On Fri, Sep 11, 2020 at 3:43 PM Jakub Kicinski <k...@kernel.org> wrote:
>
> On Fri, 11 Sep 2020 15:34:24 -0700 Michael Chan wrote:
> > On Fri, Sep 11, 2020 at 12:53 PM Jakub Kicinski <k...@kernel.org> wrote:
> > >
> > > These stats are already reported in ethtool -S.
> > > Hopefully they are equivalent to standard stats?
> >
> > Yes.
> >
> > >
> > > Signed-off-by: Jakub Kicinski <k...@kernel.org>
> > > ---
> > > .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 19 +++++++++++++++++++
> > > 1 file changed, 19 insertions(+)
> > >
> > > diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> > > b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> > > index d0928334bdc8..b5de242766e3 100644
> > > --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> > > +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> > > @@ -1778,6 +1778,24 @@ static void bnxt_get_pauseparam(struct net_device
> > > *dev,
> > > epause->tx_pause = !!(link_info->req_flow_ctrl &
> > > BNXT_LINK_PAUSE_TX);
> > > }
> > >
> > > +static void bnxt_get_pause_stats(struct net_device *dev,
> > > + struct ethtool_pause_stats *epstat)
> > > +{
> > > + struct bnxt *bp = netdev_priv(dev);
> > > + struct rx_port_stats *rx_stats;
> > > + struct tx_port_stats *tx_stats;
> > > +
> > > + if (BNXT_VF(bp) || !(bp->flags & BNXT_FLAG_PORT_STATS))
> > > + return;
> > > +
> > > + rx_stats = (void *)bp->port_stats.sw_stats;
> > > + tx_stats = (void *)((unsigned long)bp->port_stats.sw_stats +
> > > + BNXT_TX_PORT_STATS_BYTE_OFFSET);
> > > +
> > > + epstat->rx_pause_frames = rx_stats->rx_pause_frames;
> > > + epstat->tx_pause_frames = tx_stats->tx_pause_frames;
> >
> > This will work, but the types on the 2 sides don't match. On the
> > right hand side, since you are casting to the hardware struct
> > rx_port_stats and tx_port_stats, the types are __le64.
> >
> > If rx_stats and tx_stats are *u64 and you use these macros:
> >
> > BNXT_GET_RX_PORT_STATS64(rx_stats, rx_pause_frames)
> > BNXT_GET_TX_PORT_STATS64(tx_stats, tx_pause_frames)
> >
> > the results will be the same with native CPU u64 types.
>
> Thanks! My build bot just poked me about this as well.
>
> I don't see any byte swaps in bnxt_get_ethtool_stats() -
> are they not needed there? I'm slightly confused.
No, swapping is not needed since we are referencing the sw_stats.
Every counter has already been swapped when we did the copy and
overflow check from the hw struct to sw_stats. sw_stats is exactly
the same as the hw struct except that every counter is already swapped
into native CPU u64 and properly adjusted for overflow.
