Re: [PATCH 13/15] x86/kconfig/64: Enable popular scheduler, cgroups and namespaces options in the defconfig

[Yafang Shao <laoar.shao@gmail.com>]

Hello Mingo,

 > +CONFIG_VIRT_CPU_ACCOUNTING_GEN=y
 > +CONFIG_IRQ_TIME_ACCOUNTING=y

Enabling CONFIG_IRQ_TIME_ACCOUNTING=y can lead to user-visible 
behavioral changes. For more context, please refer to the related 
discussion here: 
https://lore.kernel.org/all/20241222024734.63894-1-laoar.shao@gmail.com/ .

If we decide to enable it by default, we should clearly document this 
behavior change. Below is the patch I wrote earlier but haven’t sent out 
for review yet.

----

Subject: [PATCH] init/Kconfig: document behavior change when enabling
  IRQ_TIME_ACCOUNTING

After we enabled CONFIG_IRQ_TIME_ACCOUNTING, we noticed that the IRQ
usage is not accounted to the tasks and thus not accounted to the CPU
cgroup neither. This behavior change results in issues [0] in our
production servers and finally we have to revert it.

We'd better clearly document this behavior change in case it might
matter to the user.

Link: 
https://lore.kernel.org/all/20241222024734.63894-1-laoar.shao@gmail.com/ [0]
Suggested-by: "Michal Koutný" <mkoutny@suse.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
---
  init/Kconfig | 8 ++++++++
  1 file changed, 8 insertions(+)

diff --git a/init/Kconfig b/init/Kconfig
index a20e6efd3f0f..191df0b5cf1c 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -563,6 +563,14 @@ config IRQ_TIME_ACCOUNTING
        transitions between softirq and hardirq state, so there can be a
        small performance impact.

+      Enabling IRQ_TIME_ACCOUNTING excludes IRQ usage from the CPU usage
+      statistics of individual tasks and, consequently, it is not accounted
+      for in CPU cgroups. As a result, a task's CPU usage will accurately
+      reflect only its user time and system time. IRQ usage is instead
+      attributed at the global level and can be observed in metrics such as
+      /proc/stat or, potentially, at the cgroup level in files like
+      irq.pressure.
+
        If in doubt, say N here.

  config HAVE_SCHED_AVG_IRQ