Re: [RFC 0/4] Cursor support with universal planes

[Daniel Vetter <daniel@ffwll.ch>]

On Thu, May 15, 2014 at 06:17:25PM -0700, Matt Roper wrote:
> Cursor planes are a bit trickier to support via the universal plane interface
> than primary planes were.  Legacy cursor ioctls take handles to driver buffers
> directly whereas the universal plane API takes drm_framebuffer's that represent
> a buffer; due to this mismatch it isn't possible to implement a set of cursor
> helpers than provide "free" generic universal cursor support without driver
> changes as we did with primary planes.  Drivers will need to be updated to
> receive cursor support via the universal plane API.
> 
> If a driver tries to implement two interfaces for cursor support (legacy ioctl
> and universal planes), the reference counting can get very tricky/messy since
> we need to take into account userspace that may mix and match calls to both
> interfaces.  To address that, patch #1 in this set causes legacy cursor ioctls
> to be implemented using a driver's universal plane support if the driver
> registers a primary plane.  Calls to the legacy set_cursor ioctl will wrap the
              ^cursor plane I guess?

-Daniel

> provided driver buffer in a drm_framebuffer and then pass that along to the
> universal plane interface.  From a driver's point of view, this causes all
> cursor operations to arrive on the universal plane interface, regardless of
> which userspace API was used, which simplifies things greatly for the driver.
> 
> Patch #2 makes some minor changes to ensure drivers can successfully register a
> cursor plane with the DRM core.
> 
> Patch #3 does some minor i915 refactoring in preparation for the move to
> universal planes.
> 
> Patch #4 transitions the i915 driver to universal planes.  The changes here are
> intentionally minimal for ease of review.  It's likely that we could perform
> further cleanup in the future to eliminate some of the cursor state tracked in
> intel_crtc (e.g., cursor_width/cursor_height) since that information can be
> also be derived from crtc->cursor->fb.
> 
> 
> Matt Roper (4):
>   drm: Support legacy cursor ioctls via universal planes when possible
>   drm: Allow drivers to register cursor planes with crtc
>   drm/i915: Add intel_crtc_cursor_set_obj() to set cursor buffer
>   drm/i915: Switch to unified plane cursor handling
> 
>  drivers/gpu/drm/drm_crtc.c           | 113 ++++++++++++++++++++-
>  drivers/gpu/drm/i915/intel_display.c | 188 ++++++++++++++++++++++++++++-------
>  drivers/gpu/drm/i915/intel_drv.h     |   1 -
>  include/drm/drm_crtc.h               |   6 +-
>  4 files changed, 267 insertions(+), 41 deletions(-)
> 
> -- 
> 1.8.5.1
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch