Re: [PATCH] phy: nxp-c45-tja11xx: add interrupt support

[Vladimir Oltean]

On Fri, Apr 23, 2021 at 04:42:29PM +0300, Vladimir Oltean wrote:
> On Fri, Apr 23, 2021 at 03:07:51PM +0200, Andrew Lunn wrote:
> > > +static irqreturn_t nxp_c45_handle_interrupt(struct phy_device *phydev)
> > > +{
> > > +	irqreturn_t ret = IRQ_NONE;
> > > +	int irq;
> > > +
> > > +	irq = phy_read_mmd(phydev, MDIO_MMD_VEND1, VEND1_PHY_IRQ_STATUS);
> > > +	if (irq & PHY_IRQ_LINK_EVENT) {
> > > +		phy_trigger_machine(phydev);
> > > +		phy_write_mmd(phydev, MDIO_MMD_VEND1, VEND1_PHY_IRQ_ACK,
> > > +			      PHY_IRQ_LINK_EVENT);
> > 
> > The ordering here is interesting. Could phy_trigger_machine() cause a
> > second interrupt? Which you then clear without acting upon before
> > exiting the interrupt handler? I think you should ACK the interrupt
> > before calling phy_trigger_machine().
> 
> I thought that the irqchip driver keeps the interrupt line disabled
> until the handler finishes, and that recursive interrupts aren't a thing
> in Linux? Even with threaded interrupts, I thought this is what
> IRQF_ONESHOT deals with.

Ah, you mean the driver could be ACKing an event which was not the event
that triggered this IRQ. In that case, the ordering should be reversed,
sorry for the noise.