On Tue, Oct 15, 2019 at 09:38:22PM +0200, Heiner Kallweit wrote:

On 15.10.2019 00:12, Russell King - ARM Linux admin wrote:

quoted

On Mon, Oct 14, 2019 at 10:20:15PM +0200, Heiner Kallweit wrote:

quoted

On 14.10.2019 21:51, Stefan Wahren wrote:

quoted

[add more recipients]

Am 14.10.19 um 21:25 schrieb Daniel Wagner:

quoted

Moving the phy_prepare_link() up in phy_connect_direct() ensures that
phydev->adjust_link is set when the phy_check_link_status() is called.

diff --git a/drivers/net/phy/phy_device.c

b/drivers/net/phy/phy_device.c index 9d2bbb13293e..2a61812bcb0d 100644

--- a/drivers/net/phy/phy_device.c +++ b/drivers/net/phy/phy_device.c

@@ -951,11 +951,12 @@ int phy_connect_direct(struct net_device *dev,

struct phy_device *phydev, if (!dev) return -EINVAL;

+       phy_prepare_link(phydev, handler);
+
        rc = phy_attach_direct(dev, phydev, phydev->dev_flags, interface);
        if (rc)

If phy_attach_direct() fails we may have to reset phydev->adjust_link to NULL,
as we do in phy_disconnect(). Apart from that change looks good to me.

Sorry, but it doesn't look good to me.

I think there's a deeper question here - why is the phy state machine
trying to call the link change function during attach?

After your comment I had a closer look at the lm78xx driver and few things
look suspicious:

- lan78xx_phy_init() (incl. the call to phy_connect_direct()) is called
  after register_netdev(). This may cause races.

That isn't a problem.  We have lots of network device drivers that do
this - in their open() function.

- The following is wrong, irq = 0 doesn't mean polling.
  PHY_POLL is defined as -1. Also in case of irq = 0 phy_interrupt_is_valid()
  returns true.

	/* if phyirq is not set, use polling mode in phylib */
	if (dev->domain_data.phyirq > 0)
		phydev->irq = dev->domain_data.phyirq;
	else
		phydev->irq = 0;

Also unlikely to be the cause of this problem.  phy_connect_direct() is
called with an adjust link function, which is set via
phy_prepare_link() in phy_connect_direct(), before interrupts are even
considered.

So, the window for the bug is somewhere before the call to
phy_prepare_link() in phy_connect_direct(), but after
lan78xx_mdio_init().

- Manually calling genphy_config_aneg() in lan78xx_phy_init() isn't
  needed, however this should not cause our problem.

Again, way after the point where phydev->adjust_link is non-NULL,
so this can't be it.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up