[dpdk-dev] [PATCH 03/11] net/bnxt: implement vector mode driver

[Lance Richardson]

Hi Maxime, thanks for taking the time to review this.

On Thu, May 23, 2019 at 8:18 AM Maxime Coquelin
<maxime.coquelin at redhat.com> wrote:
...
> How do we enable it with Meson build? It seems to be missing.

Ah, thanks for pointing that out, I had completely missed the Meson
piece. Will fix in v2.

>
> I think we would better have a stub for bnxt_recv_pkts_vec() when
> RTE_LIBRTE_BNXT_INC_VECTOR isn't set, that would avoid having all these
> #ifdeferies in this file.
>
I'm having trouble picturing how having stubs would reduce the number
of #ifdefs,
but maybe the resulting code could be cleaner. I'll see what can be done for v2.
(Stubs with weak linkage might help, but that approach seems to be going
out of favor in http://patchwork.dpdk.org/patch/53498/).

> > -static inline uint32_t bnxt_tx_bds_in_hw(struct bnxt_tx_queue *txq)
> > +static inline uint32_t
> > +bnxt_tx_bds_in_hw(struct bnxt_tx_queue *txq)
>
> If to be done, it should be done in patch 1.
>
> >   {
> >       return ((txq->tx_ring->tx_prod - txq->tx_ring->tx_cons) &
> >               txq->tx_ring->tx_ring_struct->ring_mask);
> >   }
> >
> > -static inline uint32_t bnxt_tx_avail(struct bnxt_tx_queue *txq)
> > +static inline uint32_t
> > +bnxt_tx_avail(struct bnxt_tx_queue *txq)
>
> Ditto
>
I'll drop both of those changes from this patch in v2 (they don't seem
to be needed
since checkpatch was happy with patch 1).

Thanks again,

     Lance