Re: [PATCH net-next 3/3] net: phy: dp83869: Support 1000Base-X SFP modules

[Romain Gantois <romain.gantois@bootlin.com>]

[-- Attachment #1: Type: text/plain, Size: 2219 bytes --]

On Wednesday, 14 May 2025 11:01:07 CEST Antoine Tenart wrote:
> On Wed, May 14, 2025 at 09:49:59AM +0200, Romain Gantois wrote:
> > +static int dp83869_port_configure_serdes(struct phy_port *port, bool
> > enable, +					 phy_interface_t interface)
> > +{
> > +	struct phy_device *phydev = port_phydev(port);
> > +	struct dp83869_private *dp83869;
> > +	int ret;
> > +
> > +	if (!enable)
> > +		return 0;
> > +
> > +	dp83869 = phydev->priv;
> > +
> > +	switch (interface) {
> > +	case PHY_INTERFACE_MODE_1000BASEX:
> > +		dp83869->mode = DP83869_RGMII_1000_BASE;
> > +		break;
> > +	default:
> > +		phydev_err(phydev, "Incompatible SFP module inserted\n");
> > +		return -EINVAL;
> > +	}
> > +
> > +	ret = dp83869_configure_mode(phydev, dp83869);
> > +	if (ret)
> > +		return ret;
> > +
> > +	/* Update advertisement */
> > +	if (mutex_trylock(&phydev->lock)) {
> > +		ret = dp83869_config_aneg(phydev);
> > +		mutex_unlock(&phydev->lock);
> > +	}
> 
> Just skimmed through this quickly and it's not clear to me why aneg is
> restarted only if there was no contention on the global phydev lock;
> it's not guaranteed a concurrent holder would do the same. If this is
> intended, a comment would be welcomed.

The reasoning here is that there are code paths which call 
dp83869_port_configure_serdes() with phydev->lock already held, for example:

phy_start() -> sfp_upstream_start() -> sfp_start() -> \
	sfp_sm_event() -> __sfp_sm_event() -> sfp_sm_module() -> \ 
	sfp_module_insert() -> phy_sfp_module_insert() -> \
	dp83869_port_configure_serdes()

so taking this lock could result in a deadlock.

mutex_trylock() is definitely not a perfect solution though, but I went with it
partly because the marvell-88x2222 driver already does it this way, and partly 
because if phydev->lock() is held, then there's a solid chance that the phy 
state machine is already taking care of reconfiguring the advertisement. 
However, I'll admit that this is a bit of a shaky argument.

If someone has a better solution in mind, I'll gladly hear it out, but for now 
I guess I'll just add a comment explaining why trylock() is being used.

Thanks!

-- 
Romain Gantois, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]