From mboxrd@z Thu Jan  1 00:00:00 1970
From: martinwguy@gmail.com (Martin Guy)
Date: Sat, 17 Apr 2010 19:43:23 +0100
Subject: [spi-devel-general] [PATCH v3 0/2] spi: driver for Cirrus EP93xx
	SPI controller
In-Reply-To: <20100417045913.GA19534@gw.healthdatacare.com>
References: <cover.1271156934.git.mika.westerberg@iki.fi>
	<0D753D10438DA54287A00B0270842697636D710785@AUSP01VMBX24.collaborationhost.net>
	<20100417045913.GA19534@gw.healthdatacare.com>
Message-ID: <s2i56d259a01004171143m74b06d85yd1cbb38ef7ce778e@mail.gmail.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

(sorry for the duplicate, Mika, I forgot to include the list)

On 4/17/10, Mika Westerberg <mika.westerberg@iki.fi> wrote:
>  However, I'm not sure what you mean by limited only to built-in GPIOs? Currently
>  the driver does:
>
>         info->cs_control(spi->chip_select, value, info->data);
>
>  when it wants to assert/deassert the chip select.

There's a further strangeness when a board has a single SPI device not
using any GPIO chip select. At present, clients seem to need to
declare that they have 1 chip select and to provide an empty
cs_control function that does nothing. For example:

static void null_cs_control(unsigned cs, unsigned value, void *data) {
       /* Sim.One only has MMC card and no GPIO chip select logic */
}

static struct ep93xx_spi_info simone_spi_info = {
       .num_chipselect = 1,
       .cs_control     = null_cs_control,
       .data           = NULL,
};
...
static void __init simone_init_machine(void)
{
...
       spi_register_board_info(simone_spi_board_info,
ARRAY_SIZE(simone_spi_board_info));
       ep93xx_register_spi(&simone_spi_info);
...

Allowing cs_control to be NULL would do away with the empty function, but
when I just tried with num_chipselect = 0 the device did not appear. There
may be a better solution.

   M