Skip to contentSkip to conversation

docs/scheduler: Change unit of cpu_time and rq_time to nanoseconds

PATCH In review brookxu.cn 4 participants 4 messages Apr 8, 2022 – Apr 16, 2022 Documentation

lore
Share & tools
lore

brookxu.cn

author
#0 lore raw

PATCH 1/1 docs/scheduler: Change unit of cpu_time and rq_time to nanoseconds

From: Chunguang Xu <[email protected]>

In the current implementation, cpu_time and rq_time should be in nanoseconds, so this document needs to be modified.

---
 Documentation/scheduler/sched-stats.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Documentation/scheduler/sched-stats.rst +2 -2
index dd9b99a..9078a27 100644
@@ -56,9 +56,9 @@ Next two are try_to_wake_up() statistics:
 
 Next three are statistics describing scheduling latency:
 
-     7) sum of all time spent running by tasks on this processor (in jiffies)
+     7) sum of all time spent running by tasks on this processor (in nanoseconds)
      8) sum of all time spent waiting to run by tasks on this processor (in
-        jiffies)
+        nanoseconds)
      9) # of timeslices run on this cpu
 
 
Signature · 1 line
1.8.3.1
#1 lore raw
Quoted 4 lines · brookxu.cn

From: Chunguang Xu <[email protected]>

In the current implementation, cpu_time and rq_time should be in nanoseconds, so this document needs to be modified.

I agree that this is wrong, but we shouldn't be changing the units of measurement reported to userspace without changing the schedstats version. I suspect this was an inadvertent change, and we should be converting from sched_clock() time (~= ns) to jiffies in show_schedstat(). Adding scheduler experts.

Quoted 24 lines
---
 Documentation/scheduler/sched-stats.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Documentation/scheduler/sched-stats.rst +2 -2
index dd9b99a..9078a27 100644
@@ -56,9 +56,9 @@ Next two are try_to_wake_up() statistics:
 
 Next three are statistics describing scheduling latency:
 
-     7) sum of all time spent running by tasks on this processor (in jiffies)
+     7) sum of all time spent running by tasks on this processor (in nanoseconds)
      8) sum of all time spent waiting to run by tasks on this processor (in
-        jiffies)
+        nanoseconds)
      9) # of timeslices run on this cpu
 
 
Signature · 1 line
1.8.3.1
#2 lore raw
Quoted 11 lines · Matthew Wilcox
Quoted 4 lines · brookxu.cn

From: Chunguang Xu <[email protected]>

In the current implementation, cpu_time and rq_time should be in nanoseconds, so this document needs to be modified.

I agree that this is wrong, but we shouldn't be changing the units of measurement reported to userspace without changing the schedstats version. I suspect this was an inadvertent change, and we should be converting from sched_clock() time (~= ns) to jiffies in show_schedstat(). Adding scheduler experts.

Seems to have happend at 425e0968a25f ("sched: move code into kernel/sched_stats.h").

Before that we have:

-static inline void
-rq_sched_info_depart(struct rq *rq, unsigned long delta_jiffies) -{
- if (rq)
- rq->rq_sched_info.cpu_time += delta_jiffies;
-}

-static inline void sched_info_depart(struct task_struct *t) -{
- unsigned long delta_jiffies = jiffies - t->sched_info.last_arrival;
-
- t->sched_info.cpu_time += delta_jiffies;
- rq_sched_info_depart(task_rq(t), delta_jiffies);
-}

afterwards we have:

+static inline void sched_info_depart(struct task_struct *t) +{
+ unsigned long long delta = sched_clock() - t->sched_info.last_arrival;
+
+ t->sched_info.cpu_time += delta;
+ rq_sched_info_depart(task_rq(t), delta);
+}

And that's 15 years and at least one SCHEDSTAT_VERSION ago (although this change itself didn't bump the version).

So I'm thinking we can update the documentation and forget about this.

maintainer
#3 lore raw
Quoted 4 lines · Peter Zijlstra

And that's 15 years and at least one SCHEDSTAT_VERSION ago (although this change itself didn't bump the version).

So I'm thinking we can update the documentation and forget about this.

OK, I've applied the patch, thanks.

jon