* Christoph Lameter <[email protected]> wrote:
> We do not care about races for the expiration logic in
> refresh_cpu_vm_stats(). Draining is a rare act after all.
> No need to create too much overhead for that.
>
> Use raw_cpu_ops there.
>
> Signed-off-by: Christoph Lameter <[email protected]>
>
> Index: linux/mm/vmstat.c
> ===================================================================
> --- linux.orig/mm/vmstat.c 2013-09-23 10:20:31.742262228 -0500
> +++ linux/mm/vmstat.c 2013-09-23 10:20:31.738262268 -0500
> @@ -439,6 +439,10 @@ static inline void fold_diff(int *diff)
> * statistics in the remote zone struct as well as the global cachelines
> * with the global counters. These could cause remote node cache line
> * bouncing and will have to be only done when necessary.
> + *
> + * Note that we have to use raw_cpu ops here. The thread is pinned
> + * to a specific processor but the preempt checking logic does not
> + * know about this.
That's not actually true - debug_smp_processor_id() does a check for the
pinning status of the current task:
/*
* Kernel threads bound to a single CPU can safely use
* smp_processor_id():
*/
if (cpumask_equal(tsk_cpus_allowed(current), cpumask_of(this_cpu)))
goto out;
You should factor out those existing debug checks and reuse them, instead
of using inferior ones.
Note that debug_smp_processor_id() can probably be optimized a bit: today
we have p->nr_cpus_allowed which tracks the pinning status, so instead of
the above line we could write this cheaper form:
if (current->nr_cpus_allowed == 1)
goto out;
(This should help on kernels configured for larger systems where the
cpumask is non-trivial.)
What we cannot do is to hide the weakness of the debug check you added by
adding various workarounds to core code.
Thanks,
Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
