Re: [PATCH 3/6] driver core: cpu: optimize print_cpus_isolated()

[Thomas Gleixner <tglx@xxxxxxxxxxxxx>]

On Mon, May 13 2024 at 15:01, Yury Norov wrote:
> The function may be called with housekeeping_cpumask == cpu_possible_mask,

How so? There is no cpumask argument in the function signature. Can you
please be precise?

> and in such case the 'isolated' cpumask would be just empty.
>
> We can call cpumask_clear() in that case, and save CPU cycles.
>
> @@ -282,8 +282,10 @@ static ssize_t print_cpus_isolated(struct device *dev,
>  	if (!alloc_cpumask_var(&isolated, GFP_KERNEL))
>  		return -ENOMEM;
>  
> -	cpumask_andnot(isolated, cpu_possible_mask,
> -		       housekeeping_cpumask(HK_TYPE_DOMAIN));
> +	if (cpu_possible_mask != housekeeping_cpumask(HK_TYPE_DOMAIN))
> +		cpumask_andnot(isolated, cpu_possible_mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
> +	else
> +		cpumask_clear(isolated);
>  	len = sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(isolated));
>  
>  	free_cpumask_var(isolated);

Seriously? You need clear() to emit an empty string via %*pbl?

	if (cpu_possible_mask != housekeeping_cpumask(HK_TYPE_DOMAIN)) {
        	if (!alloc_cpumask_var(&isolated, GFP_KERNEL))
                	return -ENOMEM;
                cpumask_andnot(isolated, cpu_possible_mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
                len = sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(isolated));
	  	free_cpumask_var(isolated);
	} else {
        	len = sysfs_emit(buf, "\n");
        }

That actually would make sense and spare way more CPU cycles, no?

Is it actually worth the larger text size? Not really convinced about that.

Thanks,

        tglx