Skip to contentSkip to conversation

sched/rt: Add pid and comm when RT throttling activated

PATCH In review Yajun Deng 2 participants 3 messages Dec 6, 2022 – Dec 12, 2022 Scheduler

lore
Share & tools
lore

Yajun Deng

author
#0 lore raw

PATCH 1/1 sched/rt: Add pid and comm when RT throttling activated

It is difficult to know which process consumes time when RT throttling activated.

So add pid and comm for this case.

---
 kernel/sched/rt.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
kernel/sched/rt.c +3 -1
index ed2a47e4ddae..1f267ab1b59a 100644
@@ -1012,6 +1012,7 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 
 	if (rt_rq->rt_time > runtime) {
 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
+		struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
 
 		/*
 		 * Don't actually throttle groups that have no runtime assigned
@@ -1019,7 +1020,8 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 		 */
 		if (likely(rt_b->rt_runtime)) {
 			rt_rq->rt_throttled = 1;
-			printk_deferred_once("sched: RT throttling activated\n");
+			printk_deferred_once("pid: %d, comm: %s, sched: RT throttling activated\n",
+					     curr->pid, curr->comm);
 		} else {
 			/*
 			 * In case we did anyway, make it go away,
Signature · 1 line
2.25.1
#1 lore raw
Quoted 5 lines · On Tue, 6 Dec 2022 16:36:41 +0800 Yajun Deng <[email protected]> wrote:

It is difficult to know which process consumes time when RT throttling activated.

So add pid and comm for this case.

Unfortunately, that only shows the process that "broke the camel's back"[1]. The things is, it is the amount of time *all* RT tasks use to trigger throttling. You are only printing the task that the throttling occurred on.

Try running https://rostedt.org/code/migrate.c . It will create a bunch of RT tasks, and they all will throttle once it hits 950ms ever second.

trace-cmd record -e sched_switch ./migrate
kernelshark

Then plot all the migrate tasks.

The above will demonstrate it very well.

-- Steve

[1] https://en.wikipedia.org/wiki/Straw_that_broke_the_camel%27s_back

Quoted 27 lines
---
 kernel/sched/rt.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
kernel/sched/rt.c +3 -1
index ed2a47e4ddae..1f267ab1b59a 100644
@@ -1012,6 +1012,7 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 
 	if (rt_rq->rt_time > runtime) {
 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
+		struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
 
 		/*
 		 * Don't actually throttle groups that have no runtime assigned
@@ -1019,7 +1020,8 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 		 */
 		if (likely(rt_b->rt_runtime)) {
 			rt_rq->rt_throttled = 1;
-			printk_deferred_once("sched: RT throttling activated\n");
+			printk_deferred_once("pid: %d, comm: %s, sched: RT throttling activated\n",
+					     curr->pid, curr->comm);
 		} else {
 			/*
 			 * In case we did anyway, make it go away,
author
#2 lore raw
Quoted 16 lines · Steven Rostedt
Quoted 4 lines · On Tue, 6 Dec 2022 16:36:41 +0800 Yajun Deng <[email protected]> wrote:

It is difficult to know which process consumes time when RT throttling activated.

So add pid and comm for this case.

Unfortunately, that only shows the process that "broke the camel's back"[1]. The things is, it is the amount of time *all* RT tasks use to trigger throttling. You are only printing the task that the throttling occurred on.

Try running https://rostedt.org/code/migrate.c . It will create a bunch of RT tasks, and they all will throttle once it hits 950ms ever second.

Quoted 6 lines

trace-cmd record -e sched_switch ./migrate
kernelshark

Then plot all the migrate tasks.

The above will demonstrate it very well.

Thanks for your patient answer, I'll test it.

Quoted 32 lines

-- Steve

[1] https://en.wikipedia.org/wiki/Straw_that_broke_the_camel's_back

Quoted 27 lines

---
kernel/sched/rt.c | 4 +++-

1 file changed, 3 insertions(+), 1 deletion(-)
kernel/sched/rt.c +3 -1
index ed2a47e4ddae..1f267ab1b59a 100644
@@ -1012,6 +1012,7 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 
 if (rt_rq->rt_time > runtime) {
 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
+ struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
 
 /*
 * Don't actually throttle groups that have no runtime assigned
@@ -1019,7 +1020,8 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 */
 if (likely(rt_b->rt_runtime)) {
 rt_rq->rt_throttled = 1;
- printk_deferred_once("sched: RT throttling activated\n");
+ printk_deferred_once("pid: %d, comm: %s, sched: RT throttling activated\n",
+ curr->pid, curr->comm);
 } else {
 /*
 * In case we did anyway, make it go away,