[0/2] Fix observed mst problems with StarTech hub

Message

Lin, Wayne May 28, 2021, 1:55 p.m. UTC

Use Startech 1to3 DP hub to do some mst hotplug tests and find some
light up issues.

1. ACT polling timeout:
   Which is due to we didn't update DPCD payload table but still try
   to send ACT and polling for "ACT Handled" bit
2. Not all monitors light up:
   Due to we wrongly set unavailable VCP ID for new streams

Wayne Lin (2):
  drm/dp_mst: Do not set proposed vcpi directly
  drm/dp_mst: Avoid to mess up payload table by ports in stale topology

 drivers/gpu/drm/drm_dp_mst_topology.c | 65 ++++++++++++++++-----------
 1 file changed, 39 insertions(+), 26 deletions(-)

Comments

Lyude Paul June 11, 2021, 10:43 p.m. UTC | #1

haha. turns out it actually was a good thing I was busy with work today,
because I ended up testing some backports and running into the exact same MST
bug these patches appear to fix. How convienent :)

anyway-I looked over this and this looks good to me (and IMO, I like these
fixes more then the workarounds they replace!). The one thing we do need to
fix here though is this appears to definitely fix a regression, so we need to
make sure we actually bisect the issue that this patch is fixing so we can add
the appropriate Fixes: and Cc: tags so that these fixes get backported to
earlier stable kernel versions.

I definitely need this fix in asap though for my own work, so I am going to
see if I can start bisecting this. If I manage to figure out what's breaking
it before my workday ends today I'll just add my R-b and push this upstream,
otherwise I'll probably just push this first thing on monday. If you see this
message beforethen and know what kernel version introduced this issue, feel
free to respond ;)

On Fri, 2021-05-28 at 21:55 +0800, Wayne Lin wrote:
> Use Startech 1to3 DP hub to do some mst hotplug tests and find some
> light up issues.
> 
> 1. ACT polling timeout:
>    Which is due to we didn't update DPCD payload table but still try
>    to send ACT and polling for "ACT Handled" bit
> 2. Not all monitors light up:
>    Due to we wrongly set unavailable VCP ID for new streams
> 
> Wayne Lin (2):
>   drm/dp_mst: Do not set proposed vcpi directly
>   drm/dp_mst: Avoid to mess up payload table by ports in stale topology
> 
>  drivers/gpu/drm/drm_dp_mst_topology.c | 65 ++++++++++++++++-----------
>  1 file changed, 39 insertions(+), 26 deletions(-)
>

Lin, Wayne June 15, 2021, 4:38 a.m. UTC | #2

[Public]

Thanks Lyude for the review!

For the 1st patch, it's trying to fix the patch
7617e9621bf2 ("drm/dp_mst: clear time slots for ports invalid").

As for the 2nd one, it's my first time to test on this hub and I not
yet know the exact regression point. I'm also not quite sure if this
regression is caused by driver or just behavior changes from upper layer.
I was thinking that this patch is just an enhancement for our mst
helper functions and should be applicable for older kernel versions.

Thanks for your time!

Regards,
Wayne