[xdp-hints] Re: [PATCH bpf-next v2 6/8] mlx4: Introduce mlx4_xdp_buff wrapper for xdp_buff

[Stanislav Fomichev <sdf@google.com>]

On Wed, Nov 23, 2022 at 6:33 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
> Stanislav Fomichev <sdf@google.com> writes:
>
> > No functional changes. Boilerplate to allow stuffing more data after xdp_buff.
> >
> > Cc: Tariq Toukan <tariqt@nvidia.com>
> > Cc: John Fastabend <john.fastabend@gmail.com>
> > Cc: David Ahern <dsahern@gmail.com>
> > Cc: Martin KaFai Lau <martin.lau@linux.dev>
> > Cc: Jakub Kicinski <kuba@kernel.org>
> > Cc: Willem de Bruijn <willemb@google.com>
> > Cc: Jesper Dangaard Brouer <brouer@redhat.com>
> > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > Cc: Alexander Lobakin <alexandr.lobakin@intel.com>
> > Cc: Magnus Karlsson <magnus.karlsson@gmail.com>
> > Cc: Maryam Tahhan <mtahhan@redhat.com>
> > Cc: xdp-hints@xdp-project.net
> > Cc: netdev@vger.kernel.org
> > Signed-off-by: Stanislav Fomichev <sdf@google.com>
> > ---
> >  drivers/net/ethernet/mellanox/mlx4/en_rx.c | 26 +++++++++++++---------
> >  1 file changed, 15 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> > index 8f762fc170b3..467356633172 100644
> > --- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> > +++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> > @@ -661,17 +661,21 @@ static int check_csum(struct mlx4_cqe *cqe, struct sk_buff *skb, void *va,
> >  #define MLX4_CQE_STATUS_IP_ANY (MLX4_CQE_STATUS_IPV4)
> >  #endif
> >
> > +struct mlx4_xdp_buff {
> > +     struct xdp_buff xdp;
> > +};
>
> This embedding trick works for drivers that put xdp_buff on the stack,
> but mlx5 supports XSK zerocopy, which uses the xsk_buff_pool for
> allocating them. This makes it a bit awkward to do the same thing there;
> and since it's probably going to be fairly common to do something like
> this, how about we just add a 'void *drv_priv' pointer to struct
> xdp_buff that the drivers can use? The xdp_buff already takes up a full
> cache line anyway, so any data stuffed after it will spill over to a new
> one; so I don't think there's much difference performance-wise.

I guess the alternative is to extend xsk_buff_pool with some new
argument for xdp_buff tailroom? (so it can kmalloc(sizeof(xdp_buff) +
xdp_buff_tailroom))
But it seems messy because there is no way of knowing what the target
device's tailroom is, so it has to be a user setting :-/
I've started with a priv pointer in xdp_buff initially, it seems fine
to go back. I'll probably convert veth/mlx4 to the same mode as well
to avoid having different approaches in different places..

> I'll send my patch to add support to mlx5 (using the drv_priv pointer
> approach) separately.

Saw them, thanks! Will include them in v3+.