drm/i915: Don't register CRT connector when it's fused off
authorVille Syrjälä <ville.syrjala@linux.intel.com>
Fri, 20 Nov 2015 20:35:41 +0000 (22:35 +0200)
committerVille Syrjälä <ville.syrjala@linux.intel.com>
Tue, 24 Nov 2015 14:23:44 +0000 (16:23 +0200)
On some machines the CRT connector may be fused off. The weird thing
about this setup is that the ADPA register works otherwise normally,
except the enable bit is hardwired to 0. No one knows of any fuse
register that would tell us if this is the case, so the only thing we
can do (apart from a blacklist) is to try and set the enable bit and see
if it sticks. If not, we don't register the connector at all. Obviously
if the bit is already set when loading the driver we can just assume it
works.

I've smoke tested this approach on several machines (GMCH and PCH),
some with actual CRT connectors, some with shadow connectors, and
obviously the machine (IVB) with the fused off connector. So far
I've not seen any ill effects from this probe.

The main benefit is that we can actually run igt on machines with
fused off connectors, without totally upsetting the state checker.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1448051741-22771-1-git-send-email-ville.syrjala@linux.intel.com
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
drivers/gpu/drm/i915/intel_crt.c

index 27b3e610e8f0e47b92240b32855009db67075729..12008af797bd78049e15a6c14a295f851d7e88e3 100644 (file)
@@ -777,11 +777,37 @@ void intel_crt_init(struct drm_device *dev)
        struct intel_crt *crt;
        struct intel_connector *intel_connector;
        struct drm_i915_private *dev_priv = dev->dev_private;
+       i915_reg_t adpa_reg;
+       u32 adpa;
 
        /* Skip machines without VGA that falsely report hotplug events */
        if (dmi_check_system(intel_no_crt))
                return;
 
+       if (HAS_PCH_SPLIT(dev))
+               adpa_reg = PCH_ADPA;
+       else if (IS_VALLEYVIEW(dev))
+               adpa_reg = VLV_ADPA;
+       else
+               adpa_reg = ADPA;
+
+       adpa = I915_READ(adpa_reg);
+       if ((adpa & ADPA_DAC_ENABLE) == 0) {
+               /*
+                * On some machines (some IVB at least) CRT can be
+                * fused off, but there's no known fuse bit to
+                * indicate that. On these machine the ADPA register
+                * works normally, except the DAC enable bit won't
+                * take. So the only way to tell is attempt to enable
+                * it and see what happens.
+                */
+               I915_WRITE(adpa_reg, adpa | ADPA_DAC_ENABLE |
+                          ADPA_HSYNC_CNTL_DISABLE | ADPA_VSYNC_CNTL_DISABLE);
+               if ((I915_READ(adpa_reg) & ADPA_DAC_ENABLE) == 0)
+                       return;
+               I915_WRITE(adpa_reg, adpa);
+       }
+
        crt = kzalloc(sizeof(struct intel_crt), GFP_KERNEL);
        if (!crt)
                return;
@@ -815,12 +841,7 @@ void intel_crt_init(struct drm_device *dev)
                connector->interlace_allowed = 1;
        connector->doublescan_allowed = 0;
 
-       if (HAS_PCH_SPLIT(dev))
-               crt->adpa_reg = PCH_ADPA;
-       else if (IS_VALLEYVIEW(dev))
-               crt->adpa_reg = VLV_ADPA;
-       else
-               crt->adpa_reg = ADPA;
+       crt->adpa_reg = adpa_reg;
 
        crt->base.compute_config = intel_crt_compute_config;
        if (HAS_PCH_SPLIT(dev) && !HAS_DDI(dev)) {