(In reply to Frank Winklmeier from comment #0)
> For the Ubuntu 4.13.0-43 kernel, the problem is fixed after reverting the
> following 4 commits:

I'm curious about the statement, because...

> fd18173 Revert "drm/i915/edp: Allow alternate fixed mode for eDP if
> available."

dc911f5bd8aa ("drm/i915/edp: Allow alternate fixed mode for eDP if available.") was first introduced in v4.14.

> 4fb94d0 Revert "drm/i915/dp: rename intel_dp_is_edp to intel_dp_is_port_edp"

7b91bf7f9196 ("drm/i915/dp: rename intel_dp_is_edp to intel_dp_is_port_edp") was first introduced in v4.15.

> 7547d0c Revert "drm/i915/dp: make is_edp non-static and rename to
> intel_dp_is_edp"

1853a9daa19e ("drm/i915/dp: make is_edp non-static and rename to intel_dp_is_edp") was first introduced in v4.15.

> cb6c81f Revert "drm/i915/edp: Do not do link training fallback or prune
> modes on EDP"

a306343bcd7d ("drm/i915/edp: Do not do link training fallback or prune modes on EDP") was first introduced in v4.16.

Does the Ubuntu kernel have backports of this stuff that you have to revert?

> which seems to be consistent with the following revert on drm-tip:
> https://cgit.freedesktop.org/drm-tip/commit/drivers/gpu/drm/
> i915?id=368b554d63948133aca05e63ff8f5f4fbc2804d4
> 
> Could the relevant fixes be included int the mainline kernel? I expect this
> to affect everybody with a Samsung NP900X3F/G laptop.

I'm wondering what the relevant fixes would be in this case. Is 368b554d6394 ("Revert "drm/i915/edp: Allow alternate fixed mode for eDP if available."") enough alone on v4.17 release to fix it for you?

(In reply to Jani Nikula from comment #1)

> Does the Ubuntu kernel have backports of this stuff that you have to revert?

Yes, it seems they backport some of this. E.g.:

http://kernel.ubuntu.com/git/ubuntu/ubuntu-xenial.git/log/?h=hwe&qt=grep&q=training

> I'm wondering what the relevant fixes would be in this case. Is 368b554d6394
> ("Revert "drm/i915/edp: Allow alternate fixed mode for eDP if available."")
> enough alone on v4.17 release to fix it for you?

The situation seems a bit more complicated. From my tests on the 4.13 kernel, I had to revert all of the following four commits:

fd18173 Revert "drm/i915/edp: Allow alternate fixed mode for eDP if available."
4fb94d0 Revert "drm/i915/dp: rename intel_dp_is_edp to intel_dp_is_port_edp"
7547d0c Revert "drm/i915/dp: make is_edp non-static and rename to intel_dp_is_edp"
cb6c81f Revert "drm/i915/edp: Do not do link training fallback or prune modes on EDP"

Only reverting the first one did not help (I assume the 2nd and 3rd commit only affect code quality and don't matter here). Looking again at drm-tip I see that "Allow alternate fixed mode..." has been indeed reverted but the "link training" commit is still present. Since I don't see the problem with drm-tip, I assume some of the other changes that went in there (but not mainline) fixed the regression?

(In reply to Frank Winklmeier from comment #2)
> The situation seems a bit more complicated. From my tests on the 4.13
> kernel, I had to revert all of the following four commits:

To be blunt, I don't care what you have to revert on a downstream kernel with backports. Please use an upstream kernel for this, i.e. a released kernel from kernel.org.

Fair point. So starting from 4.17.2 I tried adding this commit:

db7c1c2 drm/i915/dp: Add support for HBR3 and TPS4 during link training

and then also reverting this one (starting again from 4.17.2):

80a4cb3 Revert "drm/i915/edp: Do not do link training fallback or prune modes on EDP"

Neither of these solved the problem. Since there is a large amount of differences between drm-tip and 4.17.2, can someone give a hint as to which commits could be responsible for fixing the problem in drm-tip?

I just tested the latest v4.18.0-rc1 kernel and the problem is fixed there. No more horizontal lines.

I hope this solved problem you were facing.
Please don't forget to close the issue!