixes a real issue, I will add the CONFIG_PROVING_LOCK part in the
> > commit log. Thanks!
> 
> So now that we have:
> 
>    343060092585 lockdep: Move lockdep_assert_locked() under #ifdef CONFIG_PROVE_LOCKING
> 
> Do we really need:
> 
>    8148fa2e022b lockdep: Mark chain_hlock_class_idx() with __maybe_unused
> 
> As it would hide the problem fixed by 343060092585?
> 

These two commits fix different functions, so they are both needed.
Though, we could probably regroup the functions into one big #ifdef
block so that we won't need the __maybe_unused annotations.

I will take a look.

Regards,
Boqun

> __maybe_unused annotations are almost always canaries of something 
> messy being hidden.
> 
> Thanks,
> 
> 	Ingo