Commit a5183862e76fdc25f36b39c2489b816a5c66e2e5
("tick/nohz: Conditionally restart tick on idle exit") allows
a nohz_full CPU to enter idle and return from it with the
scheduler tick disabled (since the tick might be undesired noise).
The idle=poll case still unconditionally restarts the tick when entering
idle.
To reduce the noise for that case as well, stop the tick when entering
idle, for the idle=poll case.
tick_nohz_idle_exit (called if the NEED_RESCHED bit is set)
is responsible for re-enabling the tick when necessary.
Signed-off-by: Marcelo Tosatti <[email protected]>
---
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index 342f58a329f5..fdc240988ea9 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -276,7 +276,7 @@ static void do_idle(void)
* idle as we know that the IPI is going to arrive right away.
*/
if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
- tick_nohz_idle_restart_tick();
+ tick_nohz_idle_stop_tick();
cpu_idle_poll();
} else {
cpuidle_idle_call();
On Wed, Jun 07, 2023 at 10:10:46PM -0300, Marcelo Tosatti wrote:
>
> Commit a5183862e76fdc25f36b39c2489b816a5c66e2e5
> ("tick/nohz: Conditionally restart tick on idle exit") allows
> a nohz_full CPU to enter idle and return from it with the
> scheduler tick disabled (since the tick might be undesired noise).
>
> The idle=poll case still unconditionally restarts the tick when entering
> idle.
>
> To reduce the noise for that case as well, stop the tick when entering
> idle, for the idle=poll case.
>
> tick_nohz_idle_exit (called if the NEED_RESCHED bit is set)
> is responsible for re-enabling the tick when necessary.
Who cares and why?
On Wed, Jun 07, 2023 at 10:10:46PM -0300, Marcelo Tosatti wrote:
>
> Commit a5183862e76fdc25f36b39c2489b816a5c66e2e5
> ("tick/nohz: Conditionally restart tick on idle exit") allows
> a nohz_full CPU to enter idle and return from it with the
> scheduler tick disabled (since the tick might be undesired noise).
>
> The idle=poll case still unconditionally restarts the tick when entering
> idle.
>
> To reduce the noise for that case as well, stop the tick when entering
> idle, for the idle=poll case.
>
> tick_nohz_idle_exit (called if the NEED_RESCHED bit is set)
> is responsible for re-enabling the tick when necessary.
>
> Signed-off-by: Marcelo Tosatti <[email protected]>
>
> ---
>
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index 342f58a329f5..fdc240988ea9 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -276,7 +276,7 @@ static void do_idle(void)
> * idle as we know that the IPI is going to arrive right away.
> */
> if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
> - tick_nohz_idle_restart_tick();
> + tick_nohz_idle_stop_tick();
> cpu_idle_poll();
cpu_idle_poll() re-enables interrupts and only checks need_resched(). If an
interrupt happens and queues a new timer without waking up a task, the timer
gets ignored.
> } else {
> cpuidle_idle_call();
>
On Thu, Jun 08, 2023 at 10:34:00AM +0200, Peter Zijlstra wrote:
> On Wed, Jun 07, 2023 at 10:10:46PM -0300, Marcelo Tosatti wrote:
> >
> > Commit a5183862e76fdc25f36b39c2489b816a5c66e2e5
> > ("tick/nohz: Conditionally restart tick on idle exit") allows
> > a nohz_full CPU to enter idle and return from it with the
> > scheduler tick disabled (since the tick might be undesired noise).
> >
> > The idle=poll case still unconditionally restarts the tick when entering
> > idle.
> >
> > To reduce the noise for that case as well, stop the tick when entering
> > idle, for the idle=poll case.
> >
> > tick_nohz_idle_exit (called if the NEED_RESCHED bit is set)
> > is responsible for re-enabling the tick when necessary.
>
> Who cares and why?
Systems that use isolcpus/nohz_full, and use idle=poll to avoid sleeping.
For example:
https://github.com/intel/FlexRAN/blob/master/README.md
More specifically, the case which HLTs (via nanosleep usually) is:
https://www.isa.org/intech-home/2022/february-2022/departments/what-you-may-not-know-about-softplcs
Can add this to the next version of the patch (unless you have any other
comments).
Thanks.