Re: [PATCH v2] tick/broadcast: Do not set oneshot_mask except was_periodic was true

[Frederic Weisbecker <frederic@kernel.org>]

On Tue, May 02, 2023 at 02:38:57PM +0200, Thomas Gleixner wrote:
> Updated patch below.
> 
> Thanks,

Looks good from my layperson's eyes, just a doubt about a comment below:

> +
> +	/*
> +	 * When switching from periodic to oneshot mode arm the broadcast
> +	 * device for the next tick.
> +	 *
> +	 * If the broadcast device has been replaced in oneshot mode and
> +	 * the oneshot broadcast mask is not empty, then arm it to expire
> +	 * immediately in order to reevaluate the next expiring timer.
> +	 * nexttick is 0 and therefore in the past which will cause the

Is nexttick really in the past? It's set to tick_next_period...

Thanks.

> +	 * clockevent code to force an event.
> +	 *
> +	 * For both cases the programming can be avoided when the oneshot
> +	 * broadcast mask is empty.
> +	 *
> +	 * tick_broadcast_set_event() implicitly switches the broadcast
> +	 * device to oneshot state.
> +	 */
> +	if (!cpumask_empty(tick_broadcast_oneshot_mask))
> +		tick_broadcast_set_event(bc, cpu, nexttick);
>  }