Introduce infrastructure function housekeeping_update() to change
housekeeping_cpumask during runtime and adjust affinities of depended
subsystems.

Affinity adjustments of subsystems follow in subsequent patches.

Parent patch:
"sched/isolation: Exclude dynamically isolated CPUs from housekeeping masks"
https://lore.kernel.org/lkml/20240229021414.508972-2-longman@redhat.com/

Test example for cgroup2:

cd /sys/fs/cgroup/
echo +cpuset > cgroup.subtree_control
mkdir test
echo isolated > test/cpuset.cpus.partition
echo $isolate > test/cpuset.cpus

Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
---
 kernel/sched/isolation.c | 48 +++++++++++++++++++++++++++++++++++-----
 1 file changed, 43 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
index 948b9ee0dc2cc..036e48f0e7d1b 100644
--- a/kernel/sched/isolation.c
+++ b/kernel/sched/isolation.c
@@ -116,6 +116,39 @@ static void __init housekeeping_setup_type(enum hk_type type,
 		     housekeeping_staging);
 }
 
+/*
+ * housekeeping_update - change housekeeping.cpumasks[type] and propagate the
+ * change.
+ *
+ * Assuming cpuset_mutex is held in sched_partition_write or
+ * cpuset_write_resmask.
+ */
+static int housekeeping_update(enum hk_type type, cpumask_var_t update)
+{
+	struct {
+		struct cpumask changed;
+		struct cpumask enable;
+		struct cpumask disable;
+	} *masks;
+
+	masks = kmalloc(sizeof(*masks), GFP_KERNEL);
+	if (!masks)
+		return -ENOMEM;
+
+	lockdep_assert_cpus_held();
+	cpumask_xor(&masks->changed, housekeeping_cpumask(type), update);
+	cpumask_and(&masks->enable, &masks->changed, update);
+	cpumask_andnot(&masks->disable, &masks->changed, update);
+	cpumask_copy(housekeeping.cpumasks[type], update);
+	housekeeping.flags |= BIT(type);
+	if (!static_branch_unlikely(&housekeeping_overridden))
+		static_key_enable_cpuslocked(&housekeeping_overridden.key);
+
+	kfree(masks);
+
+	return 0;
+}
+
 static int __init housekeeping_setup(char *str, unsigned long flags)
 {
 	cpumask_var_t non_housekeeping_mask, housekeeping_staging;
@@ -314,9 +347,12 @@ int housekeeping_exlude_isolcpus(const struct cpumask *isolcpus, unsigned long f
 		/*
 		 * Reset housekeeping to bootup default
 		 */
-		for_each_set_bit(type, &housekeeping_boot.flags, HK_TYPE_MAX)
-			cpumask_copy(housekeeping.cpumasks[type],
-				     housekeeping_boot.cpumasks[type]);
+		for_each_set_bit(type, &housekeeping_boot.flags, HK_TYPE_MAX) {
+			int err = housekeeping_update(type, housekeeping_boot.cpumasks[type]);
+
+			if (err)
+				return err;
+		}
 
 		WRITE_ONCE(housekeeping.flags, housekeeping_boot.flags);
 		if (!housekeeping_boot.flags &&
@@ -344,9 +380,11 @@ int housekeeping_exlude_isolcpus(const struct cpumask *isolcpus, unsigned long f
 		cpumask_andnot(tmp_mask, src_mask, isolcpus);
 		if (!cpumask_intersects(tmp_mask, cpu_online_mask))
 			return -EINVAL;	/* Invalid isolated CPUs */
-		cpumask_copy(housekeeping.cpumasks[type], tmp_mask);
+		int err = housekeeping_update(type, tmp_mask);
+
+		if (err)
+			return err;
 	}
-	WRITE_ONCE(housekeeping.flags, housekeeping_boot.flags | flags);
 	excluded = true;
 	if (!static_branch_unlikely(&housekeeping_overridden))
 		static_key_enable_cpuslocked(&housekeeping_overridden.key);
-- 
2.45.0

On Thu, May 16 2024 at 22:04, Costa Shulyupin wrote:
> Introduce infrastructure function housekeeping_update() to change
> housekeeping_cpumask during runtime and adjust affinities of depended
> subsystems.
>
> Affinity adjustments of subsystems follow in subsequent patches.
>
> Parent patch:
> "sched/isolation: Exclude dynamically isolated CPUs from housekeeping masks"
> https://lore.kernel.org/lkml/20240229021414.508972-2-longman@redhat.com/
>
> Test example for cgroup2:
>
> cd /sys/fs/cgroup/
> echo +cpuset > cgroup.subtree_control
> mkdir test
> echo isolated > test/cpuset.cpus.partition
> echo $isolate > test/cpuset.cpus

This changelog is not telling me anything. Please see
Documentation/process/ what changelogs should contain.

> +/*
> + * housekeeping_update - change housekeeping.cpumasks[type] and propagate the
> + * change.
> + *
> + * Assuming cpuset_mutex is held in sched_partition_write or
> + * cpuset_write_resmask.

Locking cannot be assumed. lockdep_assert_held() is there to document
and enforce such requirements.

> + */
> +static int housekeeping_update(enum hk_type type, cpumask_var_t update)

Please us 'struct cpumask *update' as it makes it clear what this is
about. cpumask_var_t is a hack to make onstack and embedded cpumask and
their allocated counterparts possible without #ifdeffery in the code.

But any function which is not related to alloc/free of cpumask_var_t
should simply use 'struct cpumask *' as argument type.

> +	housekeeping.flags |= BIT(type);

The existing code uses WRITE_ONCE() probably for a reason. Why is that
not longer required here?

>  static int __init housekeeping_setup(char *str, unsigned long flags)
>  {
>  	cpumask_var_t non_housekeeping_mask, housekeeping_staging;
> @@ -314,9 +347,12 @@ int housekeeping_exlude_isolcpus(const struct cpumask *isolcpus, unsigned long f
>  		/*
>  		 * Reset housekeeping to bootup default
>  		 */
> -		for_each_set_bit(type, &housekeeping_boot.flags, HK_TYPE_MAX)
> -			cpumask_copy(housekeeping.cpumasks[type],
> -				     housekeeping_boot.cpumasks[type]);
> +		for_each_set_bit(type, &housekeeping_boot.flags, HK_TYPE_MAX) {
> +			int err = housekeeping_update(type, housekeeping_boot.cpumasks[type]);
> +
> +			if (err)
> +				return err;
> +		}
>  
>  		WRITE_ONCE(housekeeping.flags, housekeeping_boot.flags);
>  		if (!housekeeping_boot.flags &&
> @@ -344,9 +380,11 @@ int housekeeping_exlude_isolcpus(const struct cpumask *isolcpus, unsigned long f
>  		cpumask_andnot(tmp_mask, src_mask, isolcpus);
>  		if (!cpumask_intersects(tmp_mask, cpu_online_mask))
>  			return -EINVAL;	/* Invalid isolated CPUs */
> -		cpumask_copy(housekeeping.cpumasks[type], tmp_mask);
> +		int err = housekeeping_update(type, tmp_mask);
> +
> +		if (err)
> +			return err;

Do we really need two places to define 'int err' or might it be possible
to have one instance defined at function scope?

Thanks,

        tglx