[PATCH 3/5] SPI S3C64XX: Header for passing platform data

[jassi brar]

On Mon, Dec 7, 2009 at 11:41 PM, Ben Dooks <ben-linux at fluff.org> wrote:
> On Fri, Dec 04, 2009 at 08:06:12PM +0900, jassi brar wrote:
>> On Fri, Dec 4, 2009 at 7:06 AM, Ben Dooks <ben-linux at fluff.org> wrote:
>> > On Thu, Nov 26, 2009 at 03:48:17PM +0900, Jassi Brar wrote:
>> >> We need a way to pass controller specific information to the
>> >> SPI device driver. For that purpose a new header is made.
>> >>
>> >> Signed-off-by: Jassi Brar <jassi.brar at samsung.com>
>> >> ---
>> >>  arch/arm/plat-s3c64xx/include/plat/spi.h |   68 ++++++++++++++++++++++++++++++
>> >>  1 files changed, 68 insertions(+), 0 deletions(-)
>> >>  create mode 100644 arch/arm/plat-s3c64xx/include/plat/spi.h
>> >>
>> >> diff --git a/arch/arm/plat-s3c64xx/include/plat/spi.h b/arch/arm/plat-s3c64xx/include/plat/spi.h
>> >> new file mode 100644
>> >> index 0000000..d65ddfd
>> >> --- /dev/null
>> >> +++ b/arch/arm/plat-s3c64xx/include/plat/spi.h
>> >
>> > let's not have all these called spi.h, it will make life more difficult
>> > when trying to find which spi.h we are searching for in our platform
>> > support.
>> We can call it s3c64xx-spi.h but won't that be kinda redundant as it's
>> in plat-s3c64xx ?
>
> If it ever gets moved, then there's your first problem case.
>
> The second, is that you look at the top of the driver and see <plat/spi.h>
> and then go 'find . -type f -name spi.h' and see how many results you get
> for that. Giving it a more descriptive name makes it easier to find the
> right header without having to work out what is being included.
Ok.
Also I think, for reuse with newer SoCs, we need to divide this into two parts

  First - SPI controller specific part, which is common to all newer SoCs.
           Maybe this could go into plat-samsung/include with the
           name s3c64xx-spi.h?

  Second - SPI source clocks specific, which SoCs differ by name and number.
          Maybe this could go into plat-<soc>/include with the name
          <soc>-spi.h

   Please share your opinion.

>> >> @@ -0,0 +1,68 @@
>> >> +/* linux/arch/arm/plat-s3c64xx/include/plat/spi.h
>> >> + *
>> >> + * Copyright (C) 2009 Samsung Electronics Ltd.
>> >> + *   Jaswinder Singh <jassi.brar at samsung.com>
>> >> + *
>> >> + * This program is free software; you can redistribute it and/or modify
>> >> + * it under the terms of the GNU General Public License version 2 as
>> >> + * published by the Free Software Foundation.
>> >> + */
>> >> +
>> >> +#ifndef __S3C64XX_PLAT_SPI_H
>> >> +#define __S3C64XX_PLAT_SPI_H __FILE__
>> >> +
>> >> +#define S3C64XX_SPI_SRCCLK_PCLK              0
>> >> +#define S3C64XX_SPI_SRCCLK_SPIBUS    1
>> >> +#define S3C64XX_SPI_SRCCLK_48M               2
>> >> +
>> >> +#define BUSNUM(b)            (b)
>> >> +
>> >> +/**
>> >> + * struct s3c64xx_spi_csinfo - ChipSelect description
>> >> + * @fb_delay: Slave specific feedback delay.
>> >> + * @set_level: CS line control.
>> >> + */
>> >> +struct s3c64xx_spi_csinfo {
>> >> +     u8 fb_delay;
>> >> +     void (*set_level)(int lvl);
>> >> +};
>> >
>> > I think set_level should be called 'set_cs' to make it clearer what is
>> > being done here.
>> Well, in the driver we instantiate the structure pointer as 'cs', so all
>> the calls look like "cs->set_level" so I think that should be ok,
>> as it's quite obvious its all about cs(ChipSelect).
>>
>> >> +/**
>> >> + * struct s3c64xx_spi_cntrlr_info - SPI Controller defining structure
>> >> + * @src_clk_nr: Clock source index for the CLK_CFG[SPI_CLKSEL] field.
>> >> + * @src_clk_name: Platform name of the corresponding clock.
>> >> + * @src_clk: Pointer to the source clock.
>> >> + * @num_cs: Number of CS this controller emulates.
>> >> + * @cs: Array describing each CS.
>> >> + * @cfg_gpio: Configure pins for this SPI controller.
>> >> + * @fifo_lvl_mask: All tx fifo_lvl fields start at offset-6
>> >> + * @rx_lvl_offset: Depends on tx fifo_lvl field and bus number
>> >> + * @high_speed: If the controller supports HIGH_SPEED_EN bit
>> >> + */
>> >> +struct s3c64xx_spi_cntrlr_info {
>> >
>> > how about not bothering with the _cntrlr_ here and just call it
>> > s3c64xx_spi_info instead?
>> Sure.
>>
>> >> +     int src_clk_nr;
>> >> +     char *src_clk_name;
>> >> +     struct clk *src_clk;
>> >
>> > do not pass 'struct clk *' in via platform data.
>> Since this is not initialized in platform code: just a pointer
>> made available to the driver. So, yes, this can be made a
>> member of s3c64xx_spi_driver_data rather.
>>
>> >> +     int num_cs;
>> >> +     struct s3c64xx_spi_csinfo *cs;
>> >> +
>> >> +     int (*cfg_gpio)(struct platform_device *pdev);
>> >> +
>> >> +     /* Following two fields are for future compatibility */
>> >> +     int fifo_lvl_mask;
>> >> +     int rx_lvl_offset;
>> >> +     int high_speed;
>> >> +};
>> >
>> > I was wondering if a single 'set_cs' callback here would be in order,
>> > given each spi device can already hold a chip-select number for use
>> > with such callbacks, so:
>> >
>> > void (*set_cs)(struct s3c64xx_spi_cntrlr_info *us, struct spi_device *sel, int to);
>> In that case the machine code wud have to map the chipselect number to
>> appropriate function/switch-case. Switch-case maybe ok, but calling some
>> function to toggle CS might result in bigger lags between CS and appearance
>> of clock on the bus.
>
> The point is that we should already have a pointer to the spi device
> being initialised, and this can have a machine-set field in it specifying
> the chipselect. If it is all gpio, then this simply could be the
> number of the gpio involved.
Just for clarification:- the current implementation does make use of
spi_device->controller_data which, as u suggest, is set by machine code
to the pointer to a structure(since there is another CS
specific parameter- fb_delay) s3c64xx_spi_csinfo
We must have a callback function member, rather than void*, because
not every CS maybe simple GPIO manipulation. The driver simply call the
pointer to a function implemented in machine code.
I am unable to decide how to implement what u suggest.
If u think u understand the situation well and still stand by ur idea, please
let me know, I will modify the driver as you will explain.

> I don't see that this is going to save a lot of code time, wheras it is
> adding to the complexity of the platform data.
 Yes, that shudn't be a concern.

>> >> +/**
>> >> + * s3c64xx_spi_set_info - SPI Controller configure callback by the board
>> >> + *                           initialization code.
>> >> + * @cntrlr: SPI controller number the configuration is for.
>> >> + * @src_clk_nr: Clock the SPI controller is to use to generate SPI clocks.
>> >> + * @cs: Pointer to the array of CS descriptions.
>> >> + * @num_cs: Number of elements in the 'cs' array.
>> >> + */
>> >> +extern void s3c64xx_spi_set_info(int cntrlr, int src_clk_nr, int num_cs);
>> >> +
>> >> +#endif /* __S3C64XX_PLAT_SPI_H */
>> >> --
>> >> 1.6.2.5