Skip to contentSkip to conversation

gpio: provide for_each_gpio()

PATCH In review Bartosz Golaszewski 2 participants 6 messages Feb 15, 2024 – Feb 22, 2024 GPIO

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 gpio: provide for_each_gpio()

From: Bartosz Golaszewski <[email protected]>

We only provide iterators for requested GPIOs to provider drivers. In order to allow them to display debug information about all GPIOs, let's provide a variant for iterating over all GPIOs.

---
 include/linux/gpio/driver.h | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
include/linux/gpio/driver.h +15 -0
index 9d0023f83a57..5f915b653548 100644
@@ -551,6 +551,21 @@ DEFINE_CLASS(_gpiochip_for_each_data,
 	     }),
 	     const char **label, int *i)
 
+/**
+ * for_each_gpio - Iterates over all GPIOs for given chip.
+ * @_chip: Chip to iterate over.
+ * @_i: Loop counter.
+ * @_label: Place to store the address of the label if the GPIO is requested.
+ *          Set to NULL for unused GPIOs.
+ */
+#define for_each_gpio(_chip, _i, _label) \
+	for (CLASS(_gpiochip_for_each_data, _data)(&_label, &_i); \
+	     *_data.i < _chip->ngpio; \
+	     (*_data.i)++, kfree(*(_data.label)), *_data.label = NULL) \
+		if (IS_ERR(*_data.label = \
+			gpiochip_dup_line_label(_chip, *_data.i))) {} \
+		else
+
 /**
  * for_each_requested_gpio_in_range - iterates over requested GPIOs in a given range
  * @_chip:	the chip to query
Signature · 1 line
2.40.1
#1 lore raw

PATCH 2/2 gpio: sim: use for_each_gpio()

From: Bartosz Golaszewski <[email protected]>

Display debugfs information about all simulated GPIOs, not only the requested ones.

---
 drivers/gpio/gpio-sim.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpio/gpio-sim.c +2 -2
index c4106e37e6db..1ebd6961ffba 100644
@@ -234,10 +234,10 @@ static void gpio_sim_dbg_show(struct seq_file *seq, struct gpio_chip *gc)
 
 	guard(mutex)(&chip->lock);
 
-	for_each_requested_gpio(gc, i, label)
+	for_each_gpio(gc, i, label)
 		seq_printf(seq, " gpio-%-3d (%s) %s,%s\n",
 			   gc->base + i,
-			   label,
+			   label ?: "<unused>",
 			   test_bit(i, chip->direction_map) ? "input" :
 				test_bit(i, chip->value_map) ? "output-high" :
 							       "output-low",
Signature · 1 line
2.40.1
Reviewed-by on v1 1/2
#3 lore raw
In reply to Bartosz Golaszewski

Hi Bartosz,

Quoted 7 lines · Bartosz Golaszewski

From: Bartosz Golaszewski <[email protected]>

We only provide iterators for requested GPIOs to provider drivers. In order to allow them to display debug information about all GPIOs, let's provide a variant for iterating over all GPIOs.

(...)

Quoted 2 lines

+/**
+ * for_each_gpio - Iterates over all GPIOs for given chip.

Does this really intuitively fit with other functions named for_each_XXX()?

include/linux/gpio/driver.h DEFINE_CLASS(_gpiochip_for_each_data, · line 567

Quoted 12 lines

+ * @_chip: Chip to iterate over.
+ * @_i: Loop counter.
+ * @_label: Place to store the address of the label if the GPIO is requested.
+ * Set to NULL for unused GPIOs.
+ */

include/linux/gpio/driver.h +7 -0
+#define for_each_gpio(_chip, _i, _label) \
+       for (CLASS(_gpiochip_for_each_data, _data)(&_label, &_i); \
+            *_data.i < _chip->ngpio; \
+            (*_data.i)++, kfree(*(_data.label)), *_data.label = NULL) \
+               if (IS_ERR(*_data.label = \
+                       gpiochip_dup_line_label(_chip, *_data.i))) {} \
+               else

I would call it for_each_line_label() or something. I try to avoid using "gpio" in function names as well because of ambiguity, I could also go with for_each_hwgpio_label() I suppose.

With some more reasonable name:

Yours,
Linus Walleij

#4 lore raw
Quoted 32 lines · Linus Walleij

Hi Bartosz,

Quoted 7 lines · Bartosz Golaszewski

From: Bartosz Golaszewski <[email protected]>

We only provide iterators for requested GPIOs to provider drivers. In order to allow them to display debug information about all GPIOs, let's provide a variant for iterating over all GPIOs.

(...)

Quoted 2 lines

+/**
+ * for_each_gpio - Iterates over all GPIOs for given chip.

Does this really intuitively fit with other functions named for_each_XXX()?

include/linux/gpio/driver.h DEFINE_CLASS(_gpiochip_for_each_data, · line 567

Quoted 12 lines

+ * @_chip: Chip to iterate over.
+ * @_i: Loop counter.
+ * @_label: Place to store the address of the label if the GPIO is requested.
+ * Set to NULL for unused GPIOs.
+ */

include/linux/gpio/driver.h +7 -0
+#define for_each_gpio(_chip, _i, _label) \
+       for (CLASS(_gpiochip_for_each_data, _data)(&_label, &_i); \
+            *_data.i < _chip->ngpio; \
+            (*_data.i)++, kfree(*(_data.label)), *_data.label = NULL) \
+               if (IS_ERR(*_data.label = \
+                       gpiochip_dup_line_label(_chip, *_data.i))) {} \
+               else
Quoted 3 lines

I would call it for_each_line_label() or something. I try to avoid using "gpio" in function names as well because of ambiguity, I could also go with for_each_hwgpio_label() I suppose.

The problem is: this doesn't iterate over labels. It really goes through all offsets and if there's no consumer then the label is NULL (I should have said that in the kerneldoc).

Quoted 2 lines

With some more reasonable name:

Does for_each_hwgpio() make more sense?

Bart

Quoted 4 lines

Yours,
Linus Walleij

#5 lore raw
Quoted 37 lines · Bartosz Golaszewski
Quoted 32 lines · Linus Walleij

Hi Bartosz,

Quoted 7 lines · Bartosz Golaszewski

From: Bartosz Golaszewski <[email protected]>

We only provide iterators for requested GPIOs to provider drivers. In order to allow them to display debug information about all GPIOs, let's provide a variant for iterating over all GPIOs.

(...)

Quoted 2 lines

+/**
+ * for_each_gpio - Iterates over all GPIOs for given chip.

Does this really intuitively fit with other functions named for_each_XXX()?

include/linux/gpio/driver.h DEFINE_CLASS(_gpiochip_for_each_data, · line 567

Quoted 12 lines

+ * @_chip: Chip to iterate over.
+ * @_i: Loop counter.
+ * @_label: Place to store the address of the label if the GPIO is requested.
+ * Set to NULL for unused GPIOs.
+ */

include/linux/gpio/driver.h +7 -0
+#define for_each_gpio(_chip, _i, _label) \
+       for (CLASS(_gpiochip_for_each_data, _data)(&_label, &_i); \
+            *_data.i < _chip->ngpio; \
+            (*_data.i)++, kfree(*(_data.label)), *_data.label = NULL) \
+               if (IS_ERR(*_data.label = \
+                       gpiochip_dup_line_label(_chip, *_data.i))) {} \
+               else
Quoted 3 lines

I would call it for_each_line_label() or something. I try to avoid using "gpio" in function names as well because of ambiguity, I could also go with for_each_hwgpio_label() I suppose.

Quoted 8 lines

The problem is: this doesn't iterate over labels. It really goes through all offsets and if there's no consumer then the label is NULL (I should have said that in the kerneldoc).

Quoted 2 lines

With some more reasonable name:

Does for_each_hwgpio() make more sense?

It's better it reflects the usage, but isn't the usage to conditionally extract the label (or NULL) for each hwgpio?

What I'm after is if there is a risk that someone think this is a generic iterator for hwgpios which would be confusing.

At the same time for_each_hwgpio_attempt_extract_label() is a bit long I guess.

I don't wanna bikeshed too much so go with for_each_hwgpio() if you don't immediately see anything better.

Yours,
Linus Walleij