1 files changed, 70 insertions, 5 deletions

diff --git a/kernel/padata.c b/kernel/padata.c
index 868f947166d7..57c0074d50cc 100644
--- a/kernel/padata.c
+++ b/kernel/padata.c
@@ -131,6 +131,7 @@ int padata_do_parallel(struct padata_instance *pinst,
 	padata->cb_cpu = cb_cpu;
 
 	target_cpu = padata_cpu_hash(pd);
+	padata->cpu = target_cpu;
 	queue = per_cpu_ptr(pd->pqueue, target_cpu);
 
 	spin_lock(&queue->parallel.lock);
@@ -275,11 +276,51 @@ static void padata_reorder(struct parallel_data *pd)
 	return;
 }
 
-static void padata_reorder_timer(unsigned long arg)
+static void invoke_padata_reorder(struct work_struct *work)
 {
-	struct parallel_data *pd = (struct parallel_data *)arg;
+	struct padata_parallel_queue *pqueue;
+	struct parallel_data *pd;
 
+	local_bh_disable();
+	pqueue = container_of(work, struct padata_parallel_queue, reorder_work);
+	pd = pqueue->pd;
 	padata_reorder(pd);
+	local_bh_enable();
+}
+
+static void padata_reorder_timer(struct timer_list *t)
+{
+	struct parallel_data *pd = from_timer(pd, t, timer);
+	unsigned int weight;
+	int target_cpu, cpu;
+
+	cpu = get_cpu();
+
+	/* We don't lock pd here to not interfere with parallel processing
+	 * padata_reorder() calls on other CPUs. We just need any CPU out of
+	 * the cpumask.pcpu set. It would be nice if it's the right one but
+	 * it doesn't matter if we're off to the next one by using an outdated
+	 * pd->processed value.
+	 */
+	weight = cpumask_weight(pd->cpumask.pcpu);
+	target_cpu = padata_index_to_cpu(pd, pd->processed % weight);
+
+	/* ensure to call the reorder callback on the correct CPU */
+	if (cpu != target_cpu) {
+		struct padata_parallel_queue *pqueue;
+		struct padata_instance *pinst;
+
+		/* The timer function is serialized wrt itself -- no locking
+		 * needed.
+		 */
+		pinst = pd->pinst;
+		pqueue = per_cpu_ptr(pd->pqueue, target_cpu);
+		queue_work_on(target_cpu, pinst->wq, &pqueue->reorder_work);
+	} else {
+		padata_reorder(pd);
+	}
+
+	put_cpu();
 }
 
 static void padata_serial_worker(struct work_struct *serial_work)
@@ -323,10 +364,21 @@ void padata_do_serial(struct padata_priv *padata)
 	int cpu;
 	struct padata_parallel_queue *pqueue;
 	struct parallel_data *pd;
+	int reorder_via_wq = 0;
 
 	pd = padata->pd;
 
 	cpu = get_cpu();
+
+	/* We need to run on the same CPU padata_do_parallel(.., padata, ..)
+	 * was called on -- or, at least, enqueue the padata object into the
+	 * correct per-cpu queue.
+	 */
+	if (cpu != padata->cpu) {
+		reorder_via_wq = 1;
+		cpu = padata->cpu;
+	}
+
 	pqueue = per_cpu_ptr(pd->pqueue, cpu);
 
 	spin_lock(&pqueue->reorder.lock);
@@ -336,7 +388,13 @@ void padata_do_serial(struct padata_priv *padata)
 
 	put_cpu();
 
-	padata_reorder(pd);
+	/* If we're running on the wrong CPU, call padata_reorder() via a
+	 * kernel worker.
+	 */
+	if (reorder_via_wq)
+		queue_work_on(cpu, pd->pinst->wq, &pqueue->reorder_work);
+	else
+		padata_reorder(pd);
 }
 EXPORT_SYMBOL(padata_do_serial);
 
@@ -384,8 +442,14 @@ static void padata_init_pqueues(struct parallel_data *pd)
 	struct padata_parallel_queue *pqueue;
 
 	cpu_index = 0;
-	for_each_cpu(cpu, pd->cpumask.pcpu) {
+	for_each_possible_cpu(cpu) {
 		pqueue = per_cpu_ptr(pd->pqueue, cpu);
+
+		if (!cpumask_test_cpu(cpu, pd->cpumask.pcpu)) {
+			pqueue->cpu_index = -1;
+			continue;
+		}
+
 		pqueue->pd = pd;
 		pqueue->cpu_index = cpu_index;
 		cpu_index++;
@@ -393,6 +457,7 @@ static void padata_init_pqueues(struct parallel_data *pd)
 		__padata_list_init(&pqueue->reorder);
 		__padata_list_init(&pqueue->parallel);
 		INIT_WORK(&pqueue->work, padata_parallel_worker);
+		INIT_WORK(&pqueue->reorder_work, invoke_padata_reorder);
 		atomic_set(&pqueue->num_obj, 0);
 	}
 }
@@ -420,7 +485,7 @@ static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
 
 	padata_init_pqueues(pd);
 	padata_init_squeues(pd);
-	setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd);
+	timer_setup(&pd->timer, padata_reorder_timer, 0);
 	atomic_set(&pd->seq_nr, -1);
 	atomic_set(&pd->reorder_objects, 0);
 	atomic_set(&pd->refcnt, 0);