On Mon, 2019-06-10 at 10:43 +0200, Paolo Abeni wrote:
> Hi,
>
> On Fri, 2019-06-07 at 18:09 +0000, Saeed Mahameed wrote:
> > On Thu, 2019-06-06 at 23:56 +0200, Paolo Abeni wrote:
> > > We can avoid another indirect call per packet wrapping the rx
> > > handler call with the proper helper.
> > >
> > > To ensure that even the last listed direct call experience
> > > measurable gain, despite the additional conditionals we must
> > > traverse before reaching it, I tested reversing the order of the
> > > listed options, with performance differences below noise level.
> > >
> > > Together with the previous indirect call patch, this gives
> > > ~6% performance improvement in raw UDP tput.
> > >
> > > v1 -> v2:
> > > - update the direct call list and use a macro to define it,
> > > as per Saeed suggestion. An intermediated additional
> > > macro is needed to allow arg list expansion
> > >
> > > Signed-off-by: Paolo Abeni <pab...@redhat.com>
> > > ---
> > > drivers/net/ethernet/mellanox/mlx5/core/en.h | 4 ++++
> > > drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 5 ++++-
> > > 2 files changed, 8 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > > b/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > > index 3a183d690e23..52bcdc87cbe2 100644
> > > --- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > > +++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > > @@ -148,6 +148,10 @@ struct page_pool;
> > >
> > > #define MLX5E_MSG_LEVEL NETIF_MSG_LINK
> > >
> > > +#define MLX5_RX_INDIRECT_CALL_LIST \
> > > + mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe,
> > > mlx5i_handle_rx_cqe, \
> > > + mlx5e_ipsec_handle_rx_cqe
> > > +
> > > #define mlx5e_dbg(mlevel, priv, format,
> > > ...) \
> > > do
> > > { \
> > > if (NETIF_MSG_##mlevel & (priv)->msglevel) \
> > > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> > > b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> > > index 0fe5f13d07cc..7faf643eb1b9 100644
> > > --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> > > +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> > > @@ -1303,6 +1303,8 @@ void mlx5e_handle_rx_cqe_mpwrq(struct
> > > mlx5e_rq
> > > *rq, struct mlx5_cqe64 *cqe)
> > > mlx5_wq_ll_pop(wq, cqe->wqe_id, &wqe->next.next_wqe_index);
> > > }
> > >
> > > +#define INDIRECT_CALL_LIST(f, list, ...) INDIRECT_CALL_4(f,
> > > list,
> > > __VA_ARGS__)
> > > +
> >
> > Hi Paolo,
> >
> > This patch produces some compiler errors:
> >
> > Please note that mlx5e_ipsec_handle_rx_cqe is only defined when
> > CONFIG_MLX5_EN_IPSEC is enabled.
>
> I'm sorry, I dumbly did not fuzz vs mlx5 build options.
>
> It looks like that, to cope with all the possible mixes, a not-so-
> nice
> macro maze is required; something alike the following:
>
> #if defined(CONFIG_MLX5_EN_IPSEC) && defined (CONFIG_MLX5_CORE_IPOIB)
>
> #define MLX5_RX_INDIRECT_CALL_LIST \
> mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe,
> mlx5i_handle_rx_cqe, \
> mlx5e_ipsec_handle_rx_cqe
> #define INDIRECT_CALL_LIST(f, list, ...) INDIRECT_CALL_4(f, list,
> __VA_ARGS__)
>
> #elif defined(CONFIG_MLX5_EN_IPSEC)
>
> #define MLX5_RX_INDIRECT_CALL_LIST \
> mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe, \
> mlx5e_ipsec_handle_rx_cqe
> #define INDIRECT_CALL_LIST(f, list, ...) INDIRECT_CALL_3(f, list,
> __VA_ARGS__)
>
> #elif defined(CONFIG_MLX5_CORE_IPOIB)
>
> #define MLX5_RX_INDIRECT_CALL_LIST \
> mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe,
> mlx5i_handle_rx_cqe
> #define INDIRECT_CALL_LIST(f, list, ...) INDIRECT_CALL_3(f, list,
> __VA_ARGS__)
>
> #else
>
> #define MLX5_RX_INDIRECT_CALL_LIST \
> mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe
> #define INDIRECT_CALL_LIST(f, list, ...) INDIRECT_CALL_2(f, list,
> __VA_ARGS__)
>
> #endif
>
> If you are ok with the above, I can include it in v3, otherwise I can
> either:
>
> * drop patch 2/3 and use only the 2 alternatives
> (mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe) that are available
> regardless of the driver build options
>
yea, the above is too much, maybe we can simplify, I will revisit it
later, for now, let's have the 2 functions that are always available,
after all they are the ones that really matter.
> * drop both patches 2/3 and 3/3
>
> Any feedback welcome, thanks!
>
> Paolo
>
