[PATCH, -v2] sched/core: Disable CONFIG_SCHED_CORE by default
From: Ingo Molnar
Date: Mon Jun 28 2021 - 16:47:16 EST
* Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Jun 28, 2021 at 12:57 PM Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> > You are completely right, I missed this. Find below the patch to fix it -
> I think you should update the helptext too, which currently says that
> it's enabled if SCHED_SMT is enabled.
> And yes, it may be that there is no measurable performance downside,
> but this feature has been discussed and under development for a long
> time now, and I'd like the default to try to limit the impact since it
> makes little sense to most people.
Updated patch attached, which updates the help text and clarifies what
overhead there is.
While the worst of the overhead is behind the __sched_core_enabled static
key, there's a notable increase in pick_next_task(), which might be
measurable with the right microbenchmark.
So the 'no overhead' claim is probably not entirely true, and comes from
the context of *horrible* overhead of early iterations of the core
scheduling feature ...
From: Ingo Molnar <mingo@xxxxxxxxxx>
Date: Mon, 28 Jun 2021 21:55:16 +0200
Subject: [PATCH] sched/core: Disable CONFIG_SCHED_CORE by default
This option at minimum adds extra code to the scheduler - even if
it's default unused - and most users wouldn't want it.
Reported-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
kernel/Kconfig.preempt | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt
index bd7c4147b9a8..5876e30c5740 100644
@@ -102,7 +102,6 @@ config PREEMPT_DYNAMIC
bool "Core Scheduling for SMT"
- default y
depends on SCHED_SMT
This option permits Core Scheduling, a means of coordinated task
@@ -115,7 +114,8 @@ config SCHED_CORE
- mitigation of some (not all) SMT side channels;
- limiting SMT interference to improve determinism and/or performance.
- SCHED_CORE is default enabled when SCHED_SMT is enabled -- when
- unused there should be no impact on performance.
+ SCHED_CORE is default disabled. When it is enabled and unused,
+ which is the likely usage by Linux distributions, there should
+ be no measurable impact on performance.