[PATCH 0/9] drm: Revert general use of struct drm_gem_object.dma_buf

[Linus Torvalds]

On Fri, 11 Jul 2025 at 02:40, Thomas Zimmermann <tzimmermann at suse.de> wrote:
>
> Reverting the whole thing is the only sensible action here.

I'm assuming this is against some current drm tree, not mine, because
it doesn't apply here.

I'll try the smaller set of reverts that Simona suggested for my
testing, and will get back to you on that.

Background for others: current -git ends up having odd hangs when
Xwayland starts for me (not at boot, but when I log in). It seems to
be very timing-dependent, so presumably I'm just unlucky with my
hardware.

The timing-dependence has made my attempts to bisect it very
frustrating and non-conclusive, but reverting 5307dce878d4 ("drm/gem:
Acquire references on GEM handles for framebuffers") made things work
_better_, to the point where I thought it was fixed.

Until the hang came back.

The "hang" is not some kind of kernel hang, it seems to be user-space
getting confused by the changes. Because I can still switch back to a
text VT, and there are no messages in the kernel logs. And the system
"recovers", in that after a 30-second timeout (it feels like a minute,
but logs seem to imply it really is just half a minute), I get

   gnome-session-manager at gnome.service: start operation timed out. Terminating.
   gnome-session-manager at gnome.service: Failed with result 'timeout'.

and it restarts, and things work fine on the second try. Except when
they don't - on one reboot it went through this twice, and

Now, because it's obviously timing-sensitive, it might be something
entirely different going on and I have no guarantees, but it does seem
to be related to that commit 5307dce878d4.

And I only see this on one of my machines. Again, might be timing, but
might also be that it only shows up on certain hardware (in my case:
Radeon RX 580).

           Linus