[RFC PATCH 1/2] workqueue: allow use of realtime scheduling policies

[Rasmus Villemoes]

Prepare for allowing the administrator to set RT scheduling policies,
rather than just tweaking the nice value, of workqueues exposed via
sysfs. Subsystems that currently use, say, system_unbound_wq, can be
updated to create a separate workqueue (possibly depending on a
CONFIG_ knob or boottime parameter).

This patch merely updates the internal interfaces. The next patch will
expose a sysfs knob.

Signed-off-by: Rasmus Villemoes <linux@xxxxxxxxxxxxxxxxxx>
---
 include/linux/workqueue.h | 17 +++++++++++++++--
 kernel/workqueue.c        | 20 ++++++++++++++++++--
 2 files changed, 33 insertions(+), 4 deletions(-)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index 7fee9b6cfede..a69bdd877120 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -131,9 +131,22 @@ struct rcu_work {
  */
 struct workqueue_attrs {
 	/**
-	 * @nice: nice level
+	 * @policy: scheduling policy (SCHED_NORMAL, SCHED_FIFO, SCHED_RR)
 	 */
-	int nice;
+	int policy;
+
+	/**
+	 * @nice: nice level (SCHED_NORMAL)
+	 * @priority: priority (SCHED_FIFO, SCHED_RR)
+	 *
+	 * Letting these two fields occupy the same word simplifies
+	 * copying, hashing and equality testing of struct
+	 * workqueue_attrs.
+	 */
+	union {
+		int nice;
+		int priority;
+	};
 
 	/**
 	 * @cpumask: allowed CPUs
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 33f1106b4f99..9eb2ff7bcc04 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -51,6 +51,7 @@
 #include <linux/sched/isolation.h>
 #include <linux/nmi.h>
 #include <linux/kvm_para.h>
+#include <uapi/linux/sched/types.h>
 
 #include "workqueue_internal.h"
 
@@ -1969,7 +1970,13 @@ static struct worker *create_worker(struct worker_pool *pool)
 	if (IS_ERR(worker->task))
 		goto fail;
 
-	set_user_nice(worker->task, pool->attrs->nice);
+	if (pool->attrs->policy == SCHED_NORMAL) {
+		set_user_nice(worker->task, pool->attrs->nice);
+	} else {
+		struct sched_param sp = { .sched_priority = pool->attrs->priority };
+
+		sched_setscheduler(worker->task, pool->attrs->policy, &sp);
+	}
 	kthread_bind_mask(worker->task, pool->attrs->cpumask);
 
 	/* successful, attach the worker to the pool */
@@ -3402,6 +3409,12 @@ struct workqueue_attrs *alloc_workqueue_attrs(void)
 {
 	struct workqueue_attrs *attrs;
 
+	/*
+	 * A zeroed structure has ->policy==SCHED_NORMAL and
+	 * ->nice==0.
+	 */
+	static_assert(SCHED_NORMAL == 0);
+
 	attrs = kzalloc(sizeof(*attrs), GFP_KERNEL);
 	if (!attrs)
 		goto fail;
@@ -3418,6 +3431,7 @@ struct workqueue_attrs *alloc_workqueue_attrs(void)
 static void copy_workqueue_attrs(struct workqueue_attrs *to,
 				 const struct workqueue_attrs *from)
 {
+	to->policy = from->policy;
 	to->nice = from->nice;
 	cpumask_copy(to->cpumask, from->cpumask);
 	/*
@@ -3433,7 +3447,7 @@ static u32 wqattrs_hash(const struct workqueue_attrs *attrs)
 {
 	u32 hash = 0;
 
-	hash = jhash_1word(attrs->nice, hash);
+	hash = jhash_2words(attrs->policy, attrs->nice, hash);
 	hash = jhash(cpumask_bits(attrs->cpumask),
 		     BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash);
 	return hash;
@@ -3443,6 +3457,8 @@ static u32 wqattrs_hash(const struct workqueue_attrs *attrs)
 static bool wqattrs_equal(const struct workqueue_attrs *a,
 			  const struct workqueue_attrs *b)
 {
+	if (a->policy != b->policy)
+		return false;
 	if (a->nice != b->nice)
 		return false;
 	if (!cpumask_equal(a->cpumask, b->cpumask))
-- 
2.31.1