Re: [PATCH v2 1/2 net-next] net: phy: mscc: add support for CLKOUT ctrl reg for VSC8531 and similar

[Alexandru Ardelean <alex@shruggie.ro>]

On Sat, Jul 15, 2023 at 1:27 AM Andrew Lunn <andrew@lunn.ch> wrote:
>
> On Fri, Jul 14, 2023 at 09:09:14AM +0300, Alexandru Ardelean wrote:
> > On Thu, Jul 13, 2023 at 11:35 PM Andrew Lunn <andrew@lunn.ch> wrote:
> > >
> > > > +set_reg:
> > > > +     mutex_lock(&phydev->lock);
> > > > +     rc = phy_modify_paged(phydev, MSCC_PHY_PAGE_EXTENDED_GPIO,
> > > > +                           VSC8531_CLKOUT_CNTL, mask, set);
> > > > +     mutex_unlock(&phydev->lock);
> > >
> > > What is this mutex protecting?
> >
> > This was inspired by vsc85xx_edge_rate_cntl_set().
> > Which has the same format.

Good news.
Removing this mutex works on a 5.10 kernel, with no issues.

>
> phy_modify_paged() locks the MDIO bus while it swaps the page, so
> nothing else can use it. That also protects the read/modify/write.
>
> Nothing is modifying phydev, so the lock is not needed for that
> either.

I remembered what I was doing wrong in that version that had issues
with the lock.
I was doing some manual page changes, with
phy_base_read/()phy_base_write() functions, which are in this file.

These functions have a warning + dump_stack() for when the
"phydev->mdio.bus->mdio_lock" is not held).
That threw me off initially.

>
> > I'll re-test with this lock removed.
> > I may be misremembering (or maybe I did something silly at some
> > point), but there was a weird stack-trace warning before adding this
> > lock there.
> > This was with a 5.10.116 kernel version.
>
> This patch is for net-next, please test there.

I've been testing on a Renesas board CIP project.
Kernel version (on our board is actually 5.10.83 ; I get them confused
since 5.10.xxx seems to be used here-n-there).

The kernel is here:
https://github.com/renesas-rz/rz_linux-cip/tree/rz-5.10-cip3

I'm trying to backport some ARCH patches, so that the board boots up.
I "think" I'm half way there; now the kernel prints something to
console and then stops (that's progress from no prints).

Let's see if we get a different consensus on Rob't suggestion; this
patch may require a different V3 :)



>
> When testing for locking issues, and when doing development in
> general, it is a good idea to turn on CONFIG_PROVE_LOCKING and
> CONFIG_DEBUG_ATOMIC_SLEEP.
>
>         Andrew