[PATCH v2 0/2] drm/amd/display: Fix FreeSync infopacket refresh rates above 255 Hz

[Cooper Morgan <ext-github@cwooper.dev>]

On my MSI MAG 271QPX32 (320Hz QD-OLED over DP 1.4, RX 9070 XT), 320Hz
looks like a 720p image stretched to fill the screen whenever
Adaptive-Sync is on in the monitor's OSD. Custom modes at 305-315Hz put
it into its self-test pattern instead. Windows is fine on the same
machine and cable. It's the same bug as
https://gitlab.freedesktop.org/drm/amd/-/issues/5988, where an LG at
280Hz shows up as 24Hz.

The cause is the v1 FreeSync infopacket, which native DP sinks get
unless they're a whitelisted PCON or support Replay. It stores the min
and max refresh in one byte each, so 320 wraps to 64.

Patch 1 caps those bytes at 255 instead of letting them wrap. With
patch 1 alone, the MSI is correct at 320Hz and the LG shows every frame
at 280Hz, so I tagged it for stable.

Patch 2 sends the v3 packet to sinks that advertise AMD VSDB v3 and go
above 255Hz, so VRR gets the full range. Sinks at 255Hz and below are
unchanged.

This builds on what sunryze and oystersuki worked out in that issue.
sunryze found that the VSDB version was only saved for Replay sinks on
DP and tested sending v3 on their LG. oystersuki suggested pairing that
with a 255 cap for sinks that don't advertise v3. sunryze has no public
email, so they're credited here instead of with a tag.

Tested with the series backported to 7.2.13: 320Hz looks right, and
with VRR on, the monitor's refresh readout follows frame caps at 300,
270, 240, and 100fps. oystersuki tested both patches on an LG
27GX700A-B. Since v3 now carries the PQ bit from ebacf0a7ce54
("drm/amd/display: Signal PQ EOTF in FreeSync infopacket for DP HDR")
to these sinks in HDR, I also tested with that commit backported: HDR
looks the same. Both patches add KUnit tests.

Changes in v2:
- No functional changes; rebased onto amd-staging-drm-next.
- Patch 1: dropped the claim that an EDID range capped at 255 sends the
  bytes this patch sends (no FreeSync packet is sent then), and gave the
  result of running patch 1 alone instead. Also dropped "once the
  wrapped value lands inside their range", which the LG contradicts.
- Added tags from oystersuki and Harry.

v1: https://lore.kernel.org/amd-gfx/20261031041827.162730-1-ext-github@cwooper.dev/

Cooper Morgan (2):
  drm/amd/display: Saturate refresh rates in v1/v2 FreeSync infopackets
  drm/amd/display: Use FreeSync v3 infopacket for sinks above 255 Hz

 .../display/amdgpu_dm/amdgpu_dm_connector.c   |   2 +-
 .../display/amdgpu_dm/amdgpu_dm_freesync.c    |   7 ++
 .../tests/amdgpu_dm_connector_test.c          |  25 ++++
 .../amdgpu_dm/tests/amdgpu_dm_freesync_test.c | 116 ++++++++++++++++++
 .../amd/display/modules/freesync/freesync.c   |  12 +-
 5 files changed, 158 insertions(+), 4 deletions(-)

-- 
2.56.0