Re: [PATCH] net: phy: marvell-88q2xxx: add driver for the Marvell 88Q2220 PHY

[Dimitri Fedrau]

Am Fri, Jan 05, 2024 at 03:00:58PM +0100 schrieb Andrew Lunn:
> On Fri, Jan 05, 2024 at 01:42:21PM +0100, Dimitri Fedrau wrote:
> > Am Tue, Dec 19, 2023 at 04:57:50PM +0100 schrieb Andrew Lunn:
> > > > I am not sure that it will be accepted by the maintainers if you use a
> > > > lot of registers that are not documented.
> > > 
> > > Sometimes there is no choice, there is no documentation except the
> > > vendor crap driver which we try to clean up as much as possible, but
> > > we still end up with lots of magic numbers.
> > >
> > 
> > Hi Andrew, hi Stefan,
> > 
> > tried to reduce the init sequence. This worked for me:
> > 
> > static int mv88q222x_config_init(struct phy_device *phydev)
> > {
> > 	int ret;
> > 
> > 	/* send_s detection threshold, slave and master */
> > 	ret = phy_write_mmd(phydev, MDIO_MMD_AN, 0x8032, 0x2020);
> > 	if (ret < 0)
> > 		return ret;
> > 
> > 	ret = phy_write_mmd(phydev, MDIO_MMD_AN, 0x8031, 0xa28);
> > 	if (ret < 0)
> > 		return ret;
> > 
> > 	ret = phy_write_mmd(phydev, MDIO_MMD_AN, 0x8031, 0xc28);
> > 	if (ret < 0)
> > 		return ret;
> > 
> > 	ret = phy_write_mmd(phydev, MDIO_MMD_PCS, 0xffe4, 0xc);
> > 	if (ret < 0)
> > 		return ret;
> > 
> > 	return mv88q2xxx_config_init(phydev);
> > }
> > 
> > The four register writes were required to make the PHY work in 1000Mbit forced
> > mode. When using autonegotiation or 100Mbit forced mode they weren't needed.
> > It was enough to write them once in mv88q222x_config_init as you can
> > see. Thanks Stefan for the hint with the first three register writes, it
> > helped a lot.
> 
> Hi Dimitri
> 
Hi Andrew,

> Do we need to reduce the init sequence? Since this is all undocumented
> magic which nobody understands, it would be safer to just keep with
> the Marvell vendor crap code dump. Unless we really do need to change
> it.
>
You are right, it would be safer to use the vendor code. But when
looking at the vendor code, the init sequence changed a lot from rev. B0
to rev. B1 of the PHY. There are some additional register writes, but
mostly the order of the register writes changed. I don't know if this is
going to be worse in the future. Maintaining different revisions will
probably take some effort or at least result in bloated code. We probably
don't need all of the init sequence. I'm not sure how to deal with it,
keeping the init sequence at a minimum is probably a good idea.

> 	Andrew

Best regards,
Dimitri