[PATCH] psi: Add PSI_CPU_FULL state

From: Chengming Zhou <zhouchengming@bytedance.com>
To: linux-kernel@vger.kernel.org
Cc: hannes@cmpxchg.org, mingo@redhat.com, peterz@infradead.org,
	juri.lelli@redhat.com, vincent.guittot@linaro.org,
	dietmar.eggemann@arm.com, rostedt@goodmis.org,
	bsegall@google.com, mgorman@suse.de, bristot@redhat.com,
	zhouchengming@bytedance.com
Subject: [PATCH] psi: Add PSI_CPU_FULL state
Date: Fri, 25 Dec 2020 12:22:48 +0800	[thread overview]
Message-ID: <20201225042248.51737-1-zhouchengming@bytedance.com> (raw)

From: zhouchengming <zhouchengming@bytedance.com>

When I run a simple "perf bench sched pipe" test in a cgroup on my
machine, the output of "watch -d -n 1 cpu.pressure" of this cgroup
will report some avg10 10%-20%.
It's strange because there is not any other process in the cgroup.
Then I found that cpu contention/wait percentage came from outside
of the cgroup and the cpu idle latency, not wait for threads in
the cgroup.
So I think adding PSI_CPU_FULL state will be useful for container
workloads to distinguish between outside cgroup resource contention
and inside cgroup resource contention.
What's more, the PSI_CPU_FULL state includes the latency of the cpu
idle itself, so we can see the system sum latency introduced by cpu
idle state.

Signed-off-by: zhouchengming <zhouchengming@bytedance.com>
---
 include/linux/psi_types.h | 3 ++-
 kernel/sched/psi.c        | 9 +++++++--
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/include/linux/psi_types.h b/include/linux/psi_types.h
index b95f3211566a..0a23300d49af 100644
--- a/include/linux/psi_types.h
+++ b/include/linux/psi_types.h
@@ -50,9 +50,10 @@ enum psi_states {
 	PSI_MEM_SOME,
 	PSI_MEM_FULL,
 	PSI_CPU_SOME,
+	PSI_CPU_FULL,
 	/* Only per-CPU, to weigh the CPU in the global average: */
 	PSI_NONIDLE,
-	NR_PSI_STATES = 6,
+	NR_PSI_STATES = 7,
 };
 
 enum psi_aggregators {
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index 967732c0766c..234047e368a5 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -224,7 +224,9 @@ static bool test_state(unsigned int *tasks, enum psi_states state)
 	case PSI_MEM_FULL:
 		return tasks[NR_MEMSTALL] && !tasks[NR_RUNNING];
 	case PSI_CPU_SOME:
-		return tasks[NR_RUNNING] > tasks[NR_ONCPU];
+		return tasks[NR_RUNNING] > tasks[NR_ONCPU] && tasks[NR_ONCPU];
+	case PSI_CPU_FULL:
+		return tasks[NR_RUNNING] && !tasks[NR_ONCPU];
 	case PSI_NONIDLE:
 		return tasks[NR_IOWAIT] || tasks[NR_MEMSTALL] ||
 			tasks[NR_RUNNING];
@@ -681,6 +683,9 @@ static void record_times(struct psi_group_cpu *groupc, int cpu,
 	if (groupc->state_mask & (1 << PSI_CPU_SOME))
 		groupc->times[PSI_CPU_SOME] += delta;
 
+	if (groupc->state_mask & (1 << PSI_CPU_FULL))
+		groupc->times[PSI_CPU_FULL] += delta;
+
 	if (groupc->state_mask & (1 << PSI_NONIDLE))
 		groupc->times[PSI_NONIDLE] += delta;
 }
@@ -1018,7 +1023,7 @@ int psi_show(struct seq_file *m, struct psi_group *group, enum psi_res res)
 		group->avg_next_update = update_averages(group, now);
 	mutex_unlock(&group->avgs_lock);
 
-	for (full = 0; full < 2 - (res == PSI_CPU); full++) {
+	for (full = 0; full < 2; full++) {
 		unsigned long avg[3];
 		u64 total;
 		int w;
-- 
2.11.0

