Re: [RFC PATCH v3 net-next 2/6] net: phylink: introduce a generic method for querying PHY in-band autoneg capability

On Thu, Sep 23, 2021 at 12:31:16AM +0300, Vladimir Oltean wrote:

On Wed, Sep 22, 2021 at 10:22:19PM +0100, Russell King (Oracle) wrote:

quoted

On Wed, Sep 22, 2021 at 09:14:42PM +0300, Vladimir Oltean wrote:

quoted

+static unsigned int phylink_fixup_inband_aneg(struct phylink *pl,
+					      struct phy_device *phy,
+					      unsigned int mode)
+{
+	int ret;
+
+	ret = phy_validate_inband_aneg(phy, pl->link_interface);
+	if (ret == PHY_INBAND_ANEG_UNKNOWN) {
+		phylink_dbg(pl,
+			    "PHY driver does not report in-band autoneg capability, assuming %s\n",
+			    phylink_autoneg_inband(mode) ? "true" : "false");
+
+		return mode;
+	}
+
+	if (phylink_autoneg_inband(mode) && !(ret & PHY_INBAND_ANEG_ON)) {
+		phylink_err(pl,
+			    "Requested in-band autoneg but driver does not support this, disabling it.\n");

If we add support to the BCM84881 driver to work with
phy_validate_inband_aneg(), then this will always return
PHY_INBAND_ANEG_OFF and never PHY_INBAND_ANEG_ON. Consequently,
this will always produce this "error". It is not an error in the
SFP case, but it is if firmware is misconfigured.

So, this needs better handling - we should not be issuing an error-
level kernel message for something that is "normal".

Is this better?

		phylink_printk(phy_on_sfp(phy) ? KERN_DEBUG : KERN_ERR, pl,
			       "Requested in-band autoneg but driver does not support this, disabling it.\n");

Ah, not sure whether that was a trick question or not, but
phylink_fixup_inband_aneg function does not get called for the SFP code
path, I even noted this in the commit message but forgot:

|   So if the 3 code paths:
|   - phylink_sfp_config
|   - phylink_connect_phy
|   - phylink_fwnode_phy_connect
|
|   do more or less the same thing (adapt pl->cur_link_an_mode based on the
|   capability reported by the PHY), the intention is different. With SFP
|   modules this behavior is absolutely to be expected, and pl->cfg_link_an_mode
|   only denotes the initial operating mode. On the other hand, when the PHY
|   is on-board, the initial link AN mode should ideally also be the final
|   one. So the implementations for the three are different.

That's why phy_validate_inband_aneg is called twice, once in
phylink_sfp_config and once for the on-board case.