Re: [PATCH 1/7] sched: warn for high latency with TIF_NEED_RESCHED_LAZY

From: Shrikanth Hegde
Date: Tue Oct 22 2024 - 01:42:44 EST




On 10/22/24 00:51, Ankur Arora wrote:

Shrikanth Hegde <sshegde@xxxxxxxxxxxxx> writes:

On 10/9/24 22:24, Ankur Arora wrote:
resched_latency_warn() now also warns if TIF_NEED_RESCHED_LAZY is set
without rescheduling for more than the latency_warn_ms period.


I am bit confused here. Why do we need to warn if LAZY is set for a long time?

If lazy set, the subsequent tick, it would be set to upgraded to NEED_RESCHED.

Since the value of latency_warn_ms=100ms, that means even on system with HZ=100,
that means 10 ticks before that warning would be printed no?

That's a fair point. However, the assumption there is that there are no
bugs in upgrade on tick or that there's no situation in which the tick
is off for a prolonged period.


ok.

But if tick is off, then ticks_without_resched isn't incremented either. IIUC, this check is for situation when NR is set and tick is on.

Ankur

IIUC, the changelog c006fac556e40 ("sched: Warn on long periods of pending
need_resched") has the concern of need_resched set but if it is non-preemptible
kernel it would spend a lot of time in kernel mode. In that case print a
warning.

If someone enables Lazy, that means it is preemptible and probably this whole
notion of resched_latency_warn doesn't apply to lazy. Please correct me if i am
not understanding this correctly.

Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Peter Ziljstra <peterz@xxxxxxxxxxxxx>
Cc: Juri Lelli <juri.lelli@xxxxxxxxxx>
Cc: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
Signed-off-by: Ankur Arora <ankur.a.arora@xxxxxxxxxx>
---
kernel/sched/core.c | 2 +-
kernel/sched/debug.c | 7 +++++--
2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 694bfcf153cb..1229766b704e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5571,7 +5571,7 @@ static u64 cpu_resched_latency(struct rq *rq)
if (sysctl_resched_latency_warn_once && warned_once)
return 0;
- if (!need_resched() || !latency_warn_ms)
+ if ((!need_resched() && !tif_need_resched_lazy()) || !latency_warn_ms)
return 0;
if (system_state == SYSTEM_BOOTING)
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index 9abcc6ead11b..f0d551ba64bb 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -1293,9 +1293,12 @@ void proc_sched_set_task(struct task_struct *p)
void resched_latency_warn(int cpu, u64 latency)
{
static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
+ char *nr;
+
+ nr = tif_need_resched() ? "need_resched" : "need_resched_lazy";
WARN(__ratelimit(&latency_check_ratelimit),
- "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
+ "sched: CPU %d %s set for > %llu ns (%d ticks) "
"without schedule\n",
- cpu, latency, cpu_rq(cpu)->ticks_without_resched);
+ cpu, nr, latency, cpu_rq(cpu)->ticks_without_resched);
}


--
ankur