Re: [linux-sunxi] [PATCH v3 10/13] spi: sunxi: merge sun4i and sun6i SPI driver

[Julian Calaby]

Hi Michal,

On Tue, Jun 14, 2016 at 3:28 PM, Michal Suchanek <hramrach@xxxxxxxxx> wrote:
> On 14 June 2016 at 06:47, Julian Calaby <julian.calaby@xxxxxxxxx> wrote:
>> Hi Michal,
>>
>> On Tue, Jun 14, 2016 at 2:34 PM, Michal Suchanek <hramrach@xxxxxxxxx> wrote:
>>> Hello,
>>>
>>> On 14 June 2016 at 01:43, Julian Calaby <julian.calaby@xxxxxxxxx> wrote:
>>>> Hi Michal,
>>>>
>>>> On Tue, Jun 14, 2016 at 3:46 AM, Michal Suchanek <hramrach@xxxxxxxxx> wrote:
>>>>> The drivers are very similar and share multiple flaws which needed
>>>>> separate fixes for both drivers.
>>>>>
>>>>> Signed-off-by: Michal Suchanek <hramrach@xxxxxxxxx>
>>>>> ---
>>>>>  drivers/spi/Kconfig     |   8 +-
>>>>>  drivers/spi/Makefile    |   1 -
>>>>>  drivers/spi/spi-sun4i.c | 156 +++++++++++--
>>>>>  drivers/spi/spi-sun6i.c | 598 ------------------------------------------------
>>>>>  4 files changed, 143 insertions(+), 620 deletions(-)
>>>>>  delete mode 100644 drivers/spi/spi-sun6i.c
>>>>>
>>>>> diff --git a/drivers/spi/spi-sun4i.c b/drivers/spi/spi-sun4i.c
>>>>> index 0b8e6c6..c76f8e4 100644
>>>>> --- a/drivers/spi/spi-sun4i.c
>>>>> +++ b/drivers/spi/spi-sun4i.c
>>>>> @@ -279,9 +321,14 @@ static int sunxi_spi_transfer_one(struct spi_master *master,
>>>>>         reg = sunxi_spi_read(sspi, SUNXI_TFR_CTL_REG);
>>>>>
>>>>>         /* Reset FIFOs */
>>>>> -       sunxi_spi_write(sspi, SUNXI_TFR_CTL_REG,
>>>>> -                       reg | sspi_bits(sspi, SUNXI_CTL_RF_RST) |
>>>>> -                       sspi_bits(sspi, SUNXI_CTL_TF_RST));
>>>>> +       if (sspi->type == SPI_SUN4I)
>>>>> +               sunxi_spi_write(sspi, SUNXI_TFR_CTL_REG,
>>>>> +                               reg | sspi_bits(sspi, SUNXI_CTL_RF_RST) |
>>>>> +                               sspi_bits(sspi, SUNXI_CTL_TF_RST));
>>>>> +       else
>>>>> +               sunxi_spi_write(sspi, SUNXI_FIFO_CTL_REG,
>>>>> +                               sspi_bits(sspi, SUNXI_CTL_RF_RST) |
>>>>> +                               sspi_bits(sspi, SUNXI_CTL_TF_RST));
>>>>
>>>> If we're already doing different stuff for each generation of the IP,
>>>> why not just use the register offsets and bit definitions directly?
>>>
>>> Because having (*sspi->regmap)[SUNXI_FIFO_CTL_REG] all over the place
>>> makes my eyes bleed and you cannot use the check that you are
>>> accessing a register that actually exists.
>>
>> I mean removing SUNXI_CTL_RF_RST and SUNXI_CTL_TF_RST from all of the
>> indirection you added and using them directly, i.e.
>>
>> #define SUN4I_TFR_CTL_RF_RST BIT(x)
>> #define SUN4I_TFR_CTL_TF_RST BIT(x)
>> #define SUN6I_FIFO_CTL_RF_RST BIT(x)
>> #define SUN6I_FIFO_CTL_TF_RST BIT(x)
>>
>> then
>>
>> if (sspi->type == SPI_SUN4I)
>>     sunxi_spi_write(sspi, SUNXI_TFR_CTL_REG, reg |
>> SUN4I_TFR_CTL_RF_RST | SUN4I_TFR_CTL_TF_RST);
>> else
>>     sunxi_spi_write(sspi, SUNXI_FIFO_CTL_REG, reg |
>> SUN6I_FIFO_CTL_RF_RST | SUN6I_FIFO_CTL_TF_RST);
>>
>> I.e. the bits that need setting are in different registers, so you
>> have to have an if statement and separate calls. Therefore there's no
>> real benefit from the indirection you've introduced here, unless
>> you're expecting the SUN8I variant to use different bits in one of
>> those two registers.
>>
>
> That looks nice for this particular case.

There was another case I pointed out in this driver that could
potentially benefit from this.

> Still you will have to remember that these bits are specified directly
> while other bits on the register are mapped which is not so nice.

True. I'm not sure which path is best in this case. To my eye, your
indirection scheme seems like the "heavy" solution, however I have no
idea what form a "lighter" solution would take.

Other drivers I've seen which have tackled similar problems have used
a large struct to hold all the variant-specific stuff, whether that's
function pointers, register offsets, constants, etc. (so this code
could theoretically be re-written as sunxi_spi_write(sspi,
sspi->fifo_reg, reg | sspi->fifo_reset_arg) or sspi->reset_fifo())
however the driver I saw doing this (rtl8xxxu) was handling much more
significant differences between device variants than you are.

Thanks,

-- 
Julian Calaby

Email: julian.calaby@xxxxxxxxx
Profile: http://www.google.com/profiles/julian.calaby/