Bogus video resolution in Linux 3.5-rc4

[Sven Joachim]

After upgrading to Linux 3.5-rc4 from 3.4.4, I noticed that my monitor
switched to a resolution of 1280x960 rather than the native 1280x1024,
and nouveau has set up a framebuffer of 1680x945.  It goes without
saying that the result looks terrible.

Bisecting shows that the problem started with commit cb21aafe1:

--8<---------------cut here---------------start------------->8---
commit cb21aafe121b1c3ad4c77cc5c22320163f16ba42
Author: Adam Jackson <ajax@xxxxxxxxxx>
Date:   Fri Apr 13 16:33:36 2012 -0400

    drm/edid: Do drm_dmt_modes_for_range() for all range descriptor types
    
    EDID 1.4 retcons the meaning of the "GTF feature" bit to mean "is
    continuous frequency", and moves the set of supported timing formulas
    into the range descriptor itself.  In any event, the range descriptor
    can act as a filter on the DMT list without regard to a specific timing
    formula.
    
    Signed-off-by: Adam Jackson <ajax@xxxxxxxxxx>
    Tested-by: Takashi Iwai <tiwai@xxxxxxx>
    Reviewed-by: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx>
    Signed-off-by: Dave Airlie <airlied@xxxxxxxxxx>

diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
index cb40611..9363349 100644
--- a/drivers/gpu/drm/drm_edid.c
+++ b/drivers/gpu/drm/drm_edid.c
@@ -1042,12 +1042,13 @@ do_inferred_modes(struct detailed_timing *timing, void *c)
 {
 	struct detailed_mode_closure *closure = c;
 	struct detailed_non_pixel *data = &timing->data.other_data;
-	int gtf = (closure->edid->features & DRM_EDID_FEATURE_DEFAULT_GTF);
 
-	if (gtf && data->type == EDID_DETAIL_MONITOR_RANGE)
-		closure->modes += drm_dmt_modes_for_range(closure->connector,
-							  closure->edid,
-							  timing);
+	if (data->type != EDID_DETAIL_MONITOR_RANGE)
+		return;
+
+	closure->modes += drm_dmt_modes_for_range(closure->connector,
+						  closure->edid,
+						  timing);
 }
 
 static int
--8<---------------cut here---------------end--------------->8---

With a kernel from this commit nouveau sets up a resolution of
1400x1050; kernels built from commits cffd754 and fc48f16 complain about
an invalid EDID and use the 1024x768 fallback.

Attached is the /sys/class/drm/card0-DVI-I-1/edid file from a working
kernel; in 3.5-rc4 it is identical.

Cheers,
       Sven

Attachment: edid
Description: Binary data