[patch] workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs

[Mike Galbraith <umgwanakikbuti@gmail.com>]

On Tue, 2015-07-21 at 10:55 +0200, Mike Galbraith wrote:
> On Sun, 2015-07-19 at 10:02 +0200, Mike Galbraith wrote:
> 
> > Why do we do nothing about these allegedly unbound work items?
> 
> My box seems to think the answer is: no reason other than nobody having
> asked the source to please not do that.  Guess I'll go ask a NUMA box.

My [128] socket boxen show zero signs of caring, and it's dirt simple,
so it's no longer an experiment.  Fly or die little patchlet...


WORK_CPU_UNBOUND work items queued to a bound workqueue always run
locally.  This is a good thing normally, but not when the user has
asked us to keep unbound work away from certain CPUs.  Round robin
these to wq_unbound_cpumask CPUs instead, as perturbation avoidance
trumps performance.

Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
---
 kernel/workqueue.c |   27 +++++++++++++++++++++++++--
 1 file changed, 25 insertions(+), 2 deletions(-)

--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -301,6 +301,9 @@ static bool workqueue_freezing;		/* PL:
 
 static cpumask_var_t wq_unbound_cpumask; /* PL: low level cpumask for all unbound wqs */
 
+/* CPU where WORK_CPU_UNBOUND work was last round robin scheduled from this CPU */
+static DEFINE_PER_CPU(unsigned int, wq_unbound_rr_cpu_last);
+
 /* the per-cpu worker pools */
 static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
 				     cpu_worker_pools);
@@ -1294,6 +1297,24 @@ static bool is_chained_work(struct workq
 	return worker && worker->current_pwq->wq == wq;
 }
 
+/*
+ * When queueing WORK_CPU_UNBOUND work to a !WQ_UNBOUND queue, round
+ * robin among wq_unbound_cpumask to avoid perturbing sensitive tasks.
+ */
+static unsigned int select_round_robin_cpu(unsigned int cpu)
+{
+	if (cpumask_test_cpu(cpu, wq_unbound_cpumask))
+		return cpu;
+	if (cpumask_empty(wq_unbound_cpumask))
+		return cpu;
+	cpu = __this_cpu_read(wq_unbound_rr_cpu_last);
+	cpu = cpumask_next_and(cpu, wq_unbound_cpumask, cpu_online_mask);
+	if (cpu >= nr_cpu_ids)
+		cpu = 0;
+	__this_cpu_write(wq_unbound_rr_cpu_last, cpu);
+	return cpu;
+}
+
 static void __queue_work(int cpu, struct workqueue_struct *wq,
 			 struct work_struct *work)
 {
@@ -1322,9 +1343,11 @@ static void __queue_work(int cpu, struct
 		cpu = raw_smp_processor_id();
 
 	/* pwq which will be used unless @work is executing elsewhere */
-	if (!(wq->flags & WQ_UNBOUND))
+	if (!(wq->flags & WQ_UNBOUND)) {
+		if (req_cpu == WORK_CPU_UNBOUND)
+			cpu = select_round_robin_cpu(cpu);
 		pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
-	else
+	} else
 		pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
 
 	/*