On 10/3/26 09:11, Greg KH wrote:
On Sat, Oct 03, 2026 at 03:15:23PM +0200, Miguel Mesa wrote:
commit 7e5760f084d0 ("drm/amd/display: add HDMI 2.1 Compliance Support")
selects YCbCr 4:2:0 for a 4:2:0-only mode only when the IGT override asks
for it, and YCbCr 4:4:4 only when that override is YCbCr 4:4:4. On a
normal boot the override is unset.
A 3840x2160@60 mode that the sink lists only in the YCbCr 4:2:0 block
then falls through to 4:4:4 or RGB, and the Wayland session cannot build
the output layer: kwin_wayland_drm: Failed to create framebuffer (EINVAL)
right after login. v7.3-rc5 selects 4:2:0 again for those modes when the
color format is automatic. Do the same here, including HDMI FRL.
Keep automatic YCbCr 4:4:4 for the other HDMI modes, as 7.1 did. The TMDS
depth fallback also runs for HDMI FRL, so a mode that does not fit at
4:4:4 can still drop to 4:2:0.
This is not a backport of a mainline commit. The 7.3 fix is the rewrite
in 0b0ff65d3c ("drm/amd/display: Refactor stream validation"), which
cannot be cherry-picked to 7.2;
Why not? What's preventing that? We'll gladly take large sets of
patches to fix bugs, in fact we prefer to.
It's definitely not a clean cherry-pick because of the refactoring and
other fixes in 7.2.y already.
I let an agent chew on it a while with a /goal, and even it couldn't
come up with a working solution:
----
- `0b0ff65d3c` requires the connector split beginning with `0e967e086e75`.
- That split depends on `f5165625b8b2`, `ee55bf7d6a63`, `4734e045f49d`,
and `0618dec49415`.
- Even the earliest prerequisite, `f5165625b8b2`, conflicts with later
stable changes already in 7.2.9.
- All six available rebases of the connector extraction also conflict.
A clean backport requires an upstream prerequisite series rebased onto
this 7.2-stable tree; otherwise conflict resolution would be necessary.
----
So I'd say directionally this solution that Miguel built make sense.
Clean up the tags as Greg suggested though.
Jerry Zuo confirmed the pre-7.3 logic is
gone there. This patch restores the pre-regression selection on the 7.2
code, following the v7.3-rc5 rule. Mario Limonciello asked for it to be
sent to stable in drm/amd#5760.
Regression: black screen after login on HDMI sinks whose modes fall
through to RGB, e.g. Strix Halo (gfx1151) + TCL 4K TV whose 3840x2160@60
is YCbCr 4:2:0-only (VIC 97). 7e5760f084d0 shipped in 7.2; still unfixed
in 7.2.9.
Tested on CachyOS linux-cachyos-bore 7.2.7-3 (znver4) on that Strix Halo
box: Plasma runs at 3840x2160@60 with the same userspace and firmware
that stayed black on every stock 7.2.x kernel.
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5760
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5893
Reported-by: Michał Górny <[email protected]>
Reported-by: Miguel Mesa <[email protected]>
Suggested-by: Mario Limonciello <[email protected]>
Tested-by: Miguel Mesa <[email protected]>
Signed-off-by: Miguel Mesa <[email protected]>
Co-authored-by: Cursor <[email protected]>
A bot can not be an author. Please read the documentation for how to
use a LLM in the kernel.
thanks,
greg k-h
