Re: [PATCH v13 13/15] arm64/delay: enable testing smp_cond_load_relaxed_timeout()

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The generic synchronization macros `smp_cond_load_relaxed_timeout()` and `smp_cond_load_acquire_timeout()` (defined in `include/asm-generic/barrier.h`) are standard kernel APIs. On ARM64, these macros expand to use `cpu_poll_relax()`, which evaluates `arch_timer_evtstrm_available()` and calls `__cmpwait_relaxed_timeout()`. The latter then calls `__delay_cycles()`. Any loadable module using these generic timeout-based wait macros inherently requires both `arch_timer_evtstrm_available` and `__delay_cycles` to be exported.

However, these backend symbols are conditionally exported using `EXPORT_SYMBOL_IF_KUNIT`. If a standard non-KUnit loadable module uses these generic APIs while `CONFIG_KUNIT=n`, it will encounter 'undefined symbol' linker errors during modpost. Because these symbols satisfy a generic architecture-independent API exposed as inline macros, they must be exported unconditionally (e.g., via `EXPORT_SYMBOL_GPL`).
--

commit f9d079be600d568480e3e1fb1f712a02babf0680
Author: Ankur Arora <ankur.a.arora@oracle.com>

arm64/delay: enable testing smp_cond_load_relaxed_timeout()

This commit conditionally exports __delay_cycles and
arch_timer_evtstrm_available to KUnit to enable barrier tests to be built as a
module.

> diff --git a/arch/arm64/lib/delay.c b/arch/arm64/lib/delay.c
> index c660a7ea26dd6..dfb102ce30098 100644
> --- a/arch/arm64/lib/delay.c
> +++ b/arch/arm64/lib/delay.c
[ ... ]
> @@ -30,6 +31,7 @@ u64 notrace __delay_cycles(void)
>  	guard(preempt_notrace)();
>  	return __arch_counter_get_cntvct_stable();
>  }
> +EXPORT_SYMBOL_IF_KUNIT(__delay_cycles);

[Severity: Medium]
Will standard loadable modules encounter undefined symbol errors during modpost
if they use the generic timeout-based wait macros when CONFIG_KUNIT=n?

On ARM64, the generic synchronization macros smp_cond_load_relaxed_timeout()
and smp_cond_load_acquire_timeout() expand to use cpu_poll_relax(), which
evaluates arch_timer_evtstrm_available() and calls __cmpwait_relaxed_timeout(),
which in turn calls __delay_cycles().

Should this be exported unconditionally (e.g., via EXPORT_SYMBOL_GPL) so that
normal modules can use this architecture-independent API?

>  
>  void __delay(unsigned long cycles)
> diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
> index 90aeff44a2764..1de63e1a2cd2f 100644
> --- a/drivers/clocksource/arm_arch_timer.c
> +++ b/drivers/clocksource/arm_arch_timer.c
[ ... ]
> @@ -896,6 +897,7 @@ bool arch_timer_evtstrm_available(void)
>  	 */
>  	return cpumask_test_cpu(raw_smp_processor_id(), &evtstrm_available);
>  }
> +EXPORT_SYMBOL_IF_KUNIT(arch_timer_evtstrm_available);

[Severity: Medium]
Could this similarly break standard modules that use the generic API?

As mentioned above, arch_timer_evtstrm_available() is evaluated during the
expansion of the generic smp_cond_load_*_timeout macros. An unconditional
export might be needed here as well to prevent module linking errors.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260702013334.140905-1-ankur.a.arora@oracle.com?part=13