drm/i915: vlv: sanitize RPS interrupt mask during GPU idling
authorImre Deak <imre.deak@intel.com>
Fri, 19 Dec 2014 17:33:27 +0000 (19:33 +0200)
committerJani Nikula <jani.nikula@intel.com>
Mon, 12 Jan 2015 08:52:41 +0000 (10:52 +0200)
We apply the RPS interrupt workaround on VLV everywhere except when
writing the mask directly during idling the GPU. For consistency do this
also there.

While at it also extend the code comment about affected platforms.
I couldn't reproduce the issue on VLV fixed by this workaround, by
removing the workaround from everywhere, while it's 100% reproducible on
SNB using igt/gem_reset_stats/ban-ctx-render. So also add a note that
it hasn't been verified if the workaround really applies to VLV/CHV.

Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
drivers/gpu/drm/i915/i915_irq.c
drivers/gpu/drm/i915/intel_pm.c

index ba86dc3..b051a23 100644 (file)
@@ -295,8 +295,10 @@ void gen6_enable_rps_interrupts(struct drm_device *dev)
 u32 gen6_sanitize_rps_pm_mask(struct drm_i915_private *dev_priv, u32 mask)
 {
        /*
-        * IVB and SNB hard hangs on looping batchbuffer
+        * SNB,IVB can while VLV,CHV may hard hang on looping batchbuffer
         * if GEN6_PM_UP_EI_EXPIRED is masked.
+        *
+        * TODO: verify if this can be reproduced on VLV,CHV.
         */
        if (INTEL_INFO(dev_priv)->gen <= 7 && !IS_HASWELL(dev_priv))
                mask &= ~GEN6_PM_RP_UP_EI_EXPIRED;
index 3801ff5..bf814a6 100644 (file)
@@ -4432,7 +4432,8 @@ static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
                return;
 
        /* Mask turbo interrupt so that they will not come in between */
-       I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
+       I915_WRITE(GEN6_PMINTRMSK,
+                  gen6_sanitize_rps_pm_mask(dev_priv, ~0));
 
        vlv_force_gfx_clock(dev_priv, true);