Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 31 Aug 2022 15:49:03 -0700
Mime-Version: 1.0
Message-ID: <20220831224903.454303-1-joshdon@google.com>
Subject: [PATCH] cgroup: add pids.peak interface for pids controller
From:   Josh Don <joshdon@google.com>
To:     Tejun Heo <tj@kernel.org>, Zefan Li <lizefan.x@bytedance.com>,
        Johannes Weiner <hannes@cmpxchg.org>
Cc:     cgroups@vger.kernel.org, linux-kernel@vger.kernel.org,
        Josh Don <joshdon@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

pids.peak tracks the high watermark of usage for number of pids. This
helps give a better baseline on which to set pids.max. Polling
pids.current isn't really feasible, since it would potentially miss
short-lived spikes.

This interface is analogous to memory.peak.

Signed-off-by: Josh Don <joshdon@google.com>
---
 kernel/cgroup/pids.c | 37 +++++++++++++++++++++++++++++++++++--
 1 file changed, 35 insertions(+), 2 deletions(-)

diff --git a/kernel/cgroup/pids.c b/kernel/cgroup/pids.c
index 511af87f685e..7695e60bcb40 100644
--- a/kernel/cgroup/pids.c
+++ b/kernel/cgroup/pids.c
@@ -47,6 +47,7 @@ struct pids_cgroup {
 	 */
 	atomic64_t			counter;
 	atomic64_t			limit;
+	int64_t				watermark;
 
 	/* Handle for "pids.events" */
 	struct cgroup_file		events_file;
@@ -85,6 +86,16 @@ static void pids_css_free(struct cgroup_subsys_state *css)
 	kfree(css_pids(css));
 }
 
+static void pids_update_watermark(struct pids_cgroup *p, int64_t nr_pids)
+{
+	/*
+	 * This is racy, but we don't need perfectly accurate tallying of
+	 * the watermark, and this lets us avoid extra atomic overhead.
+	 */
+	if (nr_pids > READ_ONCE(p->watermark))
+		WRITE_ONCE(p->watermark, nr_pids);
+}
+
 /**
  * pids_cancel - uncharge the local pid count
  * @pids: the pid cgroup state
@@ -128,8 +139,11 @@ static void pids_charge(struct pids_cgroup *pids, int num)
 {
 	struct pids_cgroup *p;
 
-	for (p = pids; parent_pids(p); p = parent_pids(p))
-		atomic64_add(num, &p->counter);
+	for (p = pids; parent_pids(p); p = parent_pids(p)) {
+		int64_t new = atomic64_add_return(num, &p->counter);
+
+		pids_update_watermark(p, new);
+	}
 }
 
 /**
@@ -156,6 +170,12 @@ static int pids_try_charge(struct pids_cgroup *pids, int num)
 		 */
 		if (new > limit)
 			goto revert;
+
+		/*
+		 * Not technically accurate if we go over limit somewhere up
+		 * the hierarchy, but that's tolerable for the watermark.
+		 */
+		pids_update_watermark(p, new);
 	}
 
 	return 0;
@@ -311,6 +331,14 @@ static s64 pids_current_read(struct cgroup_subsys_state *css,
 	return atomic64_read(&pids->counter);
 }
 
+static s64 pids_peak_read(struct cgroup_subsys_state *css,
+			  struct cftype *cft)
+{
+	struct pids_cgroup *pids = css_pids(css);
+
+	return READ_ONCE(pids->watermark);
+}
+
 static int pids_events_show(struct seq_file *sf, void *v)
 {
 	struct pids_cgroup *pids = css_pids(seq_css(sf));
@@ -331,6 +359,11 @@ static struct cftype pids_files[] = {
 		.read_s64 = pids_current_read,
 		.flags = CFTYPE_NOT_ON_ROOT,
 	},
+	{
+		.name = "peak",
+		.flags = CFTYPE_NOT_ON_ROOT,
+		.read_s64 = pids_peak_read,
+	},
 	{
 		.name = "events",
 		.seq_show = pids_events_show,
-- 
2.37.2.672.g94769d06f0-goog