When the output surface of a context is imported with DRM PRIME descriptor,
and the buffer FD is already mapped to a virtual address for CPU read (may
be in another process), what is the correct way to synchronize the mapped
addresses after executing the context?
I have tried vaSyncSurface() after
vaBeginPicture()-vaRenderPicture()-vaEndPicture(), but the content read
from mapped addresses was incorrect. If I re-map the buffer FD, the content
in new mapped addresses would be correct. However I'm wondering if there's
way to avoid re-mapping the buffer to reduce overhead.