Skip to contentSkip to conversation

gpio: brcmstb: add support for gpio-ranges

PATCH Applied Doug Berger 5 participants 17 messages Apr 24, 2024 – May 6, 2024

Subsystems (4)
lore
Share & tools
lore

Doug Berger

author maintainer
#0 lore raw

cover gpio: brcmstb: add support for gpio-ranges

The Raspberry Pi 5 includes Broadcom STB GPIO IP as well as Broadcom 2712 pin controller IP.

The community has expressed interest in linking the two drivers with the "gpio-ranges" property in device tree. This commit stack implements the necessary changes.

Doug Berger (3):
dt-bindings: gpio: brcmstb: add gpio-ranges
gpio: of: support gpio-ranges for multiple gpiochip devices
gpio: brcmstb: add support for gpio-ranges

 .../bindings/gpio/brcm,brcmstb-gpio.yaml      |  3 +++
 drivers/gpio/gpio-brcmstb.c                   |  2 ++
 drivers/gpio/gpiolib-of.c                     | 23 +++++++++++++++++--
 3 files changed, 26 insertions(+), 2 deletions(-)
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH 1/3 dt-bindings: gpio: brcmstb: add gpio-ranges

Add optional gpio-ranges device-tree property to the Broadcom Set-Top-Box GPIO controller.

---
 Documentation/devicetree/bindings/gpio/brcm,brcmstb-gpio.yaml | 3 +++
 1 file changed, 3 insertions(+)
Documentation/devicetree/bindings/gpio/brcm,brcmstb-gpio.yaml +3 -0
index a1e71c974e79..f096f286da19 100644
@@ -62,6 +62,8 @@ properties:
 
   interrupt-controller: true
 
+  gpio-ranges: true
+
   wakeup-source:
     type: boolean
     description: >
@@ -88,6 +90,7 @@ examples:
         interrupt-parent = <&irq0_intc>;
         interrupts = <0x6>;
         brcm,gpio-bank-widths = <32 32 32 24>;
+        gpio-ranges = <&pinctrl 0 0 120>;
     };
 
     upg_gio_aon: gpio@f04172c0 {
Signature · 1 line
2.34.1
author maintainer
#5 lore raw

PATCH 2/3 gpio: of: support gpio-ranges for multiple gpiochip devices

In reply to Doug Berger

Some drivers (e.g. gpio-mt7621 and gpio-brcmstb) have multiple gpiochip banks within a single device. Unfortunately, the gpio-ranges property of the device node was being applied to every gpiochip of the device with device relative GPIO offset values rather than gpiochip relative GPIO offset values.

This commit makes use of the gpio_chip offset value which can be non-zero for such devices to split the device node gpio-ranges property into GPIO offset ranges that can be applied to each of the relevant gpiochips of the device.

---
 drivers/gpio/gpiolib-of.c | 23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)
drivers/gpio/gpiolib-of.c +21 -2
index cb0cefaec37e..d75f6ee37028 100644
@@ -1037,7 +1037,7 @@ static int of_gpiochip_add_pin_range(struct gpio_chip *chip)
 	struct of_phandle_args pinspec;
 	struct pinctrl_dev *pctldev;
 	struct device_node *np;
-	int index = 0, ret;
+	int index = 0, ret, trim;
 	const char *name;
 	static const char group_names_propname[] = "gpio-ranges-group-names";
 	struct property *group_names;
@@ -1059,7 +1059,14 @@ static int of_gpiochip_add_pin_range(struct gpio_chip *chip)
 		if (!pctldev)
 			return -EPROBE_DEFER;
 
+		/* Ignore ranges outside of this GPIO chip */
+		if (pinspec.args[0] >= (chip->offset + chip->ngpio))
+			continue;
+		if (pinspec.args[0] + pinspec.args[2] <= chip->offset)
+			continue;
+
 		if (pinspec.args[2]) {
+			/* npins != 0: linear range */
 			if (group_names) {
 				of_property_read_string_index(np,
 						group_names_propname,
@@ -1070,7 +1077,19 @@ static int of_gpiochip_add_pin_range(struct gpio_chip *chip)
 					break;
 				}
 			}
-			/* npins != 0: linear range */
+
+			/* Trim the range to fit this GPIO chip */
+			if (chip->offset > pinspec.args[0]) {
+				trim = chip->offset - pinspec.args[0];
+				pinspec.args[2] -= trim;
+				pinspec.args[1] += trim;
+				pinspec.args[0] = 0;
+			} else {
+				pinspec.args[0] -= chip->offset;
+			}
+			if ((pinspec.args[0] + pinspec.args[2]) > chip->ngpio)
+				pinspec.args[2] = chip->ngpio - pinspec.args[0];
+
 			ret = gpiochip_add_pin_range(chip,
 					pinctrl_dev_get_devname(pctldev),
 					pinspec.args[0],
Signature · 1 line
2.34.1
maintainer Acked-by on v1 2/3
#6 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

Quoted 6 lines · Doug Berger

Some drivers (e.g. gpio-mt7621 and gpio-brcmstb) have multiple gpiochip banks within a single device. Unfortunately, the gpio-ranges property of the device node was being applied to every gpiochip of the device with device relative GPIO offset values rather than gpiochip relative GPIO offset values.

Quoted 6 lines

This commit makes use of the gpio_chip offset value which can be non-zero for such devices to split the device node gpio-ranges property into GPIO offset ranges that can be applied to each of the relevant gpiochips of the device.

Signature · 1 line
Florian
#7 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

In reply to Doug Berger
Quoted 7 lines · Doug Berger

Some drivers (e.g. gpio-mt7621 and gpio-brcmstb) have multiple gpiochip banks within a single device. Unfortunately, the gpio-ranges property of the device node was being applied to every gpiochip of the device with device relative GPIO offset values rather than gpiochip relative GPIO offset values.

Quoted 7 lines

This commit makes use of the gpio_chip offset value which can be non-zero for such devices to split the device node gpio-ranges property into GPIO offset ranges that can be applied to each of the relevant gpiochips of the device.

---

This is a good improvement, thanks!

Bart

#8 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

In reply to Doug Berger

Hi Dough,

thanks for your patch!

I'm a bit confused here:

drivers/gpio/gpiolib-of.c static int of_gpiochip_add_pin_range(struct gpio_chip *chip) · line 1066

Quoted 5 lines · Doug Berger
drivers/gpio/gpiolib-of.c +5 -0
+               /* Ignore ranges outside of this GPIO chip */
+               if (pinspec.args[0] >= (chip->offset + chip->ngpio))
+                       continue;
+               if (pinspec.args[0] + pinspec.args[2] <= chip->offset)
+                       continue;

Here pinspec.args[0] and [2] comes directly from the device tree.

The documentation in Documentation/devicetree/bindings/gpio/gpio.txt says:

Quoted 8 lines

2.2) Ordinary (numerical) GPIO ranges
-------------------------------------

It is useful to represent which GPIOs correspond to which pins on which pin controllers. The gpio-ranges property described below represents this with a discrete set of ranges mapping pins from the pin controller local number space to pins in the GPIO controller local number space.

Quoted 5 lines

The format is: <[pin controller phandle], [GPIO controller offset],

                [pin controller offset], [number of pins]>;

The GPIO controller offset pertains to the GPIO controller node containing the range definition.

So I do not understand how pinspec[0] and [2] can ever be compared to something involving chip->offset which is a Linux-specific offset.

It rather looks like you are trying to accomodate the Linux numberspace in the ranges, which it was explicitly designed to avoid.

I just don't get it.

So NACK until I understand what is going on here.

Yours,
Linus Walleij

author maintainer
#9 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

Quoted 3 lines · Linus Walleij

Hi Dough,

thanks for your patch!

Thanks for your review!

Quoted 2 lines

I'm a bit confused here:

"Communication is hard" and I may be confused about your confusion, but hopefully we can work it out.

Quoted 10 lines · Doug Berger

drivers/gpio/gpiolib-of.c static int of_gpiochip_add_pin_range(struct gpio_chip *chip) · line 1066

Quoted 7 lines
drivers/gpio/gpiolib-of.c +5 -0
+               /* Ignore ranges outside of this GPIO chip */
+               if (pinspec.args[0] >= (chip->offset + chip->ngpio))
+                       continue;
+               if (pinspec.args[0] + pinspec.args[2] <= chip->offset)
+                       continue;
Quoted 18 lines

Here pinspec.args[0] and [2] comes directly from the device tree.

The documentation in Documentation/devicetree/bindings/gpio/gpio.txt says:

Quoted 13 lines

2.2) Ordinary (numerical) GPIO ranges
-------------------------------------

It is useful to represent which GPIOs correspond to which pins on which pin controllers. The gpio-ranges property described below represents this with a discrete set of ranges mapping pins from the pin controller local number space to pins in the GPIO controller local number space.

The format is: <[pin controller phandle], [GPIO controller offset],

                 [pin controller offset], [number of pins]>;

The GPIO controller offset pertains to the GPIO controller node containing the range definition.

I think we are in agreement here. For extra clarity, I will add that in my understanding pinspec.args[0] corresponds to [GPIO controller offset] and pinspec.args[2] corresponds to [number of pins].

Quoted 6 lines

So I do not understand how pinspec[0] and [2] can ever be compared to something involving chip->offset which is a Linux-specific offset.

It rather looks like you are trying to accomodate the Linux numberspace in the ranges, which it was explicitly designed to avoid.

The struct gpio_chip documentation in include/linux/gpio/driver.h says:

Quoted 3 lines

* @offset: when multiple gpio chips belong to the same device this
* can be used as offset within the device so friendly names can
* be properly assigned.

It is my understanding that this value represents the offset of a gpiochip relative to the GPIO controller device defined by the GPIO controller node in device tree. This puts it in the same number space as [GPIO controller offset]. I believe it was introduced for the specific purpose of translating [GPIO controller offset] values into Linux-specific offsets, which is why it is being reused for that purpose in this patch.

For GPIO Controllers that contain a single gpiochip the 'offset' member is 0 and the device tree node offsets can be applied directly to the gpiochip. However, when a GPIO Controller contains multiple gpiochips, the device tree node offsets must be translated to each individual gpiochip.

Quoted 7 lines

I just don't get it.

So NACK until I understand what is going on here.

Yours,
Linus Walleij

I hope it makes sense now, but if not please help me understand what I may be missing.

Thanks,
Doug

#10 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

Quoted 74 lines · Doug Berger
Quoted 3 lines · Linus Walleij

Hi Dough,

thanks for your patch!

Thanks for your review!

Quoted 2 lines

I'm a bit confused here:

"Communication is hard" and I may be confused about your confusion, but hopefully we can work it out.

Quoted 10 lines · Doug Berger

drivers/gpio/gpiolib-of.c static int of_gpiochip_add_pin_range(struct gpio_chip *chip) · line 1066

Quoted 7 lines
drivers/gpio/gpiolib-of.c +5 -0
+               /* Ignore ranges outside of this GPIO chip */
+               if (pinspec.args[0] >= (chip->offset + chip->ngpio))
+                       continue;
+               if (pinspec.args[0] + pinspec.args[2] <= chip->offset)
+                       continue;
Quoted 18 lines

Here pinspec.args[0] and [2] comes directly from the device tree.

The documentation in Documentation/devicetree/bindings/gpio/gpio.txt says:

Quoted 13 lines

2.2) Ordinary (numerical) GPIO ranges
-------------------------------------

It is useful to represent which GPIOs correspond to which pins on which pin controllers. The gpio-ranges property described below represents this with a discrete set of ranges mapping pins from the pin controller local number space to pins in the GPIO controller local number space.

The format is: <[pin controller phandle], [GPIO controller offset],

                 [pin controller offset], [number of pins]>;

The GPIO controller offset pertains to the GPIO controller node containing the range definition.

I think we are in agreement here. For extra clarity, I will add that in my understanding pinspec.args[0] corresponds to [GPIO controller offset] and pinspec.args[2] corresponds to [number of pins].

Quoted 6 lines

So I do not understand how pinspec[0] and [2] can ever be compared to something involving chip->offset which is a Linux-specific offset.

It rather looks like you are trying to accomodate the Linux numberspace in the ranges, which it was explicitly designed to avoid.

The struct gpio_chip documentation in include/linux/gpio/driver.h says:

Quoted 3 lines

* @offset: when multiple gpio chips belong to the same device this
* can be used as offset within the device so friendly names can
* be properly assigned.

It is my understanding that this value represents the offset of a gpiochip relative to the GPIO controller device defined by the GPIO controller node in device tree. This puts it in the same number space as [GPIO controller offset]. I believe it was introduced for the specific purpose of translating [GPIO controller offset] values into Linux-specific offsets, which is why it is being reused for that purpose in this patch.

For GPIO Controllers that contain a single gpiochip the 'offset' member is 0 and the device tree node offsets can be applied directly to the gpiochip. However, when a GPIO Controller contains multiple gpiochips, the device tree node offsets must be translated to each individual gpiochip.

Quoted 5 lines

I just don't get it.

So NACK until I understand what is going on here.

Quoted 8 lines
Quoted 2 lines

Yours,
Linus Walleij

I hope it makes sense now, but if not please help me understand what I may be missing.

Thanks,
Doug

Linus,

Please let me know if this is still a NAK, if so, I'll drop this series from my tree at least for this release.

Bart

#11 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

Quoted 2 lines · Bartosz Golaszewski

Please let me know if this is still a NAK, if so, I'll drop this series from my tree at least for this release.

No I was just wrong, I had missed Sergio's addition of local offset for several-gpiochip-per-device handling completely and confused .offset with .base...

Yours,
Linus Walleij

#12 lore raw

Re: [PATCH 2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

In reply to Doug Berger
Quoted 11 lines · Doug Berger
Quoted 2 lines · Linus Walleij

It rather looks like you are trying to accomodate the Linux numberspace in the ranges, which it was explicitly designed to avoid.

The struct gpio_chip documentation in include/linux/gpio/driver.h says:

Quoted 3 lines

* @offset: when multiple gpio chips belong to the same device this
* can be used as offset within the device so friendly names can
* be properly assigned.

Quoted 7 lines

It is my understanding that this value represents the offset of a gpiochip relative to the GPIO controller device defined by the GPIO controller node in device tree. This puts it in the same number space as [GPIO controller offset]. I believe it was introduced for the specific purpose of translating [GPIO controller offset] values into Linux-specific offsets, which is why it is being reused for that purpose in this patch.

You're right, I had it confused with .base!

It's because I missed it completely when this new property was added in 2021.

Sorry for the fuzz.

Yours,
Linus Walleij

author maintainer
#13 lore raw

PATCH 3/3 gpio: brcmstb: add support for gpio-ranges

In reply to Doug Berger

A pin controller device mapped with the gpio-ranges property will need implementations of the .request and .free members of the gpiochip.

---
 drivers/gpio/gpio-brcmstb.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpio/gpio-brcmstb.c +2 -0
index 790cb278b72a..8dce78ea7139 100644
@@ -694,6 +694,8 @@ static int brcmstb_gpio_probe(struct platform_device *pdev)
 		/* not all ngpio lines are valid, will use bank width later */
 		gc->ngpio = MAX_GPIO_PER_BANK;
 		gc->offset = bank->id * MAX_GPIO_PER_BANK;
+		gc->request = gpiochip_generic_request;
+		gc->free = gpiochip_generic_free;
 		if (priv->parent_irq > 0)
 			gc->to_irq = brcmstb_gpio_to_irq;
 
Signature · 1 line
2.34.1
#15 lore raw
In reply to Doug Berger

From: Bartosz Golaszewski <[email protected]>

Quoted 8 lines · Doug Berger

The Raspberry Pi 5 includes Broadcom STB GPIO IP as well as Broadcom 2712 pin controller IP.

The community has expressed interest in linking the two drivers with the "gpio-ranges" property in device tree. This commit stack implements the necessary changes.

[...]

Applied, thanks!

[1/3] dt-bindings: gpio: brcmstb: add gpio-ranges

      commit: 7c66f8173360556ac0c3c38a91234af5a0a5a4a9
[2/3] gpio: of: support gpio-ranges for multiple gpiochip devices
      commit: e818cd3c8a345c046edff00b5ad0be4d39f7e4d4
[3/3] gpio: brcmstb: add support for gpio-ranges
      commit: 5539287ca65686f478e058a1e939294cb5682426

Best regards,

Signature · 1 line
Bartosz Golaszewski <[email protected]>
#16 lore raw
Quoted 2 lines · Bartosz Golaszewski

[2/3] gpio: of: support gpio-ranges for multiple gpiochip devices

      commit: e818cd3c8a345c046edff00b5ad0be4d39f7e4d4

I'm not sure this is a good idea. To me it looks like the commit violates the device tree bindings, which says offsets are on the local GPIO chip and turn them into offsets in the Linux GPIO numberspace.

Yours,
Linus Walleij