Hi, I've found it odd that the GEM object has its callbacks on drm_driver and not a vtable of its own. But something being odd isn't enough to make a change (me thinks). After working on the GEM shmem helper I saw that a few drivers have runtime support for 2 memory types for their buffers (shmem,vram,cma). I have realised that if the shmem helper was self contained wrt the callbacks, it would be easier for these types of drivers to use the helper. All they needed to do was to determine the buffer type on GEM object creation time and let the helper handle the rest of the callbacks. No sure if this makes sense or if the approach is to simplistic. Hence the RFC. I've added a patch to give an example of how this would look for the CMA helper and vc4 (I'm not volunteering to refactor the CMA helper and drivers). Noralf. Noralf Trønnes (3): drm/driver: Add defaults for .gem_prime_export/import callbacks drm/gem: Add drm_gem_object_funcs drm/cma: Use drm_gem_object_funcs Documentation/gpu/todo.rst | 7 ++ drivers/gpu/drm/drm_client.c | 12 ++- drivers/gpu/drm/drm_fb_helper.c | 8 +- drivers/gpu/drm/drm_gem.c | 108 +++++++++++++++++++++++++-- drivers/gpu/drm/drm_gem_cma_helper.c | 99 +++++++++---------------- drivers/gpu/drm/drm_prime.c | 50 +++++++------ drivers/gpu/drm/vc4/vc4_bo.c | 46 ++++++------ drivers/gpu/drm/vc4/vc4_drv.c | 26 +------ drivers/gpu/drm/vc4/vc4_drv.h | 6 +- include/drm/drm_drv.h | 4 + include/drm/drm_gem.h | 138 +++++++++++++++++++++++++++++++++++ include/drm/drm_gem_cma_helper.h | 17 +---- 12 files changed, 350 insertions(+), 171 deletions(-) -- 2.15.1 _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
