[DG2] exec_object.alignment cleanup in execbuf
Apparently in execbuf userspace can abused drm_i915_gem_exec_object.alignment into algorithm complexity, and it's apparently bad enough to we should stop it.
- this is only needed on gen2/3 afai remember
- userspace might still set this, so we need a full audit of anything that ever shipped
- isolation guarantees are pretty much pointless without full ppgtt (gen8+ for us)
Ideally we should simply reject this alignment parameter on gen4+. If that would break some userspace we need to check whether on some older platforms we need to silently reject it (i.e. not fail execbuf, but also not take the alignment into account). If the offending userspace is only sna/uxa, then we can also limit this to DRM_MASTER.
Either way on anything modern (gen8+) alignment parameter must be a no-op.
Task here is
- full umd audit
- adjust the kernel to reject this, including document this uapi fully with all restrictions in kerneldoc (it's probably not so well documented right now)
- adjusting the igts