On Fri, Jul 22, 2022 at 12:43:27PM +0200, Haris Iqbal wrote: > IMHO, in addition to finding a use case for this, another question > would be, whether putting this restriction of "rxe_map_mr_sg" needed > for every reg_fast_mr/invalidate_mr() cycle makes sense in terms of > spec and/or mlx drivers. It doesn't really, the mapping of the MR is supposed to be permanent, the only thing invalidate does is block use of the mkey. revalidating without remapping shouldn't be impossible. Jason
