On Thu, Sep 04, 2014 at 10:18:31AM -0500, Christoph Lameter wrote:
> This is on top of the first patch (but still not against your tree). I
> wonder what you think about this one. No special operations.
>
>
> Since dynticks_idle is only ever modified by the local cpu we do
> not need to use atomics there.
>
> Signed-off-by: Christoph Lameter <c...@linux.com>
Sorry, as discussed earlier, this change is non-trivial. In addition,
you have not justified the change.
> Index: linux/kernel/rcu/tree.c
> ===================================================================
> --- linux.orig/kernel/rcu/tree.c
> +++ linux/kernel/rcu/tree.c
> @@ -213,7 +213,7 @@ static DEFINE_PER_CPU(struct rcu_dyntick
> .dynticks = ATOMIC_INIT(1),
> #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
> .dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE,
> - .dynticks_idle = ATOMIC_INIT(1),
> + .dynticks_idle = 1,
> #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
> };
>
> Index: linux/kernel/rcu/tree.h
> ===================================================================
> --- linux.orig/kernel/rcu/tree.h
> +++ linux/kernel/rcu/tree.h
> @@ -91,7 +91,7 @@ struct rcu_dynticks {
> #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
> long long dynticks_idle_nesting;
> /* irq/process nesting level from idle. */
> - atomic_t dynticks_idle; /* Even value for idle, else odd. */
> + long dynticks_idle; /* Even value for idle, else odd. */
> /* "Idle" excludes userspace execution. */
> unsigned long dynticks_idle_jiffies;
> /* End of last non-NMI non-idle period. */
> Index: linux/kernel/rcu/tree_plugin.h
> ===================================================================
> --- linux.orig/kernel/rcu/tree_plugin.h
> +++ linux/kernel/rcu/tree_plugin.h
> @@ -2644,9 +2644,9 @@ static void rcu_sysidle_enter(int irq)
> j = jiffies;
> ACCESS_ONCE(rdtp->dynticks_idle_jiffies) = j;
> smp_mb__before_atomic();
> - atomic_inc(&rdtp->dynticks_idle);
> + rdtp->dynticks_idle++;
> smp_mb__after_atomic();
And this is completely broken.
> - WARN_ON_ONCE(atomic_read(&rdtp->dynticks_idle) & 0x1);
> + WARN_ON_ONCE(rdtp->dynticks_idle & 0x1);
> }
>
> /*
> @@ -2712,9 +2712,9 @@ static void rcu_sysidle_exit(int irq)
>
> /* Record end of idle period. */
> smp_mb__before_atomic();
> - atomic_inc(&rdtp->dynticks_idle);
> + rdtp->dynticks_idle++;
> smp_mb__after_atomic();
As is this.
> - WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks_idle) & 0x1));
> + WARN_ON_ONCE(!(rdtp->dynticks_idle & 0x1));
>
> /*
> * If we are the timekeeping CPU, we are permitted to be non-idle
> @@ -2755,7 +2755,7 @@ static void rcu_sysidle_check_cpu(struct
> WARN_ON_ONCE(smp_processor_id() != tick_do_timer_cpu);
>
> /* Pick up current idle and NMI-nesting counter and check. */
> - cur = atomic_read(&rdtp->dynticks_idle);
> + cur = rdtp->dynticks_idle;
And this as well.
Thanx, Paul
> if (cur & 0x1) {
> *isidle = false; /* We are not idle! */
> return;
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
