Created attachment 96212 [details]
GDB backtrace and register info of the crash

For the first time ever, since I'm having a PITCAIRN GPU I was able to run a
full Piglit quick test without crashing my X (hooray for that). During the test
I came across two similar crashes in
"spec/EXT_texture_array/fbo-generatemipmap-array RGB9_E5" and
"spec/ARB_framebuffer_object/fbo-generatemipmap-3d RGB9_E5" (see the attached
GDB backtrace; the backtrace is for
spec/EXT_texture_array/fbo-generatemipmap-array RGB9_E5", but the second
crashed in the same function on the same line and as it's the same format, that
is tested, I guess, these are related, if not, I'm happy to provide a backtrace
for the second case as well and split the bugs).
I can't tell whether this is a regression or not, since this was the first time
I was able to run Piglit on this machine without getting a corrupted results
file, due to the crash of the X server.

My graphics stack is:
GPU: "PITCAIRN" (ChipID = 0x6819)
Linux: 3.13.6
libdrm: 2.4.52-1
LLVM: SVN:trunk/r204517
libclc: Git:master/1e278a7b04
Mesa: Git:master/4c79f088c0
GLAMOR: Git:master/a4fbc7732a (Standalone)
DDX: Git:master/ea6d0affe5
X: 1.15.0-2

Let me know if you need further information.