Re: [PATCH 1/1] perf/core: fix implicitly enable dynamic interrupt throttle

[Peter Zijlstra]

On Tue, May 03, 2016 at 12:26:06AM -0700, kan.liang@xxxxxxxxx wrote:
> From: Kan Liang <kan.liang@xxxxxxxxx>
> 
> This patch fixes an issue which was introduced from 'commit 91a612eea9a3
> ("perf/core: Fix dynamic interrupt throttle")'
> The old patch unconditionally sets the perf_sample_allowed_ns value to
> !0. But that could trigger an issue in the following corner case.
> The user can disable the dynamic interrupt throttle mechanism by setting
> perf_cpu_time_max_percent to 0. Then they changes
> perf_event_max_sample_rate.
> For this case, the mechanism will be enabled implicitly, because
> perf_sample_allowed_ns becomes !0.
> 
> This patch only updates the perf_sample_allowed_ns when the dynamic
> interrupt throttle mechanism is enabled.
> 
> Signed-off-by: Kan Liang <kan.liang@xxxxxxxxx>
> ---
>  kernel/events/core.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 4e2ebf6..4042a3d 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -375,6 +375,14 @@ static void update_perf_cpu_limits(void)
>  {
>  	u64 tmp = perf_sample_period_ns;
>  
> +	/*
> +	 * Don't update the perf_sample_allowed_ns,
> +	 * if the dynamic interrupt throttle mechanism is disabled.
> +	 */
> +	if (sysctl_perf_cpu_time_max_percent == 100 ||
> +	    sysctl_perf_cpu_time_max_percent == 0)
> +		return;
> +


Hmm, would it not be nicer to simply reject the write instead of
silently ignoring it?

---
 kernel/events/core.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 050a290c72c7..0a51a568d4eb 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -396,6 +396,13 @@ int perf_proc_update_handler(struct ctl_table *table, int write,
 	if (ret || !write)
 		return ret;
 
+	/*
+	 * If the throttling is disabled; don't allow the write.
+	 */
+	if (sysctl_perf_cpu_time_max_percent == 100 ||
+	    sysctl_perf_cpu_time_max_percent == 0)
+		return -EINVAL;
+
 	max_samples_per_tick = DIV_ROUND_UP(sysctl_perf_event_sample_rate, HZ);
 	perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate;
 	update_perf_cpu_limits();