2024-02-14 17:59:31

by Andrew Lunn

[permalink] [raw]
Subject: Re: [PATCH v6 net-next 09/14] net: phy: marvell-88q2xxx: add cable test support

> +static int mv88q222x_cable_test_get_status(struct phy_device *phydev,
> + bool *finished)
> +{
> + int ret;
> + u32 dist;
> +
> + ret = phy_read_mmd(phydev, MDIO_MMD_PCS, MDIO_MMD_PCS_MV_TDR_STATUS);
> + if (ret < 0)
> + return ret;
> +
> + *finished = true;

That looks odd. Is there no status bit which says it has completed? Is
it guaranteed to complete within a fixed time? How is it guaranteed that
mv88q222x_cable_test_get_status() is called at the necessary delay after
mv88q222x_cable_test_start()?

Andrew


2024-02-15 21:04:06

by Dimitri Fedrau

[permalink] [raw]
Subject: Re: [PATCH v6 net-next 09/14] net: phy: marvell-88q2xxx: add cable test support

Am Wed, Feb 14, 2024 at 06:54:58PM +0100 schrieb Andrew Lunn:
> > +static int mv88q222x_cable_test_get_status(struct phy_device *phydev,
> > + bool *finished)
> > +{
> > + int ret;
> > + u32 dist;
> > +
> > + ret = phy_read_mmd(phydev, MDIO_MMD_PCS, MDIO_MMD_PCS_MV_TDR_STATUS);
> > + if (ret < 0)
> > + return ret;
> > +
> > + *finished = true;
>
> That looks odd. Is there no status bit which says it has completed? Is
> it guaranteed to complete within a fixed time? How is it guaranteed that
> mv88q222x_cable_test_get_status() is called at the necessary delay after
> mv88q222x_cable_test_start()?
>
According to the datasheet and the Marvell API bits(0:1) can be used to
check if the test has completed. Sample code waits 500ms before checking
the bits. If the test is not completed after the delay the corresponding
function returns with an error.

I just used bits(7:4) where 2'b1000 means that the test is in progress,
and setting *finished = false. I didn't introduced any delay, relying
on the reschedule delay of the PHY state machine. I didn't notice any
problems with this approach. Anyway if the test does not complete for
whatever reasons we get stuck here, right ? Don't know if this can
happen. Probably we should take the safer path described in the Marvell
API.

Dimitri

2024-02-16 15:01:53

by Andrew Lunn

[permalink] [raw]
Subject: Re: [PATCH v6 net-next 09/14] net: phy: marvell-88q2xxx: add cable test support

On Thu, Feb 15, 2024 at 10:03:50PM +0100, Dimitri Fedrau wrote:
> Am Wed, Feb 14, 2024 at 06:54:58PM +0100 schrieb Andrew Lunn:
> > > +static int mv88q222x_cable_test_get_status(struct phy_device *phydev,
> > > + bool *finished)
> > > +{
> > > + int ret;
> > > + u32 dist;
> > > +
> > > + ret = phy_read_mmd(phydev, MDIO_MMD_PCS, MDIO_MMD_PCS_MV_TDR_STATUS);
> > > + if (ret < 0)
> > > + return ret;
> > > +
> > > + *finished = true;
> >
> > That looks odd. Is there no status bit which says it has completed? Is
> > it guaranteed to complete within a fixed time? How is it guaranteed that
> > mv88q222x_cable_test_get_status() is called at the necessary delay after
> > mv88q222x_cable_test_start()?
> >
> According to the datasheet and the Marvell API bits(0:1) can be used to
> check if the test has completed. Sample code waits 500ms before checking
> the bits. If the test is not completed after the delay the corresponding
> function returns with an error.

Thanks for the explanation.

>
> I just used bits(7:4) where 2'b1000 means that the test is in progress,
> and setting *finished = false. I didn't introduced any delay, relying
> on the reschedule delay of the PHY state machine. I didn't notice any
> problems with this approach. Anyway if the test does not complete for
> whatever reasons we get stuck here, right ?

It should not happen, and if it does, its a bug in the PHY
firmware. And if that happens, the PHY is probably dead anyway.

phylib does allow you to break out of the cable test state by calling
phy_stop(). That generally happens when you admin down the interface.

So please do check the test has completed.

Andrew