Re: [v2 PATCH 4/4] timers: logic to enable timer migration.

[Oleg Nesterov <oleg@redhat.com>]

On 03/04, Arun R Bharadwaj wrote:
>
> @@ -628,7 +629,7 @@ __mod_timer(struct timer_list *timer, un
>  {
>  	struct tvec_base *base, *new_base;
>  	unsigned long flags;
> -	int ret;
> +	int ret, current_cpu, preferred_cpu;
>  
>  	ret = 0;
>  
> @@ -649,6 +650,16 @@ __mod_timer(struct timer_list *timer, un
>  
>  	new_base = __get_cpu_var(tvec_bases);
>  
> +	current_cpu = smp_processor_id();
> +	preferred_cpu = get_nohz_load_balancer();
> +	if (enable_timer_migration && !tbase_get_pinned(timer->base) &&
> +			idle_cpu(current_cpu) && preferred_cpu != -1) {
> +		new_base = per_cpu(tvec_bases, preferred_cpu);
> +		timer_set_base(timer, new_base);
> +		timer->expires = expires;
> +		internal_add_timer(new_base, timer);
> +		goto out_unlock;

I didn't read the whole series, but this looks very wrong.

We can not do internal_add_timer/etc until we lock new_base, please
look how the current code does this under "if (base != new_base)".

I think you can do something like

	-	new_base = __get_cpu_var(tvec_bases);
	+
	+	new_cpu = smp_processor_id();
	+	if (enable_timer_migration && ....)
	+		new_cpu = preferred_cpu;
	+
	+	new_base = per_cpu(tvec_bases, new_cpu);

		if (base != new_base) {

Oleg.