[RFC 0/2] atomicify fbdev stuff

Daniel Vetter daniel at ffwll.ch
Tue Aug 25 01:53:01 PDT 2015

On Wed, Aug 19, 2015 at 06:45:15PM -0400, Rob Clark wrote:
> So the issues that Inki and Bjorn ran into with ww_acquire_fini() vs
> legacy fbdev codepaths (like restore_fbdev_mode() and pan_display())
> would be solved by using a single atomic update for these cases.  So
> I hacked up a prototype.
> (disclaimer, these are completely untested hacked-up-between-sessions-
> at-conference patches.. but at least they compile ;-))
> I think Daniel wanted something other than DRIVER_ATOMIC check, since
> i915 currently supports atomic other than async.  Other option would
> be a separate DRIVER_ATOMIC_ASYNC flag (which has some advantage, in
> that we could start using atomic modeset, but not atomic pageflip,
> on i915.. and perhaps finally start work on atomic xrandr).  Other
> option would be to, for now, just introduce an fb_helper->use_atomic
> flag, which driver could override, instead.

Not a separate driver flag, but simply something to force fbdev helper to
use atomic without DRIVER_ATOMIC. That way partially-converted drivers can
switch things on their own. With that addressed and testing done this
looks ready imo and I'll pull it in.

> For now we should probably revert the ww_acquire_fini patch, and then
> shoot for re-introducing it on top of these patches (hopefully working
> by then) for 4.4.  And maybe in the process we can fix some i915 multi-
> monitor VT switch fail.

Yeah, Dave already pushed the revert.
> Rob Clark (2):
>   drm/fb-helper: atomic restore_fbdev_mode()..
>   drm/fb-helper: atomic pan_display()..
>  drivers/gpu/drm/drm_atomic_helper.c | 129 ++++++++++++++++++++--------------
>  drivers/gpu/drm/drm_fb_helper.c     | 133 ++++++++++++++++++++++++++++++++++++
>  include/drm/drm_atomic_helper.h     |   6 ++
>  3 files changed, 215 insertions(+), 53 deletions(-)
> -- 
> 2.4.3

Daniel Vetter
Software Engineer, Intel Corporation

More information about the dri-devel mailing list