Created attachment 110508 [details]
dmesg of boot, with drm.debug=0xe

Adding video=SVIDEO-1:d as kernel parameter seems to fix (or work around) the issue. (Thanks to Chris Wilson for pointing that out in #31519)

What's the output of that sequence of xrandr?

( xrandr --output LVDS1 --off; xrandr --current; xrandr ) >& attachme.txt

Created attachment 110540 [details]
xrandr --output LVDS1 --off; xrandr --current; xrandr

This is the requested output. Note that this command sequence doesn't trigger the problem, i.e. the LVDS output stays off.
See next attachment that was generated with a sleep between the commands.

Not sure this is related, but I got the following in dmesg (during login):

[   41.996076] [drm:drm_edid_block_valid] *ERROR* EDID checksum is invalid, remainder is 26
[   41.996089] Raw EDID:
[   41.996097]  	00 ff ff ff ff ff ff 00 4c 2d 6b 03 36 32 49 4b
[   41.996104]  	17 13 01 03 0e 37 22 a0 2a fe 21 a8 53 37 ae 24
[   41.996110]  	11 50 54 bf ef 80 a9 40 81 80 81 40 71 4f 01 01
[   41.996116]  	01 01 01 01 01 01 28 3c 80 a0 70 b0 23 40 30 20
[   41.996123]  	36 00 26 57 21 00 00 1a 00 00 00 fd 00 38 4b 1e
[   41.996129]  	51 11 00 0a 41 ff ff ff ff ff ff ff ff ff ff ff
[   41.996135]  	ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
[   41.996141]  	ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff

Created attachment 110541 [details]
xrandr --output LVDS1 --off; sleep 2; xrandr --current; sleep 2; xrandr

This triggers the bug, the LVDS output is reenabled during the last call to xrandr.

From #intel-gfx

1417982504 < ickle> ukleinek: I suspect that it is you DE responding to the
    RandR notifications and resetting the outputs
[...]
1417986438  * ukleinek goes, installs openbox and plugs the external monitor
    into the X41 again
1417987536 < ukleinek> ickle: doesn't happen on openbox
1417987648 < ukleinek> ickle: output of xrandr --current and xrandr are
    identical then.

(In reply to Uwe Kleine-König from comment #6)
> From #intel-gfx
> 
> 1417982504 < ickle> ukleinek: I suspect that it is you DE responding to the
>     RandR notifications and resetting the outputs
> [...]
> 1417986438  * ukleinek goes, installs openbox and plugs the external monitor
>     into the X41 again
> 1417987536 < ukleinek> ickle: doesn't happen on openbox
> 1417987648 < ukleinek> ickle: output of xrandr --current and xrandr are
>     identical then.

In light of this, do we still consider this a bug?

(In reply to Jani Nikula from comment #7)
> (In reply to Uwe Kleine-König from comment #6)
> > From #intel-gfx
> > 
> > 1417982504 < ickle> ukleinek: I suspect that it is you DE responding to the
> >     RandR notifications and resetting the outputs
> > [...]
> > 1417986438  * ukleinek goes, installs openbox and plugs the external monitor
> >     into the X41 again
> > 1417987536 < ukleinek> ickle: doesn't happen on openbox
> > 1417987648 < ukleinek> ickle: output of xrandr --current and xrandr are
> >     identical then.
> 
> In light of this, do we still consider this a bug?

No answer, presuming it isn't.

There is still the issue that S-Video fluctuates between unknown and disconnected which causes the undesired behaviour, i.e. it would be fixed by:

diff --git a/drivers/gpu/drm/i915/intel_tv.c b/drivers/gpu/drm/i915/intel_tv.c
index 8b9d325bda3c..f15683a489fb 100644
--- a/drivers/gpu/drm/i915/intel_tv.c
+++ b/drivers/gpu/drm/i915/intel_tv.c
@@ -1337,12 +1337,12 @@ intel_tv_detect(struct drm_connector *connector, bool force)
                                connector_status_disconnected :
                                connector_status_connected;
                } else
-                       status = connector_status_unknown;
+                       status = connector->status;
 
                drm_modeset_drop_locks(&ctx);
                drm_modeset_acquire_fini(&ctx);
        } else
-               return connector->status;
+               status = connector->status;
 
        if (status != connector_status_connected)
                return status;

There's just a small debate over what is the correct status value to return when the output detection is forced but we are unable to do so. I think not wilfully converting to unknown is beneficial.

http://patchwork.freedesktop.org/patch/51176/

Please try kernel v4.4.

The request to test with latest kernel was made months ago, please update kernel and update bug.
If there is no response within a month the bug will be closed

uwe+bugsfreenode@kleine-koenig.org

This bug is now been closed because of the lack of activity and response from submitter in order to test in recent configuration.

Closing bug