drm/i915: enable irqs earlier when resuming
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 5 Mar 2013 08:50:58 +0000 (09:50 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 5 Mar 2013 09:21:36 +0000 (10:21 +0100)
We need it to restore the ilk rc6 context, since the gpu wait no
requires interrupts. But in general having interrupts around should
help in code sanity, since more and more stuff is interrupt driven.

This regression has been introduced in

commit 3e9605018ab3e333d51cc90fccfde2031886763b
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Nov 27 16:22:54 2012 +0000

    drm/i915: Rearrange code to only have a single method for waiting upon the ring

Like in the driver load code we need to make sure that hotplug
interrupts don't cause havoc with our modeset state, hence block them
with the existing infrastructure. Again we ignore races where we might
loose hotplug interrupts ...

Note that the driver load part of the regression has already been
fixed in

commit 52d7ecedac3f96fb562cb482c139015372728638
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Sat Dec 1 21:03:22 2012 +0100

    drm/i915: reorder setup sequence to have irqs for output setup

v2: Add a note to the commit message about which patch fixed the
driver load part of the regression. Stable kernels need to backport
both patches.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=54691
Cc: stable@vger.kernel.org (for 3.8 only, plese backport
    52d7ecedac3f96fb5 first)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reported-and-Tested-by: Ilya Tumaykin <itumaykin@gmail.com>
Reviewed-by: Chris wilson <chris@chris-wilson.co.uk> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_drv.c

index 2c5ee965e473faadf12258c786367ba8fe887ddb..0a8eceb7590244af233d153845a91c79e6662420 100644 (file)
@@ -495,6 +495,7 @@ static int i915_drm_freeze(struct drm_device *dev)
                intel_modeset_disable(dev);
 
                drm_irq_uninstall(dev);
+               dev_priv->enable_hotplug_processing = false;
        }
 
        i915_save_state(dev);
@@ -568,10 +569,20 @@ static int __i915_drm_thaw(struct drm_device *dev)
                error = i915_gem_init_hw(dev);
                mutex_unlock(&dev->struct_mutex);
 
+               /* We need working interrupts for modeset enabling ... */
+               drm_irq_install(dev);
+
                intel_modeset_init_hw(dev);
                intel_modeset_setup_hw_state(dev, false);
-               drm_irq_install(dev);
+
+               /*
+                * ... but also need to make sure that hotplug processing
+                * doesn't cause havoc. Like in the driver load code we don't
+                * bother with the tiny race here where we might loose hotplug
+                * notifications.
+                * */
                intel_hpd_init(dev);
+               dev_priv->enable_hotplug_processing = true;
        }
 
        intel_opregion_init(dev);