Re: [PATCH] cleanup: Fix unused guard error function with DEFINE_CLASS_IS_COND_GUARD

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Mon, Aug 11, 2025 at 03:56:10PM -0700, dan.j.williams@intel.com wrote:
> Andy Shevchenko wrote:
> > On Mon, Aug 04, 2025 at 03:09:54PM -0700, Dan Williams wrote:
> > > Andy reports that the "lock_timer" scheme in kernel/time/posix-timers.c,
> > > with its custom usage of DEFINE_CLASS_IS_COND_GUARD(), results in:
> > > 
> > > kernel/time/posix-timers.c:89:1: error: unused function 'class_lock_timer_lock_err' [-Werror,-Wunused-function]
> > >    89 | DEFINE_CLASS_IS_COND_GUARD(lock_timer);
> > > 
> > > ...with a clang W=1 build. Per Nathan, clang catches unused "static inline"
> > > functions in C files since commit 6863f5643dd7 ("kbuild: allow Clang to
> > > find unused static inline functions for W=1 build").
> > > 
> > > There are 2 ways to solve this, either mark the class_##_lock_err()
> > > function as __maybe_unused, or make sure class_##_lock_err() *is* used /
> > > gets called to disposition the lock status.
> > > 
> > > At present __lock_timer() only indicates failure with a NULL __guard_ptr().
> > > However, one could imagine that __lock_timer(), or some other custom
> > > conditional locking primitive, wants to pass an ERR_PTR() to indicate the
> > > reason for the lock acquisition failure.
> > > 
> > > Update __scoped_cond_guard() to check for ERR_PTR() in addition to NULL
> > > @scope values. This allows __lock_timer(), or another open coded
> > > DEFINE_CLASS_IS_COND_GUARD() user, to switch to passing an ERR_PTR() in the
> > > future. In the meantime, this just silences the warning.
> > 
> > Hmm... It seems fixes the timer case, but left others still fail:
> > 
> > drivers/pwm/core.c:54:1: error: unused function 'class_pwmchip_lock_err' [-Werror,-Wunused-function]
> >    54 | DEFINE_GUARD(pwmchip, struct pwm_chip *, pwmchip_lock(_T), pwmchip_unlock(_T))
> >       | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > include/linux/cleanup.h:380:2: note: expanded from macro 'DEFINE_GUARD'
> >   380 |         DEFINE_CLASS_IS_GUARD(_name)
> >       |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > include/linux/cleanup.h:372:2: note: expanded from macro 'DEFINE_CLASS_IS_GUARD'
> >   372 |         __DEFINE_GUARD_LOCK_PTR(_name, _T)
> >       |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > include/linux/cleanup.h:358:20: note: expanded from macro '__DEFINE_GUARD_LOCK_PTR'
> >   358 |         static inline int class_##_name##_lock_err(class_##_name##_t *_T)   \
> >       |                           ^~~~~~~~~~~~~~~~~~~~~~~~
> > <scratch space>:81:1: note: expanded from here
> >    81 | class_pwmchip_lock_err
> >       | ^~~~~~~~~~~~~~~~~~~~~~
> > 1 error generated.
> 
> Ok, so in these cases where the guard is unconditional, the _lock_err()
> method can be safely ignored. Here is a quick patch (below), only gcc
> compile tested, that tries to quiet the warning for unconditional
> guards, but keep the warning for conditional guards.
> 
> If you have some time to try it out, great, otherwise I will circle back
> with a clang W=1 compile test before submitting the formal patch:

I have applied it on top of the previous (incomplete) solution and tried on two
branches, one is based on v6.17-rc1 and the other on Linux Next. I do not see
problem anymore.

Tested-by: Andy Shevchenko <andriy.shevchenko@intel.com>

-- 
With Best Regards,
Andy Shevchenko