drm/i915: setup the hangcheck timer early

... together with all the other irq related resources in
intel_irq_init. I've managed to oops in the notify_ring function on my
ilk, presumably because of the powerctx setup call to i915_gpu_idle.

Note that this is only a problem with the reorder irq setup sequence
for irq-driver gmbus/dp aux.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drivers/gpu/drm/i915/i915_dma.c

@@ -1611,9 +1611,6 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
 	intel_opregion_init(dev);
 	acpi_video_register();
 
-	setup_timer(&dev_priv->hangcheck_timer, i915_hangcheck_elapsed,
-		    (unsigned long) dev);
-
 	if (IS_GEN5(dev))
 		intel_gpu_ips_init(dev_priv);
 

drivers/gpu/drm/i915/i915_irq.c

@@ -2687,6 +2687,9 @@ void intel_irq_init(struct drm_device *dev)
 	INIT_WORK(&dev_priv->rps.work, gen6_pm_rps_work);
 	INIT_WORK(&dev_priv->l3_parity.error_work, ivybridge_parity_work);
 
+	setup_timer(&dev_priv->hangcheck_timer, i915_hangcheck_elapsed,
+		    (unsigned long) dev);
+
 	dev->driver->get_vblank_counter = i915_get_vblank_counter;
 	dev->max_vblank_count = 0xffffff; /* only 24 bits of frame count */
 	if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {