On 02/14/2018 03:26 PM, Vincent Guittot wrote:
> Stopped the periodic update of blocked load when all idle CPUs have fully
> decayed. We introduce a new nohz.has_blocked that reflect if some idle
> CPUs has blocked load that have to be periodiccally updated. nohz.has_blocked
> is set everytime that a Idle CPU can have blocked load and it is then clear
> when no more blocked load has been detected during an update. We don't need
> atomic operation but only to make cure of the right ordering when updating
> nohz.idle_cpus_mask and nohz.has_blocked.
>
> Suggested-by: Peter Zijlstra (Intel) <[email protected]>
> Signed-off-by: Vincent Guittot <[email protected]>
> ---
> kernel/sched/fair.c | 122
> ++++++++++++++++++++++++++++++++++++++++++---------
> kernel/sched/sched.h | 1 +
> 2 files changed, 102 insertions(+), 21 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 7af1fa9..5a6835e 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
>
> [...]
>
> -static void update_nohz_stats(struct rq *rq)
> +static bool update_nohz_stats(struct rq *rq)
> {
> #ifdef CONFIG_NO_HZ_COMMON
> unsigned int cpu = rq->cpu;
>
> + if (!rq->has_blocked_load)
> + return false;
> +
> if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
> - return;
> + return false;
>
> if (!time_after(jiffies, rq->last_blocked_load_update_tick))
> - return;
> + return true;
>
> update_blocked_averages(cpu);
> +
> + return rq->has_blocked_load;
> +#else
> + return false;
> #endif
> }
>
(Wrongly thought that this bit was in a different patch, comment should have
been squashed in previous reply...)
I've been thinking about this, and it's a messy one if we want to
skip CPUs in idle_balance() / clear the nohz.has_blocked_flag.
AFAICT, the rq->has_blocked_load flag can be wrongly cleared: if we're
calling update_nohz_stats() for CPU A, but CPU A got out/in of
idle really quickly in that same timeframe, I'm not sure you can guarantee
the clearing of rq->has_blocked_load done in update_blocked_averages() will
always end up in memory before the setting of the flag in
nohz_balance_enter_idle().
I was going to say we don't have this problem in _nohz_idle_balance() but
actually I think we do. We have the checking of nohz.idle_cpus_mask after the
smp_mb(), which makes sure the clear of nohz.has_blocked will never
overwrite the set in nohz_balance_enter_idle(), but it doesn't
guarantee the same for the rq flag. So we can have nohz CPUs with blocked
load but with rq->has_blocked_load set to false. Which isn't a problem now
but it is if we want to use the flag to skip CPUs.
Am I correct or am I going crazy ? There's a comment about this in
nohz_balance_enter_idle() but I'm confused now:
/*
* Can be set safely without rq->lock held
* If a clear happens, it will have evaluated last additions because
* rq->lock is held during the check and the clear
*/
rq->has_blocked_load = 1;
