[PATCH 1/4 v5] of_spi: add generic binding support to specify cs gpio

On Sun, 11 Mar 2012 19:25:37 +0100, Jean-Christophe PLAGNIOL-VILLARD [off-list ref] wrote:

On 08:07 Sun 11 Mar     , Grant Likely wrote:

quoted

On Fri,  9 Mar 2012 19:25:44 +0100, Jean-Christophe PLAGNIOL-VILLARD [off-list ref] wrote:

quoted

This will allow to use gpio for chip select with no modification in the
driver binding

When use the cs-gpios, the gpio number will be passed via the cs_gpio field
and the number of chip select will automatically increased.

Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <redacted>
Cc: devicetree-discuss at lists.ozlabs.org
Cc: spi-devel-general at lists.sourceforge.net
---
v5:
	update grant comment to simplify the cs_gpio management

Best Regards,
J.
 Documentation/devicetree/bindings/spi/spi-bus.txt |    6 ++
 drivers/spi/spi.c                                 |   55 +++++++++++++++++++-
 include/linux/spi/spi.h                           |    4 ++
 3 files changed, 62 insertions(+), 3 deletions(-)

diff --git a/Documentation/devicetree/bindings/spi/spi-bus.txt b/Documentation/devicetree/bindings/spi/spi-bus.txt
index e782add..c253379 100644
--- a/Documentation/devicetree/bindings/spi/spi-bus.txt
+++ b/Documentation/devicetree/bindings/spi/spi-bus.txt

@@ -12,6 +12,7 @@ The SPI master node requires the following properties:
 - #size-cells     - should be zero.
 - compatible      - name of SPI bus controller following generic names
     		recommended practice.
+- cs-gpios	  - (optional) gpios chip select.
 No other properties are required in the SPI bus node.  It is assumed
 that a driver for an SPI bus device will understand that it is an SPI bus.
 However, the binding does not attempt to define the specific method for

@@ -21,6 +22,8 @@ assumption that board specific platform code will be used to manage
 chip selects.  Individual drivers can define additional properties to
 support describing the chip select layout.
 
+If cs-gpios is used the number of chip select will automatically increased.
+
 SPI slave nodes must be children of the SPI master node and can
 contain the following properties.
 - reg             - (required) chip select address of device.

@@ -34,6 +37,9 @@ contain the following properties.
 - spi-cs-high     - (optional) Empty property indicating device requires
     		chip select active high
 
+If a gpio chipselect is used for the SPI slave the gpio number will be passed
+via the cs_gpio
+
 SPI example for an MPC5200 SPI bus:
 	spi at f00 {
 		#address-cells = <1>;

diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c
index b2ccdea..446eee5 100644
--- a/drivers/spi/spi.c
+++ b/drivers/spi/spi.c

@@ -28,6 +28,7 @@
 #include <linux/mod_devicetable.h>
 #include <linux/spi/spi.h>
 #include <linux/of_spi.h>
+#include <linux/of_gpio.h>
 #include <linux/pm_runtime.h>
 #include <linux/export.h>
 

@@ -322,6 +323,7 @@ struct spi_device *spi_alloc_device(struct spi_master *master)
 	spi->dev.parent = &master->dev;
 	spi->dev.bus = &spi_bus_type;
 	spi->dev.release = spidev_release;
+	spi->cs_gpio = -EINVAL;
 	device_initialize(&spi->dev);
 	return spi;
 }

@@ -339,15 +341,16 @@ EXPORT_SYMBOL_GPL(spi_alloc_device);
 int spi_add_device(struct spi_device *spi)
 {
 	static DEFINE_MUTEX(spi_add_lock);
-	struct device *dev = spi->master->dev.parent;
+	struct spi_master *master = spi->master;
+	struct device *dev = master->dev.parent;
 	struct device *d;
 	int status;
 
 	/* Chipselects are numbered 0..max; validate. */
-	if (spi->chip_select >= spi->master->num_chipselect) {
+	if (spi->chip_select >= master->num_chipselect) {
 		dev_err(dev, "cs%d >= max %d\n",
 			spi->chip_select,
-			spi->master->num_chipselect);
+			master->num_chipselect);
 		return -EINVAL;
 	}
 

@@ -371,6 +374,9 @@ int spi_add_device(struct spi_device *spi)
 		goto done;
 	}
 
+	if (master->cs_gpios)
+		spi->cs_gpio = master->cs_gpios[spi->chip_select];
+

I still don't understand the need for this.  Only the master driver will need
this data, and it is already stored in the master->cs_gpios array.  Why
does it need to get duplicated into the spi_device?

yeah can drop it

quoted

quoted

 	/* Drivers may modify this initial i/o setup, but will
 	 * normally rely on the device being setup.  Devices
 	 * using SPI_CS_HIGH can't coexist well otherwise...

@@ -561,6 +567,45 @@ struct spi_master *spi_alloc_master(struct device *dev, unsigned size)
 }
 EXPORT_SYMBOL_GPL(spi_alloc_master);
 
+#ifdef CONFIG_OF
+static int of_spi_register_master(struct spi_master *master)
+{
+	int nb, i;
+	int *cs;
+	struct device_node *np = master->dev.of_node;
+
+	if (!np)
+		return 0;
+
+	nb = of_gpio_named_count(np, "cs-gpios");
+
+	if (nb < 1)
+		return 0;
+
+	cs = devm_kzalloc(&master->dev,
+			  sizeof(int) * (master->num_chipselect + nb),
+			  GFP_KERNEL);

sizeof(*cs)

quoted

+	master->cs_gpios = cs;
+
+	if (!master->cs_gpios)
+		return -ENOMEM;
+
+	memset(cs, -EINVAL, master->num_chipselect);
+	master->num_chipselect += nb;

Looks buggy.  Should be:
	if (nb > master->num_chipselect)
		master->num_chipselect = nb;

However, this is actually quite risky.  Changing the size of
num_chipselect without the driver knowing about it will most likely
cause breakage in drivers.  Drivers may use it as an array index, and
if it gets increased then there will be problems.  In fact, drivers
are probably using spi->chip_select as an array or register index so I
don't think it can be transparently extended to include gpios.

So, this function must change.  It cannot be called automatically
from spi_register_master().  It needs to be a library function that spi bus
drivers must call explicitly so they are prepared for a change in the
num_chipselect value.

I'll prefer to let the spi framework manage it

I understand your preference, but it must not carry the risk of breakage.

spi devices cannot be added if the cs number is larger than num_chipselect;
even if those devices are backed by gpios.  The controller driver still
needs to understand they are there and modify it's own num_chipselects
value.  Otherwise transfer requests will pass a CS number larger than
the bus drivers know what to do with.

quoted

Also, if we're doing this, then the gpio cs manipulation hook should be
moved out of spi_bitbang and into the core spi_master.  There should
also be a stock function for manipulating gpio chip selects. I don't want
spi driver authors under the impression that they should open code the
chip select manipulation.

yeah I was thinking about it too

how about we split it in two

num_hw_cs manage by the drivers

and

num_gpio_cs for cs gpio

and num_chipselect (hw + cs) will for internal use

the gpio cs will be after the hw cs.

As already discussed, don't split the number space.  Keep gpio cs numbers
based at zero and let them overlap the hw cs numbers.

and we move the gpio cs from spi_bitbang to spi_master

Yes, please move the gpio cs handling.

g.