Re: [PATCH] net: phy: Don't disable irqs on shutdown if WoL is enabled

From: Ioana Ciornei
Date: Thu Aug 10 2023 - 07:01:16 EST


On Wed, Aug 09, 2023 at 08:15:27PM +0100, Russell King (Oracle) wrote:
> On Wed, Aug 09, 2023 at 06:21:58PM +0200, Andrew Lunn wrote:
> > > Thinking about this, I wonder whether we could solve your issue by
> > > disabling interrupts when the PHY is probed, rather than disabling
> > > them on shutdown - something like this? (not build tested)
> > >
> > > diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
> > > index 3e9909b30938..4d1a37487923 100644
> > > --- a/drivers/net/phy/phy_device.c
> > > +++ b/drivers/net/phy/phy_device.c
> > > @@ -3216,6 +3216,8 @@ static int phy_probe(struct device *dev)
> > > goto out;
> > > }
> > >
> > > + phy_disable_interrupts(phydev);
> > > +
> > > /* Start out supporting everything. Eventually,
> > > * a controller will attach, and may modify one
> > > * or both of these values
> >
> > At some point, the interrupt is going to be enabled again. And then
> > the WoL interrupt will fire. I think some PHY drivers actually need to
> > see that WoL interrupt. e.g. the marvell driver has this comment:
> >
> > static int m88e1318_set_wol(struct phy_device *phydev,
> > struct ethtool_wolinfo *wol)
> > {
> > ....
> > /* If WOL event happened once, the LED[2] interrupt pin
> > * will not be cleared unless we reading the interrupt status
> > * register. If interrupts are in use, the normal interrupt
> > * handling will clear the WOL event. Clear the WOL event
> > * before enabling it if !phy_interrupt_is_valid()
> > */
> >
> > So it seems like just probing the marvell PHY is not enough to clear
> > the WoL interrupt.
> >
> > Can we be sure that the other PHY has reached a state it can handle
> > and clear an interrupt when we come to enable the interrupt? I think
> > not, especially in cases like NFS root, where the interface will be
> > put into use as soon as it exists, maybe before the other interface
> > has probed.
>
> I suppose the question to Ioana would be - are the two AR8031 PHYs on
> the same MDIO bus? If they are, then we're safe, because both will be
> probed consecutively (because they're using the same driver.)
>

Yes, the two AR8031 PHYs are on the same MDIO bus.

I just tested your approach to disable the interrupts at phy_probe() and
it seems to be working. I also tested with NFS boot using one of the
PHYs and it's behaving ok.

I think I'm ok with this approach as long as Uwe's problem is also
fixed. I don't know how a wake-on-lan procedure works and if it matters
if the WoL interrupt is lost before the PHY driver gets to know about
it.

Ioana