Re: NO_HZ_FULL vs NO_HZ_IDLE: ~300ns cyclictest latency regression from context tracking overhead

[Ionut Nechita (Wind River)]

On Fri, 27 Mar 2026 17:10:07 +0100, Valentin Schneider wrote:
> I can't answer for 2., but for 1. & 3.: NO_HZ_FULL is about squeezing
> as much as you can out of a CPU running a pure userspace application.
> Any kernel entry is an anomaly / interference, IOW "you lose". If you
> know your workload is going to periodically enter the kernel, put in on
> a CPU that isn't nohz_full (but maybe isolated).

Thanks Valentin, that makes sense.

For 1. and 3.: I'll test with isolated CPUs without nohz_full
(isolcpus=domain,managed_irq only) for workloads that do periodic
syscalls. This should give the isolation benefits without the context
tracking overhead.

For 2.: I've filed a related bug at bugzilla.kernel.org covering the
latency gap between 6.12-stable and 6.12-rt with PREEMPT_RT:
  https://bugzilla.kernel.org/show_bug.cgi?id=221316

Thanks,
Ionut