On 02/01/2018 12:04 PM, Maarten Lankhorst wrote:
Op 01-02-18 om 08:08 schreef Oleksandr Andrushchenko:
Hi, all!
I am working on a para-virtualized frontend DRM driver for Xen [1]
which implements display device I/O interface for Xen guest OSes [2].
At the moment I am rebasing the existing implementation from 4.9 kernel
to recent and found that when user-space DRM application exits then CRTC's
.set_config callback is not called to reset CRTC's mode to NULL.
I tracked the change down to commit 846c7dfc1193eb2f9866fe2fa0ae7d45c72f95b9
"drm/atomic: Try to preserve the crtc enabled state in drm_atomic_remove_fb, v2.
This introduces a slight behavioral change to rmfb. Instead of
disabling a crtc when the primary plane is disabled, we try to
preserve it.
<snip>
If the atomic commit is rejected by the driver then we will still
fall back to the old behavior and turn off the crtc.
"
which per my understanding is ok for real hardware, but in my case I still
need .set_config to be called with NULL: this way I can tell the backend
driver that it can release resources for this connection on its end.
I tried to implement drm_mode_config_funcs's .atomic_commit returning -EINVAL
on tear down, but at best it made CRTC's atomic state change to NOCRTC, but
still no .set_config callback seen.
Can anyone please tell me the right sequence to implement old remove FB
behavior for atomic drivers? Or what could be the problem on my side?
rmfb was never meant to mean that the crtc will be disabled as well. vmwgfx relied on that
behavior as well for old userspace support because userspace never called set_config.
If you're an atomic driver, the driver core will never call set_config any more, even if the
first attempt to disable returns -EINVAL. Even the fbcon helpers now perform direct
calls to atomic_commit().
understood, thank you for clarification
If you then still want to revert to old behavior and it doesn't make sense to have the
crtc enabled without primary plane enabled, you can copy the check from drm_simple_kms_crtc_check().
will try that, but at the moment this is not a strict requirement anymore,
just wanted to make sure I understand the flow correctly
~Maarten
Thank you,
Oleksandr
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
