On Tue, Nov 22, 2022 at 09:38:43AM +0000, Russell King (Oracle) wrote:

Also, if we get the Marvell driver implementing validate_an_inband()
then I believe we can get rid of other parts of this patch - 88E1111 is
the commonly used accessible PHY on gigabit SFPs, as this PHY implements
I2C access natively. As I mentioned, Marvell PHYs can be set to no
inband, requiring inband, or inband with bypass mode enabled. So we
need to decide how we deal with that - especially if we're going to be
changing the mode from 1000base-X to SGMII (which we do on some SFP
modules so they work at 10/100/1000.)

For the Marvell 88E1111:

- If switching into 1000base-X mode, then bypass mode is enabled by
m88e1111_config_init_1000basex(). However, if AN is disabled in the
fibre page BMCR (e.g. by firmware), then AN won't be used.

- If switching into SGMII mode, then bypass mode is left however it was
originally set by m88e1111_config_init_sgmii() - so it may be allowed
or it may be disallowed, which means it's whatever the hardware
defaulted to or firmware set it as.

For the 88e151x (x=0,2,8) it looks like bypass mode defaults to being
allowed on hardware reset, but firmware may change that.

I don't think we make much of an effort to configure other Marvell
PHYs, relying on their hardware reset defaults or firmware to set
them appropriately.

So, I think for 88e151x, we should implement something like:

	int mode, bmcr, fscr2;

	/* RGMII too? I believe RGMII can signal inband as well, so we
	 * may need to handle that as well.
	 */
	if (interface != PHY_INTERFACE_MODE_SGMII &&
	    interface != PHY_INTERFACE_MODE_1000BASE_X)
		return PHY_AN_INBAND_UNKNOWN;

	bmcr = phy_read_paged(phydev, MII_MARVELL_FIBER_PAGE, MII_BMCR);
	if (bmcr < 0)
		return SOME_ERROR?

	mode = PHY_AN_INBAND_OFF;

	if (bmcr & BMCR_ANENABLE) {
		mode = PHY_AN_INBAND_ON;

		fscr2 = phy_read_paged(phydev, MII_MARVELL_FIBER_PAGE,
				       0x1a);
		if (fscr2 & BIT(6))
			mode |= PHY_AN_INBAND_TIMEOUT;
	}

	return mode;

Obviously adding register definitions for BIT(6) and 01a.

For the 88E1111:

	int mode, hwcfg;

	/* If operating in 1000base-X mode, we always turn on inband
	 * and allow bypass.
	 */
	if (interface == PHY_INTERFACE_MODE_1000BASEX)
		return PHY_AN_INBAND_ON | PHY_AN_INBAND_TIMEOUT;

	if (interface == PHY_INTERFACE_MODE_SGMII) {
		hwcfg = phy_read(phydev, MII_M1111_PHY_EXT_SR);
		if (hwcfg < 0)
			return SOME_ERROR?

		mode = PHY_AN_INBAND_ON;
		if (hwcfg & MII_M1111_HWCFG_SERIAL_AN_BYPASS)
			mode |= PHY_AN_INBAND_TIMEOUT;

		return mode;
	}

	return PHY_AN_INBAND_UNKNOWN;

Maybe?

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!