REGRESSION from 3.8.0 to 3.8.1: Mirroring display works only with 1024x768 (4:3) whereas my laptop has 1366x768 (16:9) & external monitor 1920x1080 (16:9)

Till Kamppeter till.kamppeter at gmail.com
Wed Sep 25 00:50:56 PDT 2013 

I have a laptop with an internal 1366x768 (16:9) display and I have
connected and external 1920x1080 (16:9) monitor via HDMI. By default the
display gets mirrored with only 1024x768 (4:3) because this is the
highest common resolution of the two reported by xrandr. I can switch to
a large desktop formed by the two screens with the System Settings
application with each screen using its native resolution.

I have observed it with kernel 3.10.x and 3.11.x and with further
investigation I found out that this is a kernel regression from 3.8.0 to
3.8.1.

I can mirror with a higher resolution running the commands:

xrandr --addmode HDMI1 1366x768
xrandr --output HDMI1 --mode 1366x768
xrandr --output HDMI1 --mode 1366x768 --same-as LVDS1

This is rather awkward when using a projector.

The problem also persists when connecting the external screen to the
Mini DisplayPort (with adapter) instead of to the HDMI.

The output of "xrandr --prop" with both screens connected in
large-desktop-over-both-screens mode (no mirroring):

Screen 0: minimum 320 x 200, current 3286 x 1080, maximum 32767 x 32767
LVDS1 connected 1366x768+0+0 (normal left inverted right x axis y axis)
277mm x 156mm
 EDID:
  00ffffffffffff0030e47a0300000000
  00160103801c10780adce5975a559227
  1d505400000001010101010101010101
  010101010101241d56d4500016303020
  2500159c1000001b0000000000000000
  00000000000000000000000000fe004c
  4720446973706c61790a2020000000fe
  004c503132355748322d534c54310025
 BACKLIGHT: 1054 (0x0000041e)	range: (0,1054)
 Backlight: 1054 (0x0000041e)	range: (0,1054)
 scaling mode:	Full aspect
  supported: None Full Center Full aspect
   1366x768 59.8*+
   1360x768 59.8 60.0
   1024x768 60.0
   800x600 60.3 56.2
   640x480 59.9
VGA1 disconnected (normal left inverted right x axis y axis)
HDMI1 connected 1920x1080+1366+0 (normal left inverted right x axis y
axis) 510mm x 287mm
 EDID:
  00ffffffffffff0015c3692001010101
  2d14010380331d78eaee95a3544c9926
  0f5054a1080081808140b30081c00101
  010101010101023a801871382d40582c
  4500fe1f1100001e000000ff00353732
  32393131300a20202020000000fc0045
  5632333333570a2020202020000000fd
  003b3d1f440f000a2020202020200153
  02031b41468102031112042309070783
  01000067030c001300881e8c0ad08a20
  e02d10103e96000403000000188c0ad0
  8a20e02d10103e96001009000000188c
  0ad090204031200c4055000403000000
  188c0ad090204031200c405500100900
  000018011d007251d01e206e28550010
  090000001e000000000000000000005c
 Broadcast RGB:	Automatic
  supported: Automatic Full Limited 16:2
 audio:	auto
  supported: force-dvi off auto on
   1920x1080 60.0*+
   1680x1050 59.9
   1280x1024 60.0
   1280x960 60.0
   1280x720 60.0
   1024x768 60.0
   800x600 60.3
   720x576 50.0
   720x480 59.9
   640x480 60.0 59.9
   720x400 70.1
DP1 disconnected (normal left inverted right x axis y axis)
 Broadcast RGB:	Automatic
  supported: Automatic Full Limited 16:2
 audio:	auto
  supported: force-dvi off auto on
HDMI2 disconnected (normal left inverted right x axis y axis)
 Broadcast RGB:	Automatic
  supported: Automatic Full Limited 16:2
 audio:	auto
  supported: force-dvi off auto on
DP2 disconnected (normal left inverted right x axis y axis)
 Broadcast RGB:	Automatic
  supported: Automatic Full Limited 16:2
 audio:	auto
  supported: force-dvi off auto on


Original bug report to Ubuntu
-----------------------------

https://bugs.launchpad.net/bugs/1167301


Testing of older kernels
------------------------

I have tried several mainline kernels from 3.8.1 to 3.11: 3.11.0-rc3,
3.11.0, and also 3.10.0, 3.9.0, 3.8.1. They all show the problem.
Mainline kernel 3.8.0 is the newest where the problem does not occur.


Bisect result on upstream kernel
--------------------------------

Then I have done a bisect based on

git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git

finding that the following patch is the culprit:

till at till-twist:~/kernel/linux-stable$ git bisect good
740922ee575f8ee4daa2bfd3db5f69dd7573fc76 is the first bad commit
commit 740922ee575f8ee4daa2bfd3db5f69dd7573fc76
Author: Paulo Zanoni <paulo.r.zanoni at intel.com>
Date: Fri Feb 15 13:36:27 2013 -0200

    drm: don't add inferred modes for monitors that don't support them

    commit 196e077dc165a307efbd9e7569f81bbdbcf18f65 upstream.

    If bit 0 of the features byte (0x18) is set to 0, then, according to
    the EDID spec, "the display is non-continuous frequency (multi-mode)
    and is only specified to accept the video timing formats that are
    listed in Base EDID and certain Extension Blocks".

    For more information, please see the EDID spec, check the notes of the
    table that explains the "Feature Support" byte (18h) and also the
    notes on the tables of the section that explains "Display Range Limits
    & Additional Timing Description Definition (tag #FDh)".

    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45729
    Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
    Reviewed-by: Adam Jackson <ajax at redhat.com>
    Signed-off-by: Paulo Zanoni <paulo.r.zanoni at intel.com>
    Signed-off-by: Dave Airlie <airlied at redhat.com>
    Signed-off-by: Greg Kroah-Hartman <gregkh at linuxfoundation.org>

:040000 040000 26e271499e4ff7746011a296e569a985126aeb3f
d43c54ed20f43a1459c6adb7d922ee2be55c9317 M	drivers
till at till-twist:~/kernel/linux-stable$

Here is the patch:

till at till:~/ubuntu/linux/linux-stable$ git diff -r 740922e~..740922e
diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
index 5a3770f..584d5fd 100644
--- a/drivers/gpu/drm/drm_edid.c
+++ b/drivers/gpu/drm/drm_edid.c
@@ -2020,7 +2020,8 @@ int drm_add_edid_modes(struct drm_connector
*connector, st
        num_modes += add_cvt_modes(connector, edid);
        num_modes += add_standard_modes(connector, edid);
        num_modes += add_established_modes(connector, edid);
- num_modes += add_inferred_modes(connector, edid);
+ if (edid->features & DRM_EDID_FEATURE_DEFAULT_GTF)
+ num_modes += add_inferred_modes(connector, edid);
        num_modes += add_cea_modes(connector, edid);

        if (quirks & (EDID_QUIRK_PREFER_LARGE_60 |
EDID_QUIRK_PREFER_LARGE_75))
till at till:~/ubuntu/linux/linux-stable$


Possible fixes
--------------

The patch is simple and one can easily see that undoing it would solve
the problem.

It seems that some monitors (including my monitor) report that they are
non-continuous frequency but in reality they are. The simplest way to
fix this bug is simply to undo the patch to support such buggy monitors,
the correct fix would be an exception list with the monitors which
support resolutions which they do not report.

There should perhaps be a possibility to override this manually by a
run-time configuration bit so that for all monitors the non-reported
resolutions get listed.


As I have already found the patch causing the problem and also as I have
tested many mainline kernels I am not attaching all the /proc/... files
asked for in the "KERNEL.ORG FORMAT" on
https://wiki.ubuntu.com/KernelTeam/KernelTeamBugPolicies#Overview_on_Reporting_Bugs_Upstream

In addition, another user with different hardware has reported the same
problem in the Ubuntu bug report (Gizmo Chicken, CCed).

I hope this information helps to fix the problem. Please ask me if
further information is needed.

Note that I am not subscribed to the mailing list, please CC me.

   Till

More information about the dri-devel mailing list