While debugging the libdrm duplicate buffer object adventure, I managed to temporarily understand object lifetimes in the __DRIimage getBuffers path, and also to compare that to the DRI2 getBuffers path. Here are a couple of small fixes.
I haven't looked at the i915 code, but I suspect the first one, and parts of the second one would apply.
[PATCH 1/2] i965: Correct check for re-bound buffer in
The code was attempting to avoid re-creating the miptree when the loader handed back the same bufffer we already had, but it was confused about how to tell -- turns out the object actually shared between the loader and the driver is the BO, not the region. And so, we compare BO pointers to see if the image buffer needs to have a new miptree.
[PATCH 2/2] i965: Set fast color clear mcs_state on newly allocated
Here's a chunk of code that was just missing from the __DRIimage path as a result of rebasing across some new driver additions.
Both of these fixes are candidates for 10.0 as they affect the __DRIimage paths now used by Wayland.
-keith