drm/i915: Hide underruns from eDP PLL and port enable on ILK
authorVille Syrjälä <ville.syrjala@linux.intel.com>
Thu, 29 Oct 2015 19:25:58 +0000 (21:25 +0200)
committerVille Syrjälä <ville.syrjala@linux.intel.com>
Tue, 10 Nov 2015 14:23:36 +0000 (16:23 +0200)
We get underruns on the other pipe when enabling the CPU eDP PLL and
port on ILK.

Bspec knows about the PLL issue, and recommends doing a vblank wait just
prior to enabling the PLL. That does seem to help, but unfortunately we
get another underrun when actually enabling the CPU eDP port. Bspec
doesn't mention that at all, and the same vblank wait trick doesn't
appear to be effective there.

Since I have no better clue how to deal with this, just hide the errors.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446146763-31821-10-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
drivers/gpu/drm/i915/intel_dp.c

index 278f025f4b20bd48246eb649683ef988f198e4f9..d4e4f7a1a11b33dfc23e98ae153abcf5bc8d8fe7 100644 (file)
@@ -2578,6 +2578,8 @@ static void intel_enable_dp(struct intel_encoder *encoder)
        struct drm_i915_private *dev_priv = dev->dev_private;
        struct intel_crtc *crtc = to_intel_crtc(encoder->base.crtc);
        uint32_t dp_reg = I915_READ(intel_dp->output_reg);
+       enum port port = dp_to_dig_port(intel_dp)->port;
+       enum pipe pipe = crtc->pipe;
 
        if (WARN_ON(dp_reg & DP_PORT_EN))
                return;
@@ -2589,6 +2591,17 @@ static void intel_enable_dp(struct intel_encoder *encoder)
 
        intel_dp_enable_port(intel_dp);
 
+       if (port == PORT_A && IS_GEN5(dev_priv)) {
+               /*
+                * Underrun reporting for the other pipe was disabled in
+                * g4x_pre_enable_dp(). The eDP PLL and port have now been
+                * enabled, so it's now safe to re-enable underrun reporting.
+                */
+               intel_wait_for_vblank_if_active(dev_priv->dev, !pipe);
+               intel_set_cpu_fifo_underrun_reporting(dev_priv, !pipe, true);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, !pipe, true);
+       }
+
        edp_panel_vdd_on(intel_dp);
        edp_panel_on(intel_dp);
        edp_panel_vdd_off(intel_dp, true);
@@ -2611,7 +2624,7 @@ static void intel_enable_dp(struct intel_encoder *encoder)
 
        if (crtc->config->has_audio) {
                DRM_DEBUG_DRIVER("Enabling DP audio on pipe %c\n",
-                                pipe_name(crtc->pipe));
+                                pipe_name(pipe));
                intel_audio_codec_enable(encoder);
        }
 }
@@ -2634,13 +2647,28 @@ static void vlv_enable_dp(struct intel_encoder *encoder)
 
 static void g4x_pre_enable_dp(struct intel_encoder *encoder)
 {
+       struct drm_i915_private *dev_priv = to_i915(encoder->base.dev);
        struct intel_dp *intel_dp = enc_to_intel_dp(&encoder->base);
-       struct intel_digital_port *dport = dp_to_dig_port(intel_dp);
+       enum port port = dp_to_dig_port(intel_dp)->port;
+       enum pipe pipe = to_intel_crtc(encoder->base.crtc)->pipe;
 
        intel_dp_prepare(encoder);
 
+       if (port == PORT_A && IS_GEN5(dev_priv)) {
+               /*
+                * We get FIFO underruns on the other pipe when
+                * enabling the CPU eDP PLL, and when enabling CPU
+                * eDP port. We could potentially avoid the PLL
+                * underrun with a vblank wait just prior to enabling
+                * the PLL, but that doesn't appear to help the port
+                * enable case. Just sweep it all under the rug.
+                */
+               intel_set_cpu_fifo_underrun_reporting(dev_priv, !pipe, false);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, !pipe, false);
+       }
+
        /* Only ilk+ has port A */
-       if (dport->port == PORT_A) {
+       if (port == PORT_A) {
                ironlake_set_pll_cpu_edp(intel_dp);
                ironlake_edp_pll_on(intel_dp);
        }