[PATCH 0/3] cgroup/cpuset: Remove redundant rebuild_sched_domains_locked() calls

[PATCH 0/3] cgroup/cpuset: Remove redundant rebuild_sched_domains_locked() calls

[Waiman Long]

The main purpose of this series is to make sure that there will
be at most one rebuild_sched_domains_locked() call per cpuset
operation. Juri found that is still not the case. Make use of the
existing cpuset_force_rebuild() helper as a replacement for a direct
rebuild_sched_domains_locked() call and check force_sd_rebuild at the
operation end.

The last patch is for removing some unnecessary v1 specific code if
CONFIG_CPUSETS_V1 isn't set.

Waiman Long (3):
  cgroup/cpuset: Revert "Allow suppression of sched domain rebuild in
    update_cpumasks_hier()"
  cgroup/cpuset: Enforce at most one rebuild_sched_domains_locked() call
    per operation
  cgroup/cpuset: Further optimize code if CONFIG_CPUSETS_V1 not set

 kernel/cgroup/cpuset.c | 121 +++++++++++++++++++++--------------------
 1 file changed, 63 insertions(+), 58 deletions(-)

-- 
2.47.0

[PATCH 1/3] cgroup/cpuset: Revert "Allow suppression of sched domain rebuild in update_cpumasks_hier()"

[Waiman Long]

Revert commit 3ae0b773211e ("cgroup/cpuset: Allow suppression of sched
domain rebuild in update_cpumasks_hier()") to allow for an alternative
way to suppress unnecessary rebuild_sched_domains_locked() calls in
update_cpumasks_hier() and elsewhere in a following commit.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 39 ++++++++++++++-------------------------
 1 file changed, 14 insertions(+), 25 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index a4dd285cdf39..565280193922 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -1922,12 +1922,6 @@ static void compute_partition_effective_cpumask(struct cpuset *cs,
 	rcu_read_unlock();
 }
 
-/*
- * update_cpumasks_hier() flags
- */
-#define HIER_CHECKALL		0x01	/* Check all cpusets with no skipping */
-#define HIER_NO_SD_REBUILD	0x02	/* Don't rebuild sched domains */
-
 /*
  * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree
  * @cs:  the cpuset to consider
@@ -1942,7 +1936,7 @@ static void compute_partition_effective_cpumask(struct cpuset *cs,
  * Called with cpuset_mutex held
  */
 static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
-				 int flags)
+				 bool force)
 {
 	struct cpuset *cp;
 	struct cgroup_subsys_state *pos_css;
@@ -2007,10 +2001,10 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 		 * Skip the whole subtree if
 		 * 1) the cpumask remains the same,
 		 * 2) has no partition root state,
-		 * 3) HIER_CHECKALL flag not set, and
+		 * 3) force flag not set, and
 		 * 4) for v2 load balance state same as its parent.
 		 */
-		if (!cp->partition_root_state && !(flags & HIER_CHECKALL) &&
+		if (!cp->partition_root_state && !force &&
 		    cpumask_equal(tmp->new_cpus, cp->effective_cpus) &&
 		    (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
 		    (is_sched_load_balance(parent) == is_sched_load_balance(cp)))) {
@@ -2112,8 +2106,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 	}
 	rcu_read_unlock();
 
-	if (need_rebuild_sched_domains && !(flags & HIER_NO_SD_REBUILD) &&
-	    !force_sd_rebuild)
+	if (need_rebuild_sched_domains && !force_sd_rebuild)
 		rebuild_sched_domains_locked();
 }
 
@@ -2141,9 +2134,7 @@ static void update_sibling_cpumasks(struct cpuset *parent, struct cpuset *cs,
 	 * directly.
 	 *
 	 * The update_cpumasks_hier() function may sleep. So we have to
-	 * release the RCU read lock before calling it. HIER_NO_SD_REBUILD
-	 * flag is used to suppress rebuild of sched domains as the callers
-	 * will take care of that.
+	 * release the RCU read lock before calling it.
 	 */
 	rcu_read_lock();
 	cpuset_for_each_child(sibling, pos_css, parent) {
@@ -2159,7 +2150,7 @@ static void update_sibling_cpumasks(struct cpuset *parent, struct cpuset *cs,
 			continue;
 
 		rcu_read_unlock();
-		update_cpumasks_hier(sibling, tmp, HIER_NO_SD_REBUILD);
+		update_cpumasks_hier(sibling, tmp, false);
 		rcu_read_lock();
 		css_put(&sibling->css);
 	}
@@ -2179,7 +2170,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	struct tmpmasks tmp;
 	struct cpuset *parent = parent_cs(cs);
 	bool invalidate = false;
-	int hier_flags = 0;
+	bool force = false;
 	int old_prs = cs->partition_root_state;
 
 	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
@@ -2240,8 +2231,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	 * Check all the descendants in update_cpumasks_hier() if
 	 * effective_xcpus is to be changed.
 	 */
-	if (!cpumask_equal(cs->effective_xcpus, trialcs->effective_xcpus))
-		hier_flags = HIER_CHECKALL;
+	force = !cpumask_equal(cs->effective_xcpus, trialcs->effective_xcpus);
 
 	retval = validate_change(cs, trialcs);
 
@@ -2309,7 +2299,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	spin_unlock_irq(&callback_lock);
 
 	/* effective_cpus/effective_xcpus will be updated here */
-	update_cpumasks_hier(cs, &tmp, hier_flags);
+	update_cpumasks_hier(cs, &tmp, force);
 
 	/* Update CS_SCHED_LOAD_BALANCE and/or sched_domains, if necessary */
 	if (cs->partition_root_state)
@@ -2334,7 +2324,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	struct tmpmasks tmp;
 	struct cpuset *parent = parent_cs(cs);
 	bool invalidate = false;
-	int hier_flags = 0;
+	bool force = false;
 	int old_prs = cs->partition_root_state;
 
 	if (!*buf) {
@@ -2357,8 +2347,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	 * Check all the descendants in update_cpumasks_hier() if
 	 * effective_xcpus is to be changed.
 	 */
-	if (!cpumask_equal(cs->effective_xcpus, trialcs->effective_xcpus))
-		hier_flags = HIER_CHECKALL;
+	force = !cpumask_equal(cs->effective_xcpus, trialcs->effective_xcpus);
 
 	retval = validate_change(cs, trialcs);
 	if (retval)
@@ -2411,8 +2400,8 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	 * of the subtree when it is a valid partition root or effective_xcpus
 	 * is updated.
 	 */
-	if (is_partition_valid(cs) || hier_flags)
-		update_cpumasks_hier(cs, &tmp, hier_flags);
+	if (is_partition_valid(cs) || force)
+		update_cpumasks_hier(cs, &tmp, force);
 
 	/* Update CS_SCHED_LOAD_BALANCE and/or sched_domains, if necessary */
 	if (cs->partition_root_state)
@@ -2853,7 +2842,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	update_unbound_workqueue_cpumask(new_xcpus_state);
 
 	/* Force update if switching back to member */
-	update_cpumasks_hier(cs, &tmpmask, !new_prs ? HIER_CHECKALL : 0);
+	update_cpumasks_hier(cs, &tmpmask, !new_prs);
 
 	/* Update sched domains and load balance flag */
 	update_partition_sd_lb(cs, old_prs);
-- 
2.47.0

[PATCH 2/3] cgroup/cpuset: Enforce at most one rebuild_sched_domains_locked() call per operation

[Waiman Long]

Since commit ff0ce721ec21 ("cgroup/cpuset: Eliminate unncessary
sched domains rebuilds in hotplug"), there is only one
rebuild_sched_domains_locked() call per hotplug operation. However,
writing to the various cpuset control files may still casue more than
one rebuild_sched_domains_locked() call to happen in some cases.

Juri had found that two rebuild_sched_domains_locked() calls in
update_prstate(), one from update_cpumasks_hier() and another one from
update_partition_sd_lb() could cause cpuset partition to be created
with null total_bw for DL tasks. IOW, DL tasks may not be scheduled
correctly in such a partition.

A sample command sequence that can reproduce null total_bw is as
follows.

  # echo Y >/sys/kernel/debug/sched/verbose
  # echo +cpuset >/sys/fs/cgroup/cgroup.subtree_control
  # mkdir /sys/fs/cgroup/test
  # echo 0-7 > /sys/fs/cgroup/test/cpuset.cpus
  # echo 6-7 > /sys/fs/cgroup/test/cpuset.cpus.exclusive
  # echo root >/sys/fs/cgroup/test/cpuset.cpus.partition

Fix this double rebuild_sched_domains_locked() calls problem
by replacing existing calls with cpuset_force_rebuild() except
the rebuild_sched_domains_cpuslocked() call at the end of
cpuset_handle_hotplug(). Checking of the force_sd_rebuild flag is
now done at the end of cpuset_write_resmask() and update_prstate()
to determine if rebuild_sched_domains_locked() should be called or not.

The cpuset v1 code can still call rebuild_sched_domains_locked()
directly as double rebuild_sched_domains_locked() calls is not possible.

Reported-by: Juri Lelli <juri.lelli@redhat.com>
Closes: https://lore.kernel.org/lkml/ZyuUcJDPBln1BK1Y@jlelli-thinkpadt14gen4.remote.csb/
Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 49 ++++++++++++++++++++++++++++--------------
 1 file changed, 33 insertions(+), 16 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 565280193922..0d56a226c522 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -84,9 +84,19 @@ static bool		have_boot_isolcpus;
 static struct list_head remote_children;
 
 /*
- * A flag to force sched domain rebuild at the end of an operation while
- * inhibiting it in the intermediate stages when set. Currently it is only
- * set in hotplug code.
+ * A flag to force sched domain rebuild at the end of an operation.
+ * It can be set in
+ *  - update_partition_sd_lb()
+ *  - remote_partition_check()
+ *  - update_cpumasks_hier()
+ *  - cpuset_update_flag()
+ *  - cpuset_hotplug_update_tasks()
+ *  - cpuset_handle_hotplug()
+ *
+ * Protected by cpuset_mutex (with cpus_read_lock held) or cpus_write_lock.
+ *
+ * Note that update_relax_domain_level() in cpuset-v1.c can still call
+ * rebuild_sched_domains_locked() directly without using this flag.
  */
 static bool force_sd_rebuild;
 
@@ -990,6 +1000,7 @@ void rebuild_sched_domains_locked(void)
 
 	lockdep_assert_cpus_held();
 	lockdep_assert_held(&cpuset_mutex);
+	force_sd_rebuild = false;
 
 	/*
 	 * If we have raced with CPU hotplug, return early to avoid
@@ -1164,8 +1175,8 @@ static void update_partition_sd_lb(struct cpuset *cs, int old_prs)
 			clear_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
 	}
 
-	if (rebuild_domains && !force_sd_rebuild)
-		rebuild_sched_domains_locked();
+	if (rebuild_domains)
+		cpuset_force_rebuild();
 }
 
 /*
@@ -1512,8 +1523,8 @@ static void remote_partition_check(struct cpuset *cs, struct cpumask *newmask,
 			remote_partition_disable(child, tmp);
 			disable_cnt++;
 		}
-	if (disable_cnt && !force_sd_rebuild)
-		rebuild_sched_domains_locked();
+	if (disable_cnt)
+		cpuset_force_rebuild();
 }
 
 /*
@@ -2106,8 +2117,8 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 	}
 	rcu_read_unlock();
 
-	if (need_rebuild_sched_domains && !force_sd_rebuild)
-		rebuild_sched_domains_locked();
+	if (need_rebuild_sched_domains)
+		cpuset_force_rebuild();
 }
 
 /**
@@ -2726,9 +2737,13 @@ int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 	cs->flags = trialcs->flags;
 	spin_unlock_irq(&callback_lock);
 
-	if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed &&
-	    !force_sd_rebuild)
-		rebuild_sched_domains_locked();
+	if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) {
+		if (!IS_ENABLED(CONFIG_CPUSETS_V1) ||
+		    cgroup_subsys_on_dfl(cpuset_cgrp_subsys))
+			cpuset_force_rebuild();
+		else
+			rebuild_sched_domains_locked();
+	}
 
 	if (spread_flag_changed)
 		cpuset1_update_tasks_flags(cs);
@@ -2848,6 +2863,8 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	update_partition_sd_lb(cs, old_prs);
 
 	notify_partition_change(cs, old_prs);
+	if (force_sd_rebuild)
+		rebuild_sched_domains_locked();
 	free_cpumasks(NULL, &tmpmask);
 	return 0;
 }
@@ -3141,6 +3158,8 @@ ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
 	}
 
 	free_cpuset(trialcs);
+	if (force_sd_rebuild)
+		rebuild_sched_domains_locked();
 out_unlock:
 	mutex_unlock(&cpuset_mutex);
 	cpus_read_unlock();
@@ -3885,11 +3904,9 @@ static void cpuset_handle_hotplug(void)
 		rcu_read_unlock();
 	}
 
-	/* rebuild sched domains if cpus_allowed has changed */
-	if (force_sd_rebuild) {
-		force_sd_rebuild = false;
+	/* rebuild sched domains if necessary */
+	if (force_sd_rebuild)
 		rebuild_sched_domains_cpuslocked();
-	}
 
 	free_cpumasks(NULL, ptmp);
 }
-- 
2.47.0

[PATCH 3/3] cgroup/cpuset: Further optimize code if CONFIG_CPUSETS_V1 not set

[Waiman Long]

Currently the cpuset code uses group_subsys_on_dfl() to check if we
are running with cgroup v2. If CONFIG_CPUSETS_V1 isn't set, there is
really no need to do this check and we can optimize out some of the
unneeded v1 specific code paths. Introduce a new cpuset_v2() and use it
to replace the cgroup_subsys_on_dfl() check to further optimize the
code.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 39 +++++++++++++++++++--------------------
 1 file changed, 19 insertions(+), 20 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 0d56a226c522..655396e75b58 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -293,6 +293,12 @@ static inline void dec_attach_in_progress(struct cpuset *cs)
 	mutex_unlock(&cpuset_mutex);
 }
 
+static inline bool cpuset_v2(void)
+{
+	return !IS_ENABLED(CONFIG_CPUSETS_V1) ||
+		cgroup_subsys_on_dfl(cpuset_cgrp_subsys);
+}
+
 /*
  * Cgroup v2 behavior is used on the "cpus" and "mems" control files when
  * on default hierarchy or when the cpuset_v2_mode flag is set by mounting
@@ -303,7 +309,7 @@ static inline void dec_attach_in_progress(struct cpuset *cs)
  */
 static inline bool is_in_v2_mode(void)
 {
-	return cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
+	return cpuset_v2() ||
 	      (cpuset_cgrp_subsys.root->flags & CGRP_ROOT_CPUSET_V2_MODE);
 }
 
@@ -738,7 +744,7 @@ static int generate_sched_domains(cpumask_var_t **domains,
 	int nslot;		/* next empty doms[] struct cpumask slot */
 	struct cgroup_subsys_state *pos_css;
 	bool root_load_balance = is_sched_load_balance(&top_cpuset);
-	bool cgrpv2 = cgroup_subsys_on_dfl(cpuset_cgrp_subsys);
+	bool cgrpv2 = cpuset_v2();
 	int nslot_update;
 
 	doms = NULL;
@@ -1198,7 +1204,7 @@ static void reset_partition_data(struct cpuset *cs)
 {
 	struct cpuset *parent = parent_cs(cs);
 
-	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys))
+	if (!cpuset_v2())
 		return;
 
 	lockdep_assert_held(&callback_lock);
@@ -2017,7 +2023,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 		 */
 		if (!cp->partition_root_state && !force &&
 		    cpumask_equal(tmp->new_cpus, cp->effective_cpus) &&
-		    (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
+		    (!cpuset_v2() ||
 		    (is_sched_load_balance(parent) == is_sched_load_balance(cp)))) {
 			pos_css = css_rightmost_descendant(pos_css);
 			continue;
@@ -2091,8 +2097,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 		 * from parent if current cpuset isn't a valid partition root
 		 * and their load balance states differ.
 		 */
-		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-		    !is_partition_valid(cp) &&
+		if (cpuset_v2() && !is_partition_valid(cp) &&
 		    (is_sched_load_balance(parent) != is_sched_load_balance(cp))) {
 			if (is_sched_load_balance(parent))
 				set_bit(CS_SCHED_LOAD_BALANCE, &cp->flags);
@@ -2108,8 +2113,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 		 */
 		if (!cpumask_empty(cp->cpus_allowed) &&
 		    is_sched_load_balance(cp) &&
-		   (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
-		    is_partition_valid(cp)))
+		   (!cpuset_v2() || is_partition_valid(cp)))
 			need_rebuild_sched_domains = true;
 
 		rcu_read_lock();
@@ -2246,7 +2250,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 
 	retval = validate_change(cs, trialcs);
 
-	if ((retval == -EINVAL) && cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) {
+	if ((retval == -EINVAL) && cpuset_v2()) {
 		struct cgroup_subsys_state *css;
 		struct cpuset *cp;
 
@@ -2738,8 +2742,7 @@ int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 	spin_unlock_irq(&callback_lock);
 
 	if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) {
-		if (!IS_ENABLED(CONFIG_CPUSETS_V1) ||
-		    cgroup_subsys_on_dfl(cpuset_cgrp_subsys))
+		if (cpuset_v2())
 			cpuset_force_rebuild();
 		else
 			rebuild_sched_domains_locked();
@@ -2925,8 +2928,7 @@ static int cpuset_can_attach(struct cgroup_taskset *tset)
 		 * migration permission derives from hierarchy ownership in
 		 * cgroup_procs_write_permission()).
 		 */
-		if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
-		    (cpus_updated || mems_updated)) {
+		if (!cpuset_v2() || (cpus_updated || mems_updated)) {
 			ret = security_task_setscheduler(task);
 			if (ret)
 				goto out_unlock;
@@ -3040,8 +3042,7 @@ static void cpuset_attach(struct cgroup_taskset *tset)
 	 * in effective cpus and mems. In that case, we can optimize out
 	 * by skipping the task iteration and update.
 	 */
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-	    !cpus_updated && !mems_updated) {
+	if (cpuset_v2() && !cpus_updated && !mems_updated) {
 		cpuset_attach_nodemask_to = cs->effective_mems;
 		goto out;
 	}
@@ -3391,7 +3392,7 @@ cpuset_css_alloc(struct cgroup_subsys_state *parent_css)
 	INIT_LIST_HEAD(&cs->remote_sibling);
 
 	/* Set CS_MEMORY_MIGRATE for default hierarchy */
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys))
+	if (cpuset_v2())
 		__set_bit(CS_MEMORY_MIGRATE, &cs->flags);
 
 	return &cs->css;
@@ -3418,8 +3419,7 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 	/*
 	 * For v2, clear CS_SCHED_LOAD_BALANCE if parent is isolated
 	 */
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-	    !is_sched_load_balance(parent))
+	if (cpuset_v2() && !is_sched_load_balance(parent))
 		clear_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
 
 	cpuset_inc();
@@ -3489,8 +3489,7 @@ static void cpuset_css_offline(struct cgroup_subsys_state *css)
 	if (is_partition_valid(cs))
 		update_prstate(cs, 0);
 
-	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-	    is_sched_load_balance(cs))
+	if (!cpuset_v2() && is_sched_load_balance(cs))
 		cpuset_update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
 
 	cpuset_dec();
-- 
2.47.0

Re: [PATCH 2/3] cgroup/cpuset: Enforce at most one rebuild_sched_domains_locked() call per operation

[Juri Lelli]

Hi,

On 09/11/24 21:50, Waiman Long wrote:
> Since commit ff0ce721ec21 ("cgroup/cpuset: Eliminate unncessary
> sched domains rebuilds in hotplug"), there is only one
> rebuild_sched_domains_locked() call per hotplug operation. However,
> writing to the various cpuset control files may still casue more than
> one rebuild_sched_domains_locked() call to happen in some cases.
> 
> Juri had found that two rebuild_sched_domains_locked() calls in
> update_prstate(), one from update_cpumasks_hier() and another one from
> update_partition_sd_lb() could cause cpuset partition to be created
> with null total_bw for DL tasks. IOW, DL tasks may not be scheduled
> correctly in such a partition.
> 
> A sample command sequence that can reproduce null total_bw is as
> follows.
> 
>   # echo Y >/sys/kernel/debug/sched/verbose
>   # echo +cpuset >/sys/fs/cgroup/cgroup.subtree_control
>   # mkdir /sys/fs/cgroup/test
>   # echo 0-7 > /sys/fs/cgroup/test/cpuset.cpus
>   # echo 6-7 > /sys/fs/cgroup/test/cpuset.cpus.exclusive
>   # echo root >/sys/fs/cgroup/test/cpuset.cpus.partition
> 
> Fix this double rebuild_sched_domains_locked() calls problem
> by replacing existing calls with cpuset_force_rebuild() except
> the rebuild_sched_domains_cpuslocked() call at the end of
> cpuset_handle_hotplug(). Checking of the force_sd_rebuild flag is
> now done at the end of cpuset_write_resmask() and update_prstate()
> to determine if rebuild_sched_domains_locked() should be called or not.
> 
> The cpuset v1 code can still call rebuild_sched_domains_locked()
> directly as double rebuild_sched_domains_locked() calls is not possible.
> 
> Reported-by: Juri Lelli <juri.lelli@redhat.com>
> Closes: https://lore.kernel.org/lkml/ZyuUcJDPBln1BK1Y@jlelli-thinkpadt14gen4.remote.csb/
> Signed-off-by: Waiman Long <longman@redhat.com>

This indeed works for me and fixes things with the test above (on v2).

Tested-by: Juri Lelli <juri.lelli@redhat.com>

Thanks!
Juri

Re: [PATCH 0/3] cgroup/cpuset: Remove redundant rebuild_sched_domains_locked() calls

[Tejun Heo]

On Sat, Nov 09, 2024 at 09:50:20PM -0500, Waiman Long wrote:
> Waiman Long (3):
>   cgroup/cpuset: Revert "Allow suppression of sched domain rebuild in
>     update_cpumasks_hier()"
>   cgroup/cpuset: Enforce at most one rebuild_sched_domains_locked() call
>     per operation
>   cgroup/cpuset: Further optimize code if CONFIG_CPUSETS_V1 not set

Applied to cgroup/for-6.13.

Thanks.

-- 
tejun