On Mon, Aug 19, 2019 at 11:16:39AM -0400, Vivien Didelot wrote:
> On Mon, 19 Aug 2019 15:27:33 +0200, Andrew Lunn <[email protected]> wrote:
> > > @@ -45,7 +45,8 @@ static int mv88e6xxx_smi_direct_write(struct
> > > mv88e6xxx_chip *chip,
> > > {
> > > int ret;
> > >
> > > - ret = mdiobus_write_nested(chip->bus, dev, reg, data);
> > > + ret = mdiobus_write_sts_nested(chip->bus, dev, reg, data,
> > > + chip->ptp_sts);
> > > if (ret < 0)
> > > return ret;
> > >
> >
> > Please also make a similar change to mv88e6xxx_smi_indirect_write().
> > The last write in that function should be timestamped.
> >
> > Vivien, please could you think about these changes with respect to
> > RMU. We probably want to skip the RMU in this case, so we get slow but
> > uniform jitter, vs fast and unpredictable jitter from using the RMU.
>
> The RMU will have its own mv88e6xxx_bus_ops.
Yes, that is what i was expecting. But for this operation, triggering
a PTP timestamp, we probably want it to fall back to MDIO, which is
much more deterministic.
Andrew
