3.8.13-rt11

index d44dea3..f5ad8e1 100644
@@ -83,6 +83,7 @@ struct hotplug_pcp {
 	int refcount;
 	int grab_lock;
 	struct completion synced;
+	struct completion unplug_wait;
 #ifdef CONFIG_PREEMPT_RT_FULL
 	spinlock_t lock;
 #else

@@ -180,6 +181,7 @@ static int sync_unplug_thread(void *data)
 {
 	struct hotplug_pcp *hp = data;
 
+	wait_for_completion(&hp->unplug_wait);
 	preempt_disable();
 	hp->unplug = current;
 	wait_for_pinned_cpus(hp);

@@ -245,6 +247,14 @@ static void __cpu_unplug_sync(struct hotplug_pcp *hp)
 	wait_for_completion(&hp->synced);
 }
 
+static void __cpu_unplug_wait(unsigned int cpu)
+{
+	struct hotplug_pcp *hp = &per_cpu(hotplug_pcp, cpu);
+
+	complete(&hp->unplug_wait);
+	wait_for_completion(&hp->synced);
+}
+
 /*
  * Start the sync_unplug_thread on the target cpu and wait for it to
  * complete.

@@ -268,6 +278,7 @@ static int cpu_unplug_begin(unsigned int cpu)
 	tell_sched_cpu_down_begin(cpu);
 
 	init_completion(&hp->synced);
+	init_completion(&hp->unplug_wait);
 
 	hp->sync_tsk = kthread_create(sync_unplug_thread, hp, "sync_unplug/%d", cpu);
 	if (IS_ERR(hp->sync_tsk)) {

@@ -283,8 +294,7 @@ static int cpu_unplug_begin(unsigned int cpu)
 	 * wait for tasks that are going to enter these sections and
 	 * we must not have them block.
 	 */
-	__cpu_unplug_sync(hp);
-
+	wake_up_process(hp->sync_tsk);
 	return 0;
 }
 

@@ -535,6 +545,7 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 		.hcpu = hcpu,
 	};
 	cpumask_var_t cpumask;
+	cpumask_var_t cpumask_org;
 
 	if (num_online_cpus() == 1)
 		return -EBUSY;

@@ -545,6 +556,12 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 	/* Move the downtaker off the unplug cpu */
 	if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
 		return -ENOMEM;
+	if (!alloc_cpumask_var(&cpumask_org, GFP_KERNEL))  {
+		free_cpumask_var(cpumask);
+		return -ENOMEM;
+	}
+
+	cpumask_copy(cpumask_org, tsk_cpus_allowed(current));
 	cpumask_andnot(cpumask, cpu_online_mask, cpumask_of(cpu));
 	set_cpus_allowed_ptr(current, cpumask);
 	free_cpumask_var(cpumask);

@@ -553,7 +570,8 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 	if (mycpu == cpu) {
 		printk(KERN_ERR "Yuck! Still on unplug CPU\n!");
 		migrate_enable();
-		return -EBUSY;
+		err = -EBUSY;
+		goto restore_cpus;
 	}
 
 	cpu_hotplug_begin();

@@ -571,6 +589,8 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 				__func__, cpu);
 		goto out_release;
 	}
+
+	__cpu_unplug_wait(cpu);
 	smpboot_park_threads(cpu);
 
 	/* Notifiers are done. Don't let any more tasks pin this CPU. */

@@ -610,6 +630,9 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 	cpu_hotplug_done();
 	if (!err)
 		cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
+restore_cpus:
+	set_cpus_allowed_ptr(current, cpumask_org);
+	free_cpumask_var(cpumask_org);
 	return err;
 }
 

index 1fba5cb..81a28dd 100644
@@ -849,9 +849,6 @@ static void irq_thread_dtor(struct callback_head *unused)
 static int irq_thread(void *data)
 {
 	struct callback_head on_exit_work;
-	static const struct sched_param param = {
-		.sched_priority = MAX_USER_RT_PRIO/2,
-	};
 	struct irqaction *action = data;
 	struct irq_desc *desc = irq_to_desc(action->irq);
 	irqreturn_t (*handler_fn)(struct irq_desc *desc,

@@ -863,8 +860,6 @@ static int irq_thread(void *data)
 	else
 		handler_fn = irq_thread_fn;
 
-	sched_setscheduler(current, SCHED_FIFO, &param);
-
 	init_task_work(&on_exit_work, irq_thread_dtor);
 	task_work_add(current, &on_exit_work, false);
 

@@ -965,6 +960,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
 	 */
 	if (new->thread_fn && !nested) {
 		struct task_struct *t;
+		static const struct sched_param param = {
+			.sched_priority = MAX_USER_RT_PRIO/2,
+		};
 
 		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
 				   new->name);

@@ -972,6 +970,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
 			ret = PTR_ERR(t);
 			goto out_mput;
 		}
+
+		sched_setscheduler(t, SCHED_FIFO, &param);
+
 		/*
 		 * We keep the reference to the task struct even if
 		 * the thread dies to avoid that the interrupt code

index 374e7b1..2f1c8d3 100644
@@ -76,7 +76,9 @@ struct tvec_root {
 struct tvec_base {
 	spinlock_t lock;
 	struct timer_list *running_timer;
+#ifdef CONFIG_PREEMPT_RT_FULL
 	wait_queue_head_t wait_for_running_timer;
+#endif
 	unsigned long timer_jiffies;
 	unsigned long next_timer;
 	unsigned long active_timers;

@@ -982,7 +984,7 @@ static void wait_for_running_timer(struct timer_list *timer)
 			   base->running_timer != timer);
 }
 
-# define wakeup_timer_waiters(b)	wake_up(&(b)->wait_for_tunning_timer)
+# define wakeup_timer_waiters(b)	wake_up(&(b)->wait_for_running_timer)
 #else
 static inline void wait_for_running_timer(struct timer_list *timer)
 {

@@ -1236,7 +1238,7 @@ static inline void __run_timers(struct tvec_base *base)
 			}
 		}
 	}
-	wake_up(&base->wait_for_running_timer);
+	wakeup_timer_waiters(base);
 	spin_unlock_irq(&base->lock);
 }
 

@@ -1755,7 +1757,9 @@ static int __cpuinit init_timers_cpu(int cpu)
 	}
 
 	spin_lock_init(&base->lock);
+#ifdef CONFIG_PREEMPT_RT_FULL
 	init_waitqueue_head(&base->wait_for_running_timer);
+#endif
 
 	for (j = 0; j < TVN_SIZE; j++) {
 		INIT_LIST_HEAD(base->tv5.vec + j);

index 11285e4..0d49ddf 100644
@@ -1637,8 +1637,11 @@ __acquires(&gcwq->lock)
 		 * it races with cpu hotunplug operation.  Verify
 		 * against GCWQ_DISASSOCIATED.
 		 */
-		if (!(gcwq->flags & GCWQ_DISASSOCIATED))
+		if (!(gcwq->flags & GCWQ_DISASSOCIATED)) {
 			set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu));
+			if (WARN_ON(!(task->flags & PF_THREAD_BOUND)))
+				task->flags |= PF_THREAD_BOUND;
+		}
 
 		spin_lock_irq(&gcwq->lock);
 		if (gcwq->flags & GCWQ_DISASSOCIATED)

index 339540d..dce41f4 100644
@@ -35,6 +35,7 @@
 #include <linux/swap.h>
 #include <linux/pci.h>
 #include <linux/dma-buf.h>
+#include <linux/module.h>
 
 static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj);
 static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj);

@@ -2656,6 +2657,10 @@ static inline int fence_number(struct drm_i915_private *dev_priv,
 	return fence - dev_priv->fence_regs;
 }
 
+static bool do_wbinvd = true;
+module_param(do_wbinvd, bool, 0644);
+MODULE_PARM_DESC(do_wbinvd, "Do expensive synchronization. Say no after you pin each GPU process to the same CPU in order to lower the latency.");
+
 static void i915_gem_write_fence__ipi(void *data)
 {
 	wbinvd();

@@ -2679,8 +2684,16 @@ static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj,
 	 * on each processor in order to manually flush all memory
 	 * transactions before updating the fence register.
 	 */
-	if (HAS_LLC(obj->base.dev))
-		on_each_cpu(i915_gem_write_fence__ipi, NULL, 1);
+	if (HAS_LLC(obj->base.dev)) {
+		if (do_wbinvd) {
+#ifdef CONFIG_PREEMPT_RT_FULL
+			pr_err_once("WARNING! The i915 invalidates all caches which increases the latency.");
+			pr_err_once("As a workaround use 'i915.do_wbinvd=no' and PIN each process doing ");
+			pr_err_once("any kind of GPU activity to the same CPU to avoid problems.");
+#endif
+			on_each_cpu(i915_gem_write_fence__ipi, NULL, 1);
+		}
+	}
 	i915_gem_write_fence(dev, fence_reg, enable ? obj : NULL);
 
 	if (enable) {