Re: [PATCH net-next 00/15] net: dsa: mv88e6xxx: convert to phylink pcs

["Marek Behún" <kabel@kernel.org>]

On Mon, 13 Jun 2022 13:59:03 +0100
"Russell King (Oracle)" <linux@armlinux.org.uk> wrote:

> Hi,
> 
> This series converts mv88e6xxx to use phylink pcs, which I believe is
> the last DSA driver that needs to be converted before we can declare
> the whole of DSA as non-phylink legacy.
> 
> Briefly:
> Patches 1 and 2 introduce a new phylink_pcs_inband() helper to indicate
> whether inband AN should be used. Note that the first patch fixes a bug
> in the current c22 helper where the SGMII exchange with the PHY would
> be disabled when AN is turned off on the PHY copper side.
> 
> Patch 3 gets rid of phylink's internal pcs_ops member, preferring
> instead to always use the version in the phylink_pcs structure.
> Changing this pointer is now no longer supported.
> 
> Patch 4 makes PCS polling slightly cleaner, avoiding the poll being
> triggered while we're making changes to the configuration.
> 
> Patch 5 and 6 introduce several PCS methods that are fundamentally
> necessary for mv88e6xxx to work around various issues - for example, in
> some devices, the PCS must be powered down when the CMODE field in the
> port control register is changed. In other devices, there are
> workarounds that need to be performed.
> 
> Patch 7 adds unlocked mdiobus and mdiodev accessors to complement the
> locking versions that are already there - which are needed for some of
> the mv88e6xxx conversions.
> 
> Patch 8 prepares DSA as a whole, adding support for the phylink
> mac_prepare() and mac_finish() methods. These two methods are used to
> force the link down over a major reconfiguration event, which has been
> found by people to be necessary on mv88e6xxx devices. These haven't
> been required until now as everything has been done via the
> mac_config() callback - which won't be true once we switch to
> phylink_pcs.
> 
> Patch 9 implements patch 8 on this driver.
> 
> Patches 10 and 11 prepare mv88e6xxx for the conversion.
> 
> Patches 12 through to 14 convert each "serdes" to phylink_pcs.
> 
> Patch 15 cleans up after the conversion.
> 
>  drivers/net/dsa/mv88e6xxx/Makefile   |    3 +
>  drivers/net/dsa/mv88e6xxx/chip.c     |  480 ++++----------
>  drivers/net/dsa/mv88e6xxx/chip.h     |   25 +-
>  drivers/net/dsa/mv88e6xxx/pcs-6185.c |  158 +++++
>  drivers/net/dsa/mv88e6xxx/pcs-6352.c |  383 +++++++++++
>  drivers/net/dsa/mv88e6xxx/pcs-639x.c |  834 ++++++++++++++++++++++++
>  drivers/net/dsa/mv88e6xxx/port.c     |   30 -
>  drivers/net/dsa/mv88e6xxx/serdes.c   | 1164 ++--------------------------------
>  drivers/net/dsa/mv88e6xxx/serdes.h   |  110 +---
>  drivers/net/phy/mdio_bus.c           |   24 +-
>  drivers/net/phy/phylink.c            |  141 ++--
>  include/linux/mdio.h                 |   26 +
>  include/linux/phylink.h              |   44 ++
>  include/net/dsa.h                    |    6 +
>  net/dsa/port.c                       |   32 +
>  15 files changed, 1826 insertions(+), 1634 deletions(-)
> 

Tested on Turris MOX, no regressions discovered so far.

Tested-by: Marek Behún <kabel@kernel.org>

But patches 06/15 and 14/15 need testing on CN9130-CRB: the SFP cage
needs to be tested in 2500base-x mode, and also switching between
2500base-x, sgmii, 5gbase-r and 10gbase-r.

Pali, could you find some time for this? I can direct you about how to
do this.

Marek