Re: [PATCH net v3 1/1] net: pcs: xpcs: fix incorrect CL37 AN sequence

From: Vladimir Oltean <vladimir.oltean@nxp.com>
To: "Russell King (Oracle)" <linux@armlinux.org.uk>
Cc: Wong Vee Khee <vee.khee.wong@linux.intel.com>,
	"David S . Miller" <davem@davemloft.net>,
	Jose Abreu <Jose.Abreu@synopsys.com>,
	Andrew Lunn <andrew@lunn.ch>,
	Heiner Kallweit <hkallweit1@gmail.com>,
	Jakub Kicinski <kuba@kernel.org>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Wong Vee Khee <veekhee@gmail.com>
Subject: Re: [PATCH net v3 1/1] net: pcs: xpcs: fix incorrect CL37 AN sequence
Date: Fri, 1 Oct 2021 10:12:20 +0000	[thread overview]
Message-ID: <20211001101219.g5llt6biexzij7ev@skbuf> (raw)
In-Reply-To: <YVbbitGJDZhd6eW/@shell.armlinux.org.uk>

On Fri, Oct 01, 2021 at 10:57:30AM +0100, Russell King (Oracle) wrote:
> On Fri, Oct 01, 2021 at 12:19:52AM +0000, Vladimir Oltean wrote:
> > static int xpcs_config_aneg_c37_sgmii(struct dw_xpcs *xpcs, unsigned int mode)
> > {
> > 	int ret, mdio_ctrl1, old_an_ctrl, an_ctrl, old_dig_ctrl1, dig_ctrl1;
> > 
> > 	/* Disable SGMII AN in case it is already enabled */
> > 	mdio_ctrl1 = xpcs_read(xpcs, MDIO_MMD_VEND2, DW_VR_MII_MMD_CTRL);
> > 	if (mdio_ctrl1 < 0)
> > 		return mdio_ctrl1;
> > 
> > 	if (mdio_ctrl1 & AN_CL37_EN) {
> > 		ret = xpcs_write(xpcs, MDIO_MMD_VEND2, DW_VR_MII_MMD_CTRL,
> > 				 mdio_ctrl1 & ~AN_CL37_EN);
> > 		if (ret < 0)
> > 			return ret;
> > 	}
> 
> This is fine...
> 
> > 	if (!(mdio_ctrl1 & AN_CL37_EN) && phylink_autoneg_inband(mode)) {
> > 		ret = xpcs_write(xpcs, MDIO_MMD_VEND2, DW_VR_MII_MMD_CTRL,
> > 				 mdio_ctrl1 | AN_CL37_EN);
> > 		if (ret)
> > 			return ret;
> > 	}
> 
> This is not. If the control register had AN_CL37_EN set initially, then
> in the first test above, we clear the bit. However, mdio_ctrl1 will
> still contain the bit set. When we get here, we will skip setting the
> register bit back to one even if in-band mode was requested.
> 
> As I said in a previous email, at this point there is no reason to check
> the previous state, because if it was set on entry, we will have cleared
> it, so the register state at this point has the bit clear no matter
> what. If we need to set it, then we /always/ need to write it here - it
> doesn't matter what the initial state was.

Correct, it should have been:

	if (phylink_autoneg_inband(mode)) {
		ret = xpcs_write(xpcs, MDIO_MMD_VEND2, DW_VR_MII_MMD_CTRL,
				 mdio_ctrl1 | AN_CL37_EN);
		if (ret)
			return ret;
	}

For the record, just in case this code gets copied anywhere, there's
another mistake in my placement of one of the comments.

	/* If using in-band autoneg, enable automatic speed/duplex mode change
	 * by HW after SGMII AN complete.
	 * 5) VR_MII_MMD_CTRL Bit(12) [AN_ENABLE] = 1b (Enable SGMII AN)            <- this part of the comment
	 */
	old_dig_ctrl1 = xpcs_read(xpcs, MDIO_MMD_VEND2, DW_VR_MII_DIG_CTRL1);
	if (old_dig_ctrl1 < 0)
		return old_dig_ctrl1;

really belongs here:

	/* If using SGMII AN, enable it here */
	if (phylink_autoneg_inband(mode)) {
		ret = xpcs_write(xpcs, MDIO_MMD_VEND2, DW_VR_MII_MMD_CTRL,
				 mdio_ctrl1 | AN_CL37_EN);
		if (ret)
			return ret;
	}