Bug 111505

Created attachment 145181 [details]
dmesg from booting good commit

Created attachment 145182 [details]
dmesg from booting bad commit

Created attachment 145183 [details]
xrandr --verbose (from good boot)

Considering this as a regression setting the priority to Highest. 

(In reply to Geoffrey Bennett from comment #0)
> Steps to reproduce the issue:
> - Fedora 30, Lenovo ThinkPad T480s laptop, internal + 2x external DisplayPort
> - External monitors: LG 24EB23 1920x1200 (DP-2-2) and LG 32UK550 3840x2160
> (DP-2-3)
> - Boot kernel commit f1477219869cc943ef9e0248b308da7b900c65ec or Fedora
> stock 5.2.9
> - At the LUKS password prompt, only two monitors display it (mirrored); the
> 3840x2160 monitor only displays a text cursor in the top-left
> - After logging in, the third monitor cannot be enabled
> - Boot kernel before the mentioned commit (or Fedora stock 5.1.19); all
> three monitors are working (mirrored password prompt)
> 
> How often do these steps trigger the issue? Always.
> 
> If I did the bisect right, this commit is the issue:
> 
> commit f1477219869cc943ef9e0248b308da7b900c65ec
> Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
> Date:   Tue Mar 26 16:25:53 2019 +0200
> 
>     drm/i915: Remove the 8bpc shackles from DP MST
>     
>     Allow DP MST to output any color depth. This means deep color as
>     well as falling back to 6bpc if we would otherwise require too
>     much bandwidth.
>     
>     TODO: We should probably extend bw_contstrained scheme to force
>     all streams on the link to 6bpc if we can't fit the new stream(s)
>     otherwise.
> 
> Bad commit: f1477219869cc943ef9e0248b308da7b900c65ec
> Good commit: 765bdb0b3948747dda8c9bab300013e34f76fe88
> 
> Comparing dmesg output between good & bad commits with drm.debug=0x1e ...
> 
> Good commit, I see messages like:
> 
> [    7.215513] [drm:intel_atomic_check [i915]] hw max bpp: 30, pipe bpp: 24,
> dithering: 0
> [    7.215977] [drm:intel_dump_pipe_config [i915]] cpu_transcoder: C, pipe
> bpp: 24, dithering: 0 
> [    7.216361] [drm:drm_dp_mst_atomic_check [drm_kms_helper]] [MST
> PORT:000000006766b031] requires 48 vcpi slots
> [    7.216365] [drm:drm_dp_mst_atomic_check [drm_kms_helper]] [MST
> PORT:0000000048ae3b07] requires 14 vcpi slots
> 
> Bad commit, I see messages like:
> 
> [    7.151412] [drm:intel_atomic_check [i915]] hw max bpp: 30, pipe bpp: 30,
> dithering: 0
> [    7.151878] [drm:intel_dump_pipe_config [i915]] cpu_transcoder: C, pipe
> bpp: 30, dithering: 0
> [    7.152272] [drm:drm_dp_mst_atomic_check [drm_kms_helper]] [MST
> PORT:00000000f27944b7] requires 60 vcpi slots
> [    7.152276] [drm:drm_dp_mst_atomic_check [drm_kms_helper]] [MST
> PORT:00000000f43eff2e] requires 14 vcpi slots
> [    7.152281] [drm:drm_dp_mst_atomic_check [drm_kms_helper]] [MST
> PORT:00000000f43eff2e] not enough VCPI slots in mst state 00000000c1ef406a
> (avail=3)
> [    7.152293] [drm:drm_atomic_check_only [drm]] atomic driver check for
> 00000000a9c59397 failed: -28
> 
> I also tested drm-tip, and neither external monitor worked then.

@Ville, can you help here?

Dang. I was hoping no one would hit this in practice, at least any more than they already have. We need to beef up the code to reduce the bpp of the MST streams going over the same link until they fit.

As a stopgap we probable want a revert. I'll send one out.

This should get us back to the old behaviour:
https://patchwork.freedesktop.org/series/65919/

I tested kernel v5.3-rc6 with the patch at https://patchwork.freedesktop.org/series/65919/ and am happy to report that it fixes the problem.

Thanks for the quick turnaround!

(In reply to Ville Syrjala from comment #6)
> This should get us back to the old behaviour:
> https://patchwork.freedesktop.org/series/65919/

Thanks for providing the fix immediately Ville.

(In reply to Geoffrey Bennett from comment #7)
> I tested kernel v5.3-rc6 with the patch at
> https://patchwork.freedesktop.org/series/65919/ and am happy to report that
> it fixes the problem.
> 
> Thanks for the quick turnaround!

Thanks for the feedback.

I will wait for the fix to land in drmtip and then I can close this issue. Meanwhile I will drop the priority and assign it to myself.

Now fixed.

commit 75427b2a2bffc083d51dec389c235722a9c69b05
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Wed Aug 28 13:20:59 2019 +0300

    drm/i915: Limit MST to <= 8bpc once again