Re: [PATCH v2 1/2] pinctrl: sh-pfc: Add r8a77470 PFC support

[Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>]

Hi Sergei,

On Sat, Apr 28, 2018 at 8:30 PM, Sergei Shtylyov
<sergei.shtylyov@xxxxxxxxxxxxxxxxxx> wrote:
> On 04/28/2018 08:40 PM, Sergei Shtylyov wrote:
>>>>> Subject: Re: [PATCH v2 1/2] pinctrl: sh-pfc: Add r8a77470 PFC support
>>>>> On Wed, Apr 4, 2018 at 5:22 PM, Biju Das <biju.das@xxxxxxxxxxxxxx> wrote:
>>>>>> Add PFC support for the R8A77470 SoC including pin groups for some
>>>>>> on-chip devices such as SCIF, AVB and MMC.
>>>>>>
>>>>>> Signed-off-by: Biju Das <biju.das@xxxxxxxxxxxxxx>
>>>>>> Reviewed-by: Fabrizio Castro <fabrizio.castro@xxxxxxxxxxxxxx>
>>>
>>>>>> --- /dev/null
>>>>>> +++ b/drivers/pinctrl/sh-pfc/pfc-r8a77470.c
>>>
>>>>>> +/* - AVB
>>>>>> +-------------------------------------------------------------------- */ static const
>>>>> unsigned int avb_link_pins[] = {
>>>>>> +       RCAR_GP_PIN(5, 14),
>>>>>> +};
>>>>>> +static const unsigned int avb_link_mux[] = {
>>>>>> +       AVB_LINK_MARK,
>>>>>> +};
>>>>>> +static const unsigned int avb_magic_pins[] = {
>>>>>> +       RCAR_GP_PIN(5, 15),
>>>>>> +};
>>>>>> +static const unsigned int avb_magic_mux[] = {
>>>>>> +       AVB_MAGIC_MARK,
>>>>>> +};
>>>>>> +static const unsigned int avb_phy_int_pins[] = {
>>>>>> +       RCAR_GP_PIN(5, 16),
>>>>>> +};
>>>>>> +static const unsigned int avb_phy_int_mux[] = {
>>>>>> +       AVB_PHY_INT_MARK,
>>>>>> +};
>>>>>> +static const unsigned int avb_mdio_pins[] = {
>>>>>> +       RCAR_GP_PIN(5, 12), RCAR_GP_PIN(5, 13), }; static const
>>>>>> +unsigned int avb_mdio_mux[] = {
>>>>>> +       AVB_MDC_MARK, AVB_MDIO_MARK,
>>>>>> +};
>>>>>> +static const unsigned int avb_mii_pins[] = {
>>>>>> +       RCAR_GP_PIN(3, 14), RCAR_GP_PIN(3, 15), RCAR_GP_PIN(3, 16),
>>>>>> +       RCAR_GP_PIN(3, 27),
>>>>>> +
>>>>>> +       RCAR_GP_PIN(3, 2), RCAR_GP_PIN(3, 3), RCAR_GP_PIN(3, 4),
>>>>>> +       RCAR_GP_PIN(3, 5),
>>>>>> +
>>>>>> +       RCAR_GP_PIN(3, 10), RCAR_GP_PIN(3, 0), RCAR_GP_PIN(3, 1),
>>>>>> +       RCAR_GP_PIN(5, 17), RCAR_GP_PIN(3, 13), RCAR_GP_PIN(5, 23),
>>>>>> +       RCAR_GP_PIN(3, 12),
>>>>>> +};
>>>>>> +static const unsigned int avb_mii_mux[] = {
>>>>>> +       AVB_TXD0_MARK, AVB_TXD1_MARK, AVB_TXD2_MARK,
>>>>>> +       AVB_TXD3_MARK,
>>>>>> +
>>>>>> +       AVB_RXD0_MARK, AVB_RXD1_MARK, AVB_RXD2_MARK,
>>>>>> +       AVB_RXD3_MARK,
>>>>>> +
>>>>>> +       AVB_RX_ER_MARK, AVB_RX_CLK_MARK, AVB_RX_DV_MARK,
>>>>>> +       AVB_CRS_MARK, AVB_TX_EN_MARK, AVB_TX_ER_MARK,
>>>>>> +       AVB_TX_CLK_MARK,
>>>>>
>>>>> You forgot AVB_COL, which is GP5_18?
>>>>
>>>> GP5_18 is shred between the signals AVB_COL and VI0_CLK. As per the RZ/G1C board schematic /hardware user guide
>>>> Ethernet Phy collision pin(AVB_COL) pin  is not populated on this board by default. It is populated only for
>>>> Video Input Channel0 pixel clock.
>>>>
>>>> Since it is initial submission, I thought of adding only board specific pins first and later
>>>> add this collision pin alone as a separate pin entry  in the avb group. That way we can support existing board
>>>> as well as any future RZ/G1C board which populates this pin. Are you ok for this approach?
>>
>>> Oops. That means our grouping is suboptimal. Perhaps we should revisit
>>> (for all SoCs, while keeping backwards compatibility)?
>>
>>
>>> After reading some wikipedia, I came up with:
>>>
>>>     mii_tx
>>>     mii_tx_er (optional)
>>>     mii_rx
>>
>>    I don't see why we should have the separate TX/RX subgroups just because they're
>> described separately in the Wikipedia. Or did you have some special reason to do it?

OK, so for MII:

    mii_tx_rx
    mii_tx_er

and for GMII:

    gmii_tx_rx

>>>     mii_col_crs (optional, half duplex)
>>>
>>> However, given your board uses AVB_CRS without AVB_COL, that would still
>>> not be sufficient, so the last group should be split to cover your use case?
>>
>>> BTW, how does it work with AVB_COL not wired to anything by default, and thus
>>> floating? Do you enable pull-up/down for that pin in the PFC, or is the pin
>>> just ignored in full-duplex mode?
>>
>>    Well, I'm seeing the strong possibility the half-duplex mode just doesn't work as
>> the collision indication isn't available.
>
>    And indeed, the manuals tell me that RZ/G (and R-Car gen2) EtherAVB only supports
> the full-duplex mode.

Indeed. Also on R-Car Gen3.

Then, what's the purpose of the CRS and COL signals in full-duplex mode?
Apart from being shown in the timing diagrams (Figures 33A.72 and later in
the RZ/G1C Hardware User Manual Rev. 1.00), there's no further information
about them.

Gigabit Ethernet has carrier extension, perhaps the CRS signal is needed
to detect invalid carrier extension? That means it could be used without COL,
and thus we may need separate pin groups for CRS and COL?

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html