mirror of
https://github.com/tiagovignatti/intel-gpu-tools.git
synced 2025-06-08 08:26:10 +00:00
Current userptr kernel implementation downgrades tracking VMA ranges (real userspace ones) to an inefficient linear walk for any process which has instantiated overlapping userptr objects. This adds a test which shows the performance cliff on, most visibly, generic userspace mmap(2) and munmap(2) operations between unsync, non-overlapping and overlapping userptr objects. Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Thomas Daniel <thomas.daniel@intel.com>