[DISCUSSION] spi multi chipselect support

[DISCUSSION] spi multi chipselect support

[Sebastian Reichel]

[-- Attachment #1: Type: text/plain, Size: 1140 bytes --]

Hi Mark,

Some SPI slave devices have more than one chip select [0]. Rob suggested
to use the following DT notation for this [1]:

spi-controller {
    slave@0 {
        reg = <0>, <1>;
    };
};

I think this makes sense from a DT point of view, but this needs
quite some changes in the SPI core. I tried to find out, how to
implement the suggestion in the last few days and I see two
possibilities:

1. Register two SPI Linux devices from one DT node. This needs the
   fewest changes in SPI core. But we loose the one-to-one mapping
   of DT nodes to devices. I have a feeling, that this will backfire.

2. Add support for multiple chip-select to spi_device and modify
   the transfer functions to make use of this. Unfortunately this
   would require changing all spi controller as far as I can see?

Can you please advise me how to proceed?

Thanks,

-- Sebastian

[0] For example microcontrollers with ISP (in-system programming)
    interface, like the NXP Kinetis K20 series or the MRF89XA,
    which has one chip-select for configuration register access
    and one for FIFO access
[1] https://patchwork.kernel.org/patch/10310109/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [DISCUSSION] spi multi chipselect support

[Lars-Peter Clausen]

On 07/18/2018 06:35 PM, Sebastian Reichel wrote:
> Hi Mark,
> 
> Some SPI slave devices have more than one chip select [0]. Rob suggested
> to use the following DT notation for this [1]:
> 
> spi-controller {
>     slave@0 {
>         reg = <0>, <1>;
>     };
> };
> 
> I think this makes sense from a DT point of view, but this needs
> quite some changes in the SPI core. I tried to find out, how to
> implement the suggestion in the last few days and I see two
> possibilities:
> 
> 1. Register two SPI Linux devices from one DT node. This needs the
>    fewest changes in SPI core. But we loose the one-to-one mapping
>    of DT nodes to devices. I have a feeling, that this will backfire.
> 
> 2. Add support for multiple chip-select to spi_device and modify
>    the transfer functions to make use of this. Unfortunately this
>    would require changing all spi controller as far as I can see?
> 

Hi,

I believe that has been discussed before recently, but I can't find it in
the archives.

My preferred solution would to have something like
i2c_new_secondary_device(), but for SPI. This way the driver that binds to
the compatible string can allocate devices for additional chip selects as
necessary. Also make use of the reg-names property the same way as in I2C.

The only catch here is that locking in the SPI core needs to be re-worked,
since you can't allocate SPI devices from within another SPI device's probe
function. Something like
https://github.com/analogdevicesinc/linux/commit/b5cc8460b959e530413ffbf9b93d8012c80d05df

- Lars

Re: [DISCUSSION] spi multi chipselect support

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 477 bytes --]

On Wed, Jul 18, 2018 at 07:19:19PM +0200, Lars-Peter Clausen wrote:

> My preferred solution would to have something like
> i2c_new_secondary_device(), but for SPI. This way the driver that binds to
> the compatible string can allocate devices for additional chip selects as
> necessary. Also make use of the reg-names property the same way as in I2C.

Yes, that's at least the most consistent way forwards if nothing else.
I don't really have any more appealing ideas anyway.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]