Skip to contentSkip to conversation

sched/core: add forced idle accounting for cgroups

PATCH Superseded Josh Don 2 participants 4 messages Jun 21, 2022 – Jun 28, 2022

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Josh Don

author
#0 lore raw

PATCH v3 1/1 sched/core: add forced idle accounting for cgroups

4feee7d1260 previously added per-task forced idle accounting. This patch extends this to also include cgroups.

rstat is used for cgroup accounting, except for the root, which uses kcpustat in order to bypass the need for doing an rstat flush when reading root stats.

Only cgroup v2 is supported. Similar to the task accounting, the cgroup accounting requires that schedstats is enabled.

---
v3: Fix build error, and revert back to __schedstat_*
v2: Per Tejun's suggestion, move the forceidle stat to cgroup_base_stat directly.

 include/linux/cgroup-defs.h |  4 ++++
 include/linux/kernel_stat.h |  7 ++++++
 kernel/cgroup/rstat.c       | 44 ++++++++++++++++++++++++++++++++-----
 kernel/sched/core_sched.c   |  6 ++++-
 kernel/sched/cputime.c      | 15 +++++++++++++
 5 files changed, 69 insertions(+), 7 deletions(-)
include/linux/cgroup-defs.h +4 -0
index 1bfcfb1af352..025fd0e84a31 100644
@@ -287,6 +287,10 @@ struct css_set {
 
 struct cgroup_base_stat {
 	struct task_cputime cputime;
+
+#ifdef CONFIG_SCHED_CORE
+	u64 forceidle_sum;
+#endif
 };
 
 /*
include/linux/kernel_stat.h +7 -0
index 69ae6b278464..ddb5a358fd82 100644
@@ -28,6 +28,9 @@ enum cpu_usage_stat {
 	CPUTIME_STEAL,
 	CPUTIME_GUEST,
 	CPUTIME_GUEST_NICE,
+#ifdef CONFIG_SCHED_CORE
+	CPUTIME_FORCEIDLE,
+#endif
 	NR_STATS,
 };
 
@@ -115,4 +118,8 @@ extern void account_process_tick(struct task_struct *, int user);
 
 extern void account_idle_ticks(unsigned long ticks);
 
+#ifdef CONFIG_SCHED_CORE
+extern void __account_forceidle_time(struct task_struct *tsk, u64 delta);
+#endif
+
 #endif /* _LINUX_KERNEL_STAT_H */
kernel/cgroup/rstat.c +38 -6

97 lines not on this page · raw message

kernel/sched/core_sched.c +5 -1
index 38a2cec21014..5103502da7ba 100644
@@ -277,7 +277,11 @@ void __sched_core_account_forceidle(struct rq *rq)
 		if (p == rq_i->idle)
 			continue;
 
-		__schedstat_add(p->stats.core_forceidle_sum, delta);
+		/*
+		 * Note: this will account forceidle to the current cpu, even
+		 * if it comes from our SMT sibling.
+		 */
+		__account_forceidle_time(p, delta);
 	}
 }
 
kernel/sched/cputime.c +15 -0
index 78a233d43757..95fc77853743 100644
@@ -226,6 +226,21 @@ void account_idle_time(u64 cputime)
 		cpustat[CPUTIME_IDLE] += cputime;
 }
 
+
+#ifdef CONFIG_SCHED_CORE
+/*
+ * Account for forceidle time due to core scheduling.
+ *
+ * REQUIRES: schedstat is enabled.
+ */
+void __account_forceidle_time(struct task_struct *p, u64 delta)
+{
+	__schedstat_add(p->stats.core_forceidle_sum, delta);
+
+	task_group_account_field(p, CPUTIME_FORCEIDLE, delta);
+}
+#endif
+
 /*
  * When a guest is interrupted for a longer amount of time, missed clock
  * ticks are not redelivered later. Due to that, this function may on
Signature · 1 line
2.37.0.rc0.104.g0611611a94-goog
maintainer
#1 lore raw

Hello,

Quoted 9 lines · Josh Don

4feee7d1260 previously added per-task forced idle accounting. This patch extends this to also include cgroups.

rstat is used for cgroup accounting, except for the root, which uses kcpustat in order to bypass the need for doing an rstat flush when reading root stats.

Only cgroup v2 is supported. Similar to the task accounting, the cgroup accounting requires that schedstats is enabled.

Generally looks fine from cgroup side. One nitpick follow.

kernel/cgroup/rstat.c void cgroup_base_stat_cputime_show(struct seq_file *seq) · line 501

Quoted 1 line
kernel/cgroup/rstat.c +1 -0
+	seq_printf(seq, "forceidle_usec %llu\n", forceidle_time);

Would it make sense to namespace the name to reflect the fact that this is tied to core scheduling? e.g. something like core.force_idle_usec (and yeah, underscore between words, please). I kinda hate that the feature is named "core". The word is so overloaded.

Thanks.

Signature · 1 line
tejun
author
#2 lore raw

Thanks Tejun,

Quoted 16 lines · Tejun Heo

Hello,

Quoted 9 lines · Josh Don

4feee7d1260 previously added per-task forced idle accounting. This patch extends this to also include cgroups.

rstat is used for cgroup accounting, except for the root, which uses kcpustat in order to bypass the need for doing an rstat flush when reading root stats.

Only cgroup v2 is supported. Similar to the task accounting, the cgroup accounting requires that schedstats is enabled.

Generally looks fine from cgroup side. One nitpick follow.

Quoted 7 lines
Quoted 1 line

+ seq_printf(seq, "forceidle_usec %llu\n", forceidle_time);

Would it make sense to namespace the name to reflect the fact that this is tied to core scheduling? e.g. something like core.force_idle_usec (and yeah, underscore between words, please). I kinda hate that the feature is named "core". The word is so overloaded.

Sure, although a namespace of "core_sched" would be a bit clearer, since as you point out "core" is pretty overloaded :)

Lack of underscore for forceidle was to be consistent with "core_forceidle_sum" being dumped from /proc/pid/sched, but I'm fine with it either way.

So,

core_sched.force_idle_usec ?

Quoted 5 lines

Thanks.

--
tejun

maintainer
#3 lore raw

Hello,

Quoted 8 lines · Josh Don
Quoted 5 lines

Would it make sense to namespace the name to reflect the fact that this is tied to core scheduling? e.g. something like core.force_idle_usec (and yeah, underscore between words, please). I kinda hate that the feature is named "core". The word is so overloaded.

Sure, although a namespace of "core_sched" would be a bit clearer, since as you point out "core" is pretty overloaded :)

Yeah, core_sched is probabaly better and hopefully it'll get better as we get used to the name.

Quoted 7 lines

Lack of underscore for forceidle was to be consistent with "core_forceidle_sum" being dumped from /proc/pid/sched, but I'm fine with it either way.

So,

core_sched.force_idle_usec ?

Yeah, that'd be my preference.

Thanks.

Signature · 1 line
tejun