Re: [dpdk-dev] [PATCH] net/i40e: fix Tx fn selection when using new ethdev offloads

[Bruce Richardson <bruce.richardson@intel.com>]

On Tue, May 01, 2018 at 02:24:39PM +0100, Ananyev, Konstantin wrote:
> Hi Bruce,
> 
> > 
> > The Tx function selection code in the driver only used the older txq
> > flags values to check whether the scalar or vector functions should be
> > used. This caused performance regressions with testpmd io-fwd as the
> > scalar path rather than the vector one was being used in the default
> > case. Fix this by changing the code to take account of new offloads and
> > deleting the defines used for the old ones.
> > 
> > Fixes: 7497d3e2f777 ("net/i40e: convert to new Tx offloads API")
> > 
> > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > ---
> >  drivers/net/i40e/i40e_rxtx.c | 45 +++++++++++++++++++++++---------------------
> >  1 file changed, 24 insertions(+), 21 deletions(-)
> > 
> > diff --git a/drivers/net/i40e/i40e_rxtx.c b/drivers/net/i40e/i40e_rxtx.c
> > index ec1ce54ca..c523af575 100644
> > --- a/drivers/net/i40e/i40e_rxtx.c
> > +++ b/drivers/net/i40e/i40e_rxtx.c
> > @@ -40,9 +40,6 @@
> >  /* Base address of the HW descriptor ring should be 128B aligned. */
> >  #define I40E_RING_BASE_ALIGN	128
> > 
> > -#define I40E_SIMPLE_FLAGS ((uint32_t)ETH_TXQ_FLAGS_NOMULTSEGS | \
> > -					ETH_TXQ_FLAGS_NOOFFLOADS)
> > -
> >  #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
> > 
> >  #ifdef RTE_LIBRTE_IEEE1588
> > @@ -70,6 +67,12 @@
> >  #define I40E_TX_OFFLOAD_NOTSUP_MASK \
> >  		(PKT_TX_OFFLOAD_MASK ^ I40E_TX_OFFLOAD_MASK)
> > 
> > +static const uint64_t i40e_simple_ol_mask = (DEV_TX_OFFLOAD_MULTI_SEGS |
> > +		DEV_TX_OFFLOAD_VLAN_INSERT |
> > +		DEV_TX_OFFLOAD_SCTP_CKSUM |
> > +		DEV_TX_OFFLOAD_UDP_CKSUM |
> > +		DEV_TX_OFFLOAD_TCP_CKSUM);
> > +
> 
> Seems incomplete.
> From i40e_ethdev.c full-featured tx supports:
> dev_info->tx_offload_capa =
>                 DEV_TX_OFFLOAD_VLAN_INSERT |
>                 DEV_TX_OFFLOAD_QINQ_INSERT |
>                 DEV_TX_OFFLOAD_IPV4_CKSUM |
>                 DEV_TX_OFFLOAD_UDP_CKSUM |
>                 DEV_TX_OFFLOAD_TCP_CKSUM |
>                 DEV_TX_OFFLOAD_SCTP_CKSUM |
>                 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
>                 DEV_TX_OFFLOAD_TCP_TSO |
>                 DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
>                 DEV_TX_OFFLOAD_GRE_TNL_TSO |
>                 DEV_TX_OFFLOAD_IPIP_TNL_TSO |
>                 DEV_TX_OFFLOAD_GENEVE_TNL_TSO;
> 
> So we probably need the same here plus multiseg.
> BTW, it is really strange that we don't have multiseg in tx_offload_capa.
> Should be present I think.
> Might be worse to create a new define for it, or just use dev_info->tx_offload_capa directly.
> Konstantin
> 
Thinking about this more, it seems that right now we don't need a masks at
all. Any bits set in the offloads is going to cause us to use the scalar
path or to error out with an invalid offload requested. Yes, it's not
future-proofed in that it will need to be changed if we do end up
supporting some offloads with the vector path in future, but then the same
problem would occur if we just re-use the advertised capabilities too, like
you suggest.

Therefore I think for V2 we'll just check for a non-zero offloads value.

/Bruce