pinctrl: mcp23s08: work around GPIO line naming

This driver is a bit weird because it can hide several gpio_chip instances underneath a single SPI slave. One cannot put the gpio-line-names DT stanza directly to the SPI slave when the spi-present-mask has more than one bit set.

I'm making up the gpio-bank DT child name as well as its address property. We need something to match the DT entries with the SPI address.

Signed-off-by Jan Kundrát <[email protected]>

Reviewed-by Phil Reid <[email protected]>

Cc Linus Walleij <[email protected]>

---
v2: Depend on exported devprop_gpiochip_set_names

---
 .../bindings/pinctrl/pinctrl-mcp23s08.txt     | 35 +++++++++++++++++++
 drivers/pinctrl/pinctrl-mcp23s08.c            | 11 ++++++
 2 files changed, 46 insertions(+)

Documentation/devicetree/bindings/pinctrl/pinctrl-mcp23s08.txt +7 -0

index 625a22e2f211..a0b1eb9aedad 100644
@@ -144,3 +144,38 @@ gpio21: gpio@21 {
 		bias-pull-up;
 	};
 };
+
+Line naming
+===========
+
+Because several gpio_chip instances are hidden below a single device tree
+node, it is necessary to split the names into several child nodes. Ensure
+that the configured addresses match those in the 

microchip,spi-present-mask:

Documentation/devicetree/bindings/pinctrl/pinctrl-mcp23s08.txt +28 -0

+
+gpio@0 {
+	compatible = "microchip,mcp23s17";
+	gpio-controller;
+	#gpio-cells = <2>;
+	/* this bitmask has bits #0 (0x01) and #2 (0x04) set */
+	spi-present-mask = <0x05>;
+	reg = <0>;
+	spi-max-frequency = <1000000>;
+
+	gpio-bank@1 {
+		address = <0>;
+		gpio-line-names =
+			"GPA0",
+			"GPA1",
+			...
+			"GPA7",
+			"GPB0",
+			"GPB1",
+			...
+			"GPB7";
+	};
+
+	gpio-bank@2 {
+		address = <2>;
+		gpio-line-names = ...
+	};
+};

drivers/pinctrl/pinctrl-mcp23s08.c +11 -0

index fd9d6f026d70..7810f56f8dd1 100644
@@ -1102,6 +1102,7 @@ static int mcp23s08_probe(struct spi_device *spi)
 	int				status, type;
 	unsigned			ngpio = 0;
 	const struct			of_device_id *match;
+	struct device_node		*np;
 
 	match = of_match_device(of_match_ptr(mcp23s08_spi_of_match), &spi->dev);
 	if (match)

@@ -1170,6 +1171,16 @@ static int mcp23s08_probe(struct spi_device *spi)
 		if (pdata->base != -1)
 			pdata->base += data->mcp[addr]->chip.ngpio;
 		ngpio += data->mcp[addr]->chip.ngpio;
+
+		for_each_available_child_of_node(spi->dev.of_node, np) {
+			u32 chip_addr;
+			status = of_property_read_u32(np, "address", &chip_addr);
+			if (status)
+				continue;
+			if (chip_addr != addr)
+				continue;
+			devprop_gpiochip_set_names(&data->mcp[addr]->chip, 
 of_fwnode_handle(np));
+		}
 	}
 	data->ngpio = ngpio;

Hello Jan,

On Thu, 7 Mar 2019 14:31:02 +0100
Jan Kundrát <[email protected]> wrote:

Review

Quoted 4 lines

(quoted diff) +4 -0

+Line naming
+===========
+
+Because several gpio_chip instances are hidden below a single device tree

Mentioning a Linux specific data structure (gpio_chip here) in a Device Tree binding documentation does not seem like a good idea. Bindings are supposed to be OS-agnostic.