public inbox for [email protected]
 help / color / mirror / Atom feed
From: Waiman Long <[email protected]>
To: Tejun Heo <[email protected]>, Zefan Li <[email protected]>,
	Johannes Weiner <[email protected]>,
	Jonathan Corbet <[email protected]>, Shuah Khan <[email protected]>
Cc: [email protected], [email protected],
	[email protected], [email protected],
	"Andrew Morton" <[email protected]>,
	"Roman Gushchin" <[email protected]>, "Phil Auld" <[email protected]>,
	"Peter Zijlstra" <[email protected]>,
	"Juri Lelli" <[email protected]>,
	"Frederic Weisbecker" <[email protected]>,
	"Marcelo Tosatti" <[email protected]>,
	"Michal Koutný" <[email protected]>,
	"Waiman Long" <[email protected]>
Subject: [PATCH v10 5/8] cgroup/cpuset: Add a new isolated cpus.partition type
Date: Tue,  3 May 2022 12:21:46 -0400	[thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>

Cpuset v1 uses the sched_load_balance control file to determine if load
balancing should be enabled.  Cpuset v2 gets rid of sched_load_balance
as its use may require disabling load balancing at cgroup root.

For workloads that require very low latency like DPDK, the latency
jitters caused by periodic load balancing may exceed the desired
latency limit.

When cpuset v2 is in use, the only way to avoid this latency cost is to
use the "isolcpus=" kernel boot option to isolate a set of CPUs. After
the kernel boot, however, there is no way to add or remove CPUs from
this isolated set. For workloads that are more dynamic in nature, that
means users have to provision enough CPUs for the worst case situation
resulting in excess idle CPUs.

To address this issue for cpuset v2, a new cpuset.cpus.partition type
"isolated" is added which allows the creation of a cpuset partition
without load balancing. This will allow system administrators to
dynamically adjust the size of isolated partition to the current need
of the workload without rebooting the system.

Signed-off-by: Waiman Long <[email protected]>
---
 kernel/cgroup/cpuset.c | 74 +++++++++++++++++++++++++++++++++++-------
 1 file changed, 63 insertions(+), 11 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 1f1d76013046..073db69ac8fc 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -177,11 +177,15 @@ struct cpuset {
  *
  *   0 - member (not a partition root)
  *   1 - partition root
+ *   2 - partition root without load balancing (isolated)
  *  -1 - invalid partition root
+ *  -2 - invalid isolated partition root
  */
 #define PRS_MEMBER		0
 #define PRS_ROOT		1
+#define PRS_ISOLATED		2
 #define PRS_INVALID_ROOT	-1
+#define PRS_INVALID_ISOLATED	-2
 
 static inline bool is_prs_invalid(int prs_state)
 {
@@ -278,7 +282,8 @@ static inline int is_partition_invalid(const struct cpuset *cs)
 
 static inline void set_partition_invalid(struct cpuset *cs)
 {
-	cs->partition_root_state = PRS_INVALID_ROOT;
+	if (is_partition_valid(cs))
+		cs->partition_root_state = -cs->partition_root_state;
 }
 
 /*
@@ -1369,17 +1374,19 @@ static int update_parent_subparts_cpumask(struct cpuset *cs, int cmd,
 
 	if (cmd == partcmd_update) {
 		/*
-		 * Check for possible transition between PRS_ROOT
-		 * and PRS_INVALID_ROOT.
+		 * Check for possible transition between valid and invalid
+		 * partition root.
 		 */
 		switch (cs->partition_root_state) {
 		case PRS_ROOT:
+		case PRS_ISOLATED:
 			if (part_error)
-				new_prs = PRS_INVALID_ROOT;
+				new_prs = -old_prs;
 			break;
 		case PRS_INVALID_ROOT:
+		case PRS_INVALID_ISOLATED:
 			if (!part_error)
-				new_prs = PRS_ROOT;
+				new_prs = -old_prs;
 			break;
 		}
 	}
@@ -1389,7 +1396,7 @@ static int update_parent_subparts_cpumask(struct cpuset *cs, int cmd,
 
 	/*
 	 * Transitioning from invalid to valid (partcmd_update) may require
-	 * setting CS_CPU_EXCLUSIVE.
+	 * setting CS_CPU_EXCLUSIVE and clearing CS_SCHED_LOAD_BALANCE later.
 	 */
 	if ((old_prs != new_prs) && is_prs_invalid(old_prs)) {
 		if (!is_cpu_exclusive(cs) &&
@@ -1430,8 +1437,17 @@ static int update_parent_subparts_cpumask(struct cpuset *cs, int cmd,
 	if (adding || deleting)
 		update_tasks_cpumask(parent);
 
+	/*
+	 * Set or clear CS_SCHED_LOAD_BALANCE when partcmd_update, if necessary.
+	 * rebuild_sched_domains_locked() may be called.
+	 */
+	if ((old_prs != new_prs) && (cmd == partcmd_update)) {
+		if (old_prs == PRS_ISOLATED)
+			update_flag(CS_SCHED_LOAD_BALANCE, cs, 1);
+		else if (new_prs == PRS_ISOLATED)
+			update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
+	}
 	notify_partition_change(cs, old_prs);
-
 	return 0;
 }
 
@@ -1506,6 +1522,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 		if ((cp != cs) && old_prs) {
 			switch (parent->partition_root_state) {
 			case PRS_ROOT:
+			case PRS_ISOLATED:
 				update_parent = true;
 				break;
 
@@ -1515,7 +1532,8 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 				 * invalid, child partition roots become
 				 * invalid too.
 				 */
-				new_prs = PRS_INVALID_ROOT;
+				if (is_partition_valid(cp))
+					new_prs = -cp->partition_root_state;
 				break;
 			}
 		}
@@ -2097,6 +2115,7 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 static int update_prstate(struct cpuset *cs, int new_prs)
 {
 	int err = 0, old_prs = cs->partition_root_state;
+	bool sched_domain_rebuilt = false;
 	struct cpuset *parent = parent_cs(cs);
 	struct tmpmasks tmpmask;
 
@@ -2107,8 +2126,10 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	 * For a previously invalid partition root, leave it at being
 	 * invalid if new_prs is not "member".
 	 */
-	if (new_prs && is_prs_invalid(old_prs))
+	if (new_prs && is_prs_invalid(old_prs)) {
+		cs->partition_root_state = -new_prs;
 		return 0;
+	}
 
 	if (alloc_cpumasks(NULL, &tmpmask))
 		return -ENOMEM;
@@ -2134,6 +2155,22 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 			update_flag(CS_CPU_EXCLUSIVE, cs, 0);
 			goto out;
 		}
+
+		if (new_prs == PRS_ISOLATED) {
+			/*
+			 * Disable the load balance flag should not return an
+			 * error unless the system is running out of memory.
+			 */
+			update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
+			sched_domain_rebuilt = true;
+		}
+	} else if (old_prs && new_prs) {
+		/*
+		 * A change in load balance state only, no change in cpumasks.
+		 */
+		update_flag(CS_SCHED_LOAD_BALANCE, cs, (new_prs != PRS_ISOLATED));
+		sched_domain_rebuilt = true;
+		goto out;	/* Sched domain is rebuilt in update_flag() */
 	} else {
 		/*
 		 * Switching back to member is always allowed even if it
@@ -2155,6 +2192,12 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 
 		/* Turning off CS_CPU_EXCLUSIVE will not return error */
 		update_flag(CS_CPU_EXCLUSIVE, cs, 0);
+
+		if (!is_sched_load_balance(cs)) {
+			/* Make sure load balance is on */
+			update_flag(CS_SCHED_LOAD_BALANCE, cs, 1);
+			sched_domain_rebuilt = true;
+		}
 	}
 
 	update_tasks_cpumask(parent);
@@ -2162,13 +2205,14 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	if (parent->child_ecpus_count)
 		update_sibling_cpumasks(parent, cs, &tmpmask);
 
-	rebuild_sched_domains_locked();
+	if (!sched_domain_rebuilt)
+		rebuild_sched_domains_locked();
 out:
 	/*
 	 * Make partition invalid if an error happen
 	 */
 	if (err)
-		new_prs = PRS_INVALID_ROOT;
+		new_prs = -new_prs;
 	spin_lock_irq(&callback_lock);
 	cs->partition_root_state = new_prs;
 	spin_unlock_irq(&callback_lock);
@@ -2680,12 +2724,18 @@ static int sched_partition_show(struct seq_file *seq, void *v)
 	case PRS_ROOT:
 		seq_puts(seq, "root\n");
 		break;
+	case PRS_ISOLATED:
+		seq_puts(seq, "isolated\n");
+		break;
 	case PRS_MEMBER:
 		seq_puts(seq, "member\n");
 		break;
 	case PRS_INVALID_ROOT:
 		seq_puts(seq, "root invalid\n");
 		break;
+	case PRS_INVALID_ISOLATED:
+		seq_puts(seq, "isolated invalid\n");
+		break;
 	}
 	return 0;
 }
@@ -2706,6 +2756,8 @@ static ssize_t sched_partition_write(struct kernfs_open_file *of, char *buf,
 		val = PRS_ROOT;
 	else if (!strcmp(buf, "member"))
 		val = PRS_MEMBER;
+	else if (!strcmp(buf, "isolated"))
+		val = PRS_ISOLATED;
 	else
 		return -EINVAL;
 
-- 
2.27.0



  parent reply	other threads:[~2022-05-03 16:23 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-05-03 16:21 [PATCH v10 0/8] cgroup/cpuset: Major cpu partition code restructuring Waiman Long
2022-05-03 16:21 ` [PATCH v10 1/8] cgroup/cpuset: Add top_cpuset check in update_tasks_cpumask() Waiman Long
2022-05-03 16:21 ` [PATCH v10 2/8] cgroup/cpuset: Miscellaneous cleanups & add helper functions Waiman Long
2022-05-03 16:21 ` [PATCH v10 3/8] cgroup/cpuset: Allow no-task partition to have empty cpuset.cpus.effective Waiman Long
2022-05-03 16:21 ` [PATCH v10 4/8] cgroup/cpuset: Relax constraints to partition & cpus changes Waiman Long
2022-05-03 16:21 ` Waiman Long [this message]
2022-05-03 16:21 ` [PATCH v10 6/8] cgroup/cpuset: Show invalid partition reason string Waiman Long
2022-05-03 16:21 ` [PATCH v10 7/8] cgroup/cpuset: Update description of cpuset.cpus.partition in cgroup-v2.rst Waiman Long
2022-05-03 16:21 ` [PATCH v10 8/8] kselftest/cgroup: Add cpuset v2 partition root state test Waiman Long

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox