Re: [PATCH 1/1] cpu: Add annotation inside clear_tasks_mm_cpumask()

[Peter Zijlstra]

On Wed, Apr 15, 2020 at 07:49:37PM +0100, Jules Irenge wrote:
> Sparse reports a warning
> 
> warning: context imbalance in clear_tasks_mm_cpumask() - different lock contexts for basic block

Does this sparse crap actually ever catch a real problem, or does it
only result in horrible code like this?

> The root cause is the missing annotation inside clear_tasks_mm_cpumask()
> 
> Add the missing __acquire(&t->alloc_lock) annotation.

That's just wrong; the actual cause is find_lock_task_mm() not being
annotated, and the reason for that is because __cond_lock() is a
horrible piece of crap.

Barring any evidence that these annotations actually help anybody, can't
we just remove it all instead of making an ever bigger mess of things?

> Signed-off-by: Jules Irenge <jbi.octave@xxxxxxxxx>
> ---
>  kernel/cpu.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/kernel/cpu.c b/kernel/cpu.c
> index 9c706af713fb..d8c452a8dd09 100644
> --- a/kernel/cpu.c
> +++ b/kernel/cpu.c
> @@ -834,6 +834,7 @@ void clear_tasks_mm_cpumask(int cpu)
>  		t = find_lock_task_mm(p);
>  		if (!t)
>  			continue;
> +		__acquire(&t->alloc_lock);
>  		cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
>  		task_unlock(t);
>  	}
> -- 
> 2.24.1
>