[RFC PATCH 2/2] workqueue: Add WQ_PREFER_PERCPU and system_prefer_percpu_wq

[Marco Crivellari]

There may be situations where local execution is not strictly needed for
correctness, but it is preferred for performance gains.

The Workqueue API currently do not distinguish between these two instances.

So add WQ_PREFER_PERCPU and system_prefer_percpu_wq, so that it would be the
first choice for workload who don't strictly need local execution for
correctness but only to maintain locality.

Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@xxxxxxxxxxxxx/
Suggested-by: Tejun Heo <tj@xxxxxxxxxx>
Signed-off-by: Marco Crivellari <marco.crivellari@xxxxxxxx>
---
 include/linux/workqueue.h | 7 +++++++
 kernel/workqueue.c        | 6 +++++-
 2 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index f46379d937c9..be65df3dea5b 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -404,6 +404,7 @@ enum wq_flags {
 	 */
 	WQ_POWER_EFFICIENT	= 1 << 7,
 	WQ_PERCPU		= 1 << 8, /* bound to a specific cpu */
+	WQ_PREFER_PERCPU	= 1 << 9, /* prefer local cpu, but it doesn't require it */
 
 	__WQ_DESTROYING		= 1 << 15, /* internal: workqueue is destroying */
 	__WQ_DRAINING		= 1 << 16, /* internal: workqueue is draining */
@@ -460,6 +461,9 @@ enum wq_consts {
  *
  * system_bh[_highpri]_wq are convenience interface to softirq. BH work items
  * are executed in the queueing CPU's BH context in the queueing order.
+ *
+ * system_prefer_percpu_wq is for work items which prefer to be local but
+ * doesn't require it
  */
 extern struct workqueue_struct *system_wq; /* use system_percpu_wq, this will be removed */
 extern struct workqueue_struct *system_percpu_wq;
@@ -473,6 +477,7 @@ extern struct workqueue_struct *system_freezable_power_efficient_wq;
 extern struct workqueue_struct *system_bh_wq;
 extern struct workqueue_struct *system_bh_highpri_wq;
 extern struct workqueue_struct *system_dfl_long_wq;
+extern struct workqueue_struct *system_prefer_percpu_wq;
 
 void workqueue_softirq_action(bool highpri);
 void workqueue_softirq_dead(unsigned int cpu);
@@ -873,6 +878,8 @@ extern void __warn_flushing_systemwide_wq(void)
 	     _wq == system_freezable_wq) ||				\
 	    (__builtin_constant_p(_wq == system_power_efficient_wq) &&	\
 	     _wq == system_power_efficient_wq) ||			\
+	    (__builtin_constant_p(_wq == system_prefer_percpu_wq) &&	\
+	     _wq == system_prefer_percpu_wq) ||				\
 	    (__builtin_constant_p(_wq == system_freezable_power_efficient_wq) && \
 	     _wq == system_freezable_power_efficient_wq))		\
 		__warn_flushing_systemwide_wq();			\
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 5f747f241a5f..abba222a2d58 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -545,6 +545,8 @@ struct workqueue_struct *system_bh_highpri_wq;
 EXPORT_SYMBOL_GPL(system_bh_highpri_wq);
 struct workqueue_struct *system_dfl_long_wq __ro_after_init;
 EXPORT_SYMBOL_GPL(system_dfl_long_wq);
+struct workqueue_struct *system_prefer_percpu_wq __ro_after_init;
+EXPORT_SYMBOL_GPL(system_prefer_percpu_wq);
 
 static int worker_thread(void *__worker);
 static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
@@ -8029,11 +8031,13 @@ void __init workqueue_init_early(void)
 	system_bh_highpri_wq = alloc_workqueue("events_bh_highpri",
 					       WQ_BH | WQ_HIGHPRI | WQ_PERCPU, 0);
 	system_dfl_long_wq = alloc_workqueue("events_dfl_long", WQ_UNBOUND, WQ_MAX_ACTIVE);
+	system_prefer_percpu_wq = alloc_workqueue("events_prefer_percpu", WQ_PREFER_PERCPU, 0);
 	BUG_ON(!system_wq || !system_percpu_wq|| !system_highpri_wq || !system_long_wq ||
 	       !system_unbound_wq || !system_freezable_wq || !system_dfl_wq ||
 	       !system_power_efficient_wq ||
 	       !system_freezable_power_efficient_wq ||
-	       !system_bh_wq || !system_bh_highpri_wq || !system_dfl_long_wq);
+	       !system_bh_wq || !system_bh_highpri_wq || !system_dfl_long_wq ||
+	       !system_prefer_percpu_wq);
 }
 
 static void __init wq_cpu_intensive_thresh_init(void)
-- 
2.53.0