On Fri, Aug 06, 2021 at 03:17:56PM -0700, Saeed Mahameed wrote: > On Fri, 2021-08-06 at 14:50 -0700, Kees Cook wrote: > > [...] > > So, split the memcpy() so the compiler can reason about the buffer > > sizes. > > > > "pahole" shows no size nor member offset changes to struct > > mlx5e_tx_wqe > > nor struct mlx5e_umr_wqe. "objdump -d" shows no meaningful object > > code changes (i.e. only source line number induced differences and > > optimizations). > > spiting the memcpy doesn't induce any performance degradation ? extra > instruction to copy the 1st 2 bytes ? Not meaningfully, but strictly speaking, yes, it's a different series of instructions. > [...] > > --- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c > > +++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c > > why only here ? mlx5 has at least 3 other places where we use this > unbound memcpy .. Can you point them out? I've been fixing only the ones I've been able to find through instrumentation (generally speaking, those with fixed sizes). -- Kees Cook
