Re: [PATCH net-next] net: pcs: tse: port to pcs-lynx

[Sean Anderson]

On 2/10/23 15:02, Andrew Lunn wrote:
> On Fri, Feb 10, 2023 at 09:31:59PM +0200, Vladimir Oltean wrote:
>> On Fri, Feb 10, 2023 at 08:09:49PM +0100, Maxime Chevallier wrote:
>> > When submitting the initial driver for the Altera TSE PCS, Russell King
>> > noted that the register layout for the TSE PCS is very similar to the
>> > Lynx PCS. The main difference being that TSE PCS's register space is
>> > memory-mapped, whereas Lynx's is exposed over MDIO.
>> > 
>> > Convert the TSE PCS to reuse the whole logic from Lynx, by allowing
>> > the creation of a dummy MDIO bus, and a dummy MDIO device located at
>> > address 0 on that bus. The MAC driver that uses this PCS must provide
>> > callbacks to read/write the MMIO.
>> > 
>> > Also convert the Altera TSE MAC driver to this new way of using the TSE
>> > PCS.
>> > 
>> > Signed-off-by: Maxime Chevallier <maxime.chevallier@xxxxxxxxxxx>
>> > ---
>> >  drivers/net/ethernet/altera/altera_tse.h      |   2 +-
>> >  drivers/net/ethernet/altera/altera_tse_main.c |  50 ++++-
>> >  drivers/net/pcs/Kconfig                       |   4 +
>> >  drivers/net/pcs/pcs-altera-tse.c              | 194 +++++++-----------
>> >  include/linux/pcs-altera-tse.h                |  22 +-
>> >  5 files changed, 142 insertions(+), 130 deletions(-)
>> 
>> The glue layer is larger than the duplicated PCS code? :(
> 
> I was wondering if the glue could actually be made generic. The kernel
> has a number of reasonably generic MMIO device drivers, which are just
> given an address range and assume a logical mapping.
> 
> Could this be made into a generic MDIO MMIO bus driver, which just
> gets configured with a base address, and maybe a stride between
> registers?

That's what I originally considered for MACB [1], but I ended up creating
the various encode/decode functions because the PCS didn't have an MDIO
counterpart [2].

--Sean

[1] https://lore.kernel.org/netdev/20211004191527.1610759-11-sean.anderson@xxxxxxxx/
    This patch doesn't have a mmap-mdio, as I dropped it before submission.
[2] https://lore.kernel.org/netdev/20211004191527.1610759-8-sean.anderson@xxxxxxxx/