GitLab

The original CL didn't account for most clients passing modifier 0 to
mean "don't know modifier".  The 0 modifier is DRM_FORMAT_MOD_LINEAR,
while DRM_FORMAT_MOD_INVALID is to be used to indicate that the client
doesn't know the modifier and triggers the legacy implicit layout
discovery:

https://gitlab.freedesktop.org/wayland/wayland-protocols/-/blob/master/unstable/linux-dmabuf/linux-dmabuf-unstable-v1.xml#L133

Specifically, what broke was Intel GPUs using tiled buffers and
importing those as EGLImages without specifying a modifier.  Without a
modifier, the GLES driver will get the tiling information from the
kernel.  One we started forwarding the DRM_FORMAT_MOD_LINEAR modifier
coming from the clients, that would take precedence and the driver
would view the buffer as linear, even though it was tiled.

We'll have to fix the various clients, but in the interim, we can add
support for non-linear modifiers by treating 0 as
DRM_FORMAT_MOD_INVALID like before.  This does the right thing for
linear buffers for all GPUs and keeps the mismatch in the Intel case
working.  At the same time it allows modifier-aware clients to pass in
explicit modifiers so that we can enable more efficient buffer
layouts.

BUG=b:145579089, b:79682290, b:162704898, chromium:1112017
TEST=wayland_rects_client --use-drm=msm and verify modifiers get passed

Change-Id: Ib00d4d6fe96e957c1f66205f2f5f515c09f3aabd
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2340164
Commit-Queue: Kristian H. Kristensen <hoegsberg@chromium.org>
Reviewed-by: Daniele Castagna <dcastagna@chromium.org>
Cr-Commit-Position: refs/heads/master@{#795280}