[PATCH/for-next 1/2] cgroup/cpuset: Defer housekeeping_update() call from CPU hotplug to task_work

[Waiman Long <longman@redhat.com>]

The update_isolation_cpumasks() function can be called either directly
from regular cpuset control file write with cpuset_full_lock() called
or via the CPU hotplug path with cpus_write_lock and cpuset_mutex held.

As we are going to enable dynamic update to the nozh_full housekeeping
cpumask (HK_TYPE_KERNEL_NOISE) soon with the help of CPU hotplug,
allowing the CPU hotplug path to call into housekeeping_update()
directly from update_isolation_cpumasks() will cause deadlock. So we
have to defer any call to housekeeping_update() after the CPU hotplug
operation has finished. This can be done via the task_work_add(...,
TWA_RESUME) API where the actual housekeeping_update() call, if needed,
will happen right before existing back to userspace.

Since the HK_TYPE_DOMAIN housekeeping cpumask should now track the
changes in "cpuset.cpus.isolated", add a check in test_cpuset_prs.sh to
confirm that the CPU hotplug deferral, if needed, is working as expected.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c                        | 49 ++++++++++++++++++-
 .../selftests/cgroup/test_cpuset_prs.sh       |  9 ++++
 2 files changed, 56 insertions(+), 2 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 7b7d12ab1006..98c7cb732206 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -84,6 +84,10 @@ static cpumask_var_t	isolated_cpus;
  */
 static bool isolated_cpus_updating;
 
+/* Both cpuset_mutex and cpus_read_locked acquired */
+static bool cpuset_full_locked;
+static bool isolation_task_work_queued;
+
 /*
  * A flag to force sched domain rebuild at the end of an operation.
  * It can be set in
@@ -285,10 +289,12 @@ void cpuset_full_lock(void)
 {
 	cpus_read_lock();
 	mutex_lock(&cpuset_mutex);
+	cpuset_full_locked = true;
 }
 
 void cpuset_full_unlock(void)
 {
+	cpuset_full_locked = false;
 	mutex_unlock(&cpuset_mutex);
 	cpus_read_unlock();
 }
@@ -1285,25 +1291,64 @@ static bool prstate_housekeeping_conflict(int prstate, struct cpumask *new_cpus)
 	return false;
 }
 
+static void __update_isolation_cpumasks(bool twork);
+static void isolation_task_work_fn(struct callback_head *cb)
+{
+	cpuset_full_lock();
+	__update_isolation_cpumasks(true);
+	cpuset_full_lock();
+}
+
 /*
- * update_isolation_cpumasks - Update external isolation related CPU masks
+ * __update_isolation_cpumasks - Update external isolation related CPU masks
+ * @twork - set if call from isolation_task_work_fn()
  *
  * The following external CPU masks will be updated if necessary:
  * - workqueue unbound cpumask
  */
-static void update_isolation_cpumasks(void)
+static void __update_isolation_cpumasks(bool twork)
 {
 	int ret;
 
+	if (twork)
+		isolation_task_work_queued = false;
+
 	if (!isolated_cpus_updating)
 		return;
 
+	/*
+	 * This function can be reached either directly from regular cpuset
+	 * control file write (cpuset_full_locked) or via hotplug
+	 * (cpus_write_lock && cpuset_mutex held). In the later case, we
+	 * defer the housekeeping_update() call to a task_work to avoid
+	 * the possibility of deadlock. The task_work will be run right
+	 * before exiting back to userspace.
+	 */
+	if (!cpuset_full_locked) {
+		static struct callback_head twork_cb;
+
+		if (!isolation_task_work_queued) {
+			init_task_work(&twork_cb, isolation_task_work_fn);
+			if (!task_work_add(current, &twork_cb, TWA_RESUME))
+				isolation_task_work_queued = true;
+			else
+				/* Current task shouldn't be exiting */
+				WARN_ON_ONCE(1);
+		}
+		return;
+	}
+
 	ret = housekeeping_update(isolated_cpus);
 	WARN_ON_ONCE(ret < 0);
 
 	isolated_cpus_updating = false;
 }
 
+static inline void update_isolation_cpumasks(void)
+{
+	__update_isolation_cpumasks(false);
+}
+
 /**
  * rm_siblings_excl_cpus - Remove exclusive CPUs that are used by sibling cpusets
  * @parent: Parent cpuset containing all siblings
diff --git a/tools/testing/selftests/cgroup/test_cpuset_prs.sh b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
index 5dff3ad53867..af4a2532cb3e 100755
--- a/tools/testing/selftests/cgroup/test_cpuset_prs.sh
+++ b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
@@ -773,6 +773,7 @@ check_isolcpus()
 	EXPECTED_ISOLCPUS=$1
 	ISCPUS=${CGROUP2}/cpuset.cpus.isolated
 	ISOLCPUS=$(cat $ISCPUS)
+	HKICPUS=$(cat /sys/devices/system/cpu/isolated)
 	LASTISOLCPU=
 	SCHED_DOMAINS=/sys/kernel/debug/sched/domains
 	if [[ $EXPECTED_ISOLCPUS = . ]]
@@ -810,6 +811,14 @@ check_isolcpus()
 	ISOLCPUS=
 	EXPECTED_ISOLCPUS=$EXPECTED_SDOMAIN
 
+	#
+	# The inverse of HK_TYPE_DOMAIN cpumask in $HKICPUS should match $ISOLCPUS
+	#
+	[[ "$ISOLCPUS" != "$HKICPUS" ]] && {
+		echo "Housekeeping isolated CPUs mismatch - $HKICPUS"
+		return 1
+	}
+
 	#
 	# Use the sched domain in debugfs to check isolated CPUs, if available
 	#
-- 
2.52.0