Re: [PATCH net-next] net: phy: at803x: add cable diagnostics support

[Michael Walle]

Am 2020-05-10 16:52, schrieb Andrew Lunn:
> More of a note to self:
>
> Now we have three implementations, we start to see bits of common code
> which could be pulled out and shared.
>
> > +static bool at803x_cdt_fault_length_valid(u16 status)
> > +{
> > +	switch (FIELD_GET(AT803X_CDT_STATUS_STAT_MASK, status)) {
> > +	case AT803X_CDT_STATUS_STAT_OPEN:
> > +	case AT803X_CDT_STATUS_STAT_SHORT:
> > +		return true;
> > +	}
> > +	return false;
> > +}
>
> If we uses the netlink attribute values, not the PHY specific values,
> this could be put in the core.
>
> > +
> > +static int at803x_cdt_fault_length(u16 status)
> > +{
> > +	int dt;
> > +
> > +	/* According to the datasheet the distance to the fault is
> > +	 * DELTA_TIME * 0.824 meters.
> > +	 *
> > +	 * The author suspect the correct formula is:
> > +	 *
> > +	 *   fault_distance = DELTA_TIME * (c * VF) / 125MHz / 2
> > +	 *
> > +	 * where c is the speed of light, VF is the velocity factor of
> > +	 * the twisted pair cable, 125MHz the counter frequency and
> > +	 * we need to divide by 2 because the hardware will measure the
> > +	 * round trip time to the fault and back to the PHY.
> > +	 *
> > +	 * With a VF of 0.69 we get the factor 0.824 mentioned in the
> > +	 * datasheet.
> > +	 */
> > +	dt = FIELD_GET(AT803X_CDT_STATUS_DELTA_TIME_MASK, status);
> > +
> > +	return (dt * 824) / 10;
> > +}
>
> There seems to be a general consensus of 0.824 meters. So we could
> have helpers to convert back and forth in the core.

Yeah, but I don't know why VF is 0.69 (and it differs per cable, so
its just some kind of sensible default value). According to
https://en.wikipedia.org/wiki/Velocity_factor it should be something
around 0.64 or 0.65.

-michael

> > +static int at803x_cable_test_start(struct phy_device *phydev)
> > +{
> > +	/* Enable auto-negotiation, but advertise no capabilities, no link
> > +	 * will be established. A restart of the auto-negotiation is not
> > +	 * required, because the cable test will automatically break the 
> > link.
> > +	 */
> > +	phy_write(phydev, MII_BMCR, BMCR_ANENABLE);
> > +	phy_write(phydev, MII_ADVERTISE, ADVERTISE_CSMA);
> > +	phy_write(phydev, MII_CTRL1000, 0);
>
> Could be a genphy_ helper.
>
> Lets get the code merged, when we can come back and do some
> refactoring.
>
> 	Andrew