The buffers being submitted for software cursor were not allocated in a HWC-appropriate way, and that seems the likely culprit of this bug. fences aren't involved, it just seems that powervr and mali care more about the mapping bits on the buffer than adreno does.
lp: #1391261 is likely due to the BufferQueue and HwcDevice interaction handing out a buffer to the client that is fenced (the buffer is onscreen), and the client waiting on that fence.
The buffers being submitted for software cursor were not allocated in a HWC-appropriate way, and that seems the likely culprit of this bug. fences aren't involved, it just seems that powervr and mali care more about the mapping bits on the buffer than adreno does.
lp: #1391261 is likely due to the BufferQueue and HwcDevice interaction handing out a buffer to the client that is fenced (the buffer is onscreen), and the client waiting on that fence.