Re: [PATCH] sched: Provide iowait counters

[Peter Zijlstra <peterz@infradead.org>]

On Mon, 2009-07-20 at 11:31 -0700, Arjan van de Ven wrote:
> From cb80fe74bab69449c828d7433fbbea807816eee8 Mon Sep 17 00:00:00 2001
> From: Arjan van de Ven <arjan@linux.intel.com>
> Date: Mon, 20 Jul 2009 11:26:58 -0700
> Subject: [PATCH] sched: Provide iowait counters
> 
> For counting how long an application has been waiting for (disk) IO,
> there currently is only the HZ sample driven information available, while
> for all other counters in this class, a high resolution version is
> available via CONFIG_SCHEDSTATS.
> 
> In order to make an improved bootchart tool possible, we also need
> a higher resolution version of the iowait time.
> 
> This patch below adds this scheduler statistic to the kernel.
> 
> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>

Seems sane enough, I'll pick her up. Thanks!

> +++ b/kernel/sched_fair.c
> @@ -633,6 +633,11 @@ static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
>   		se->block_start = 0;
>   		se->sum_sleep_runtime += delta;
> 
> +		if (tsk->in_iowait) {
> +			se->iowait_sum += delta;
> +			se->iowait_count++;
> +		}
> +
>   		/*
>   		 * Blocking time is in units of nanosecs, so shift by 20 to
>   		 * get a milliseconds-range estimation of the amount of

It might be nice to put a tracepoint there as well, now if there was a
way to specify perf counter attributes in the TRACE_EVENT() magic so
that we can feed stuff into perf_tpcounter_event().

TP_perf_addr()  -- defaults to 0 when not specified
TP_perf_count() -- defaults to 1 when not specified.

Steve, Frederic, is there any way to make that happen?

Failing that we could put an actual swcounter in there I suppose.

That way we could profile applications based on IO-wait, which would be
cool.