25/06/2025 17:00, Maayan Kashani: > Cross NUMA support means that when there is no > memory on the local NUMA, use other available > NUMA memory for port initialization and start. > > If mlx5 malloc with specified socket failed, > malloc function would return an error. > To support cross NUMA, > NUMA tolerant wrapper will be used on supported toolchains, > and it will log the exact place where fallback allocation happened. > NUMA tolerant flags in mlx5_malloc were added to support cases > where the toolchain does not support statement expressions. > > Add a NUMA tolerant wrapper to mlx5 malloc calls in > ipool and devx/memory region initializations > to support cross NUMA in device probing and port start stage. > > For Tx/Rx initializations, use socket-num given by the user > and when SOCKET_ID_ANY is passed, default to device NUMA > node with fallback to SOCKET_ID_ANY. > > Signed-off-by: Maayan Kashani <mkash...@nvidia.com> > Acked-by: Dariusz Sosnowski <dsosnow...@nvidia.com>
Applied, thanks.