On Wed, Oct 31, 2018 at 4:37 AM Tariq Toukan <tar...@mellanox.com> wrote:
>
>
>
> On 30/10/2018 1:25 AM, Eric Dumazet wrote:
> > This patch has two changes :
> >
> > 1) Use netdev_tx_sent_queue_more() for skbs with xmit_more
> > This avoids mangling BQL status, since we only need to
> > take care of it for the last skb of the batch.
> >
> > 2) doorbel only depends on xmit_more and netif_tx_queue_stopped()
> >
> > While not strictly necessary after 1), it is more consistent
> > this way.
> >
> > Signed-off-by: Eric Dumazet <eduma...@google.com>
> > Cc: Tariq Toukan <tar...@mellanox.com>
> > ---
> > drivers/net/ethernet/mellanox/mlx4/en_tx.c | 10 ++++++++--
> > 1 file changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/mellanox/mlx4/en_tx.c
> > b/drivers/net/ethernet/mellanox/mlx4/en_tx.c
> > index
> > 1857ee0f0871d48285a6d3711f7c3e9a1e08a05f..3acce02ade6a115881ecd72e4710e332d3f380cb
> > 100644
> > --- a/drivers/net/ethernet/mellanox/mlx4/en_tx.c
> > +++ b/drivers/net/ethernet/mellanox/mlx4/en_tx.c
> > @@ -1006,7 +1006,6 @@ netdev_tx_t mlx4_en_xmit(struct sk_buff *skb, struct
> > net_device *dev)
> > ring->packets++;
> > }
> > ring->bytes += tx_info->nr_bytes;
> > - netdev_tx_sent_queue(ring->tx_queue, tx_info->nr_bytes);
> > AVG_PERF_COUNTER(priv->pstats.tx_pktsz_avg, skb->len);
> >
> > if (tx_info->inl)
> > @@ -1044,7 +1043,14 @@ netdev_tx_t mlx4_en_xmit(struct sk_buff *skb, struct
> > net_device *dev)
> > netif_tx_stop_queue(ring->tx_queue);
> > ring->queue_stopped++;
> > }
> > - send_doorbell = !skb->xmit_more || netif_xmit_stopped(ring->tx_queue);
> > +
> > + if (skb->xmit_more) {
> > + netdev_tx_sent_queue_more(ring->tx_queue, tx_info->nr_bytes);
> > + send_doorbell = netif_tx_queue_stopped(ring->tx_queue);
> > + } else {
> > + netdev_tx_sent_queue(ring->tx_queue, tx_info->nr_bytes);
> > + send_doorbell = true;
> > + }
> >
> > real_size = (real_size / 16) & 0x3f;
> >
> >
>
> The drivers' code template would be nicer if we unify the two functions
> netdev_tx_sent_queue/netdev_tx_sent_queue_more to a single one with a
> parameter.
>
> Currently, all drivers that would want to benefit from this optimization
> will have to repeat these if/else blocks.
I can add a helper sure, but I can not change drivers that I am not
able to test.
So I can not change existing helper.
This patch series shows the problem and fixes one driver, a common
helper can be added when a second
driver is updated, there is no hurry.
