Unmapping part of a cpu mapped GEM BO corrupts the mapping
In this test : https://gitlab.freedesktop.org/llandwerlin/igt-gpu-tools/-/commits/wip/gem_mmap_offset-partial-unmap
We're mapping twice a GEM BO, writing the first page with a pattern A, and the last page with a pattern B.
Then we unmap part of one of the mapping and that appears to corrupt the binding location, resetting it to the beginning of the BO.
Reproduced on TGL kernel 6.1.
Edited by Lionel Landwerlin