[PATCH 0/2] pinctrl: core: fix untreated named gpio ranges in pinctrl_pins_show()

[PATCH 0/2] pinctrl: core: fix untreated named gpio ranges in pinctrl_pins_show()

[Léo DUBOIN]

This series covers errors I encountered with the pinctrl_pins_show()
function when dealing with named gpio ranges generated through the
device tree using 'gpio-ranges-group-names'.

These errors were introduced with the original implementation in 
f1b206cf7c57561ea156798f323b0541a783bd2f.

Léo DUBOIN (2):
  pinctrl: core: take into account the pins array in pinctrl_pins_show()
  pinctrl: core: reset gpio_device in loop in pinctrl_pins_show()

 drivers/pinctrl/core.c | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)

-- 
2.42.0

[PATCH 1/2] pinctrl: core: take into account the pins array in pinctrl_pins_show()

[Léo DUBOIN]

We previously only looked at the 'pin_base' of the pinctrl_gpio_ranges
struct for determining if a pin matched a GPIO number.

This value is present only if the 'pins' array is not NULL,
and is 0 otherwise. This means that GPIO ranges declared using
gpiochip_add_pingroup_range(), thus making use of pins, were always matched
by the pins in the range [0-npins] even if they contained pins in a
completely separate range.

Signed-off-by: Léo DUBOIN <lduboin@freebox.fr>
---
 drivers/pinctrl/core.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/pinctrl/core.c b/drivers/pinctrl/core.c
index 6649357637ff..901f2f9bf850 100644
--- a/drivers/pinctrl/core.c
+++ b/drivers/pinctrl/core.c
@@ -1672,11 +1672,20 @@ static int pinctrl_pins_show(struct seq_file *s, void *what)
 #ifdef CONFIG_GPIOLIB
 		gpio_num = -1;
 		list_for_each_entry(range, &pctldev->gpio_ranges, node) {
-			if ((pin >= range->pin_base) &&
-			    (pin < (range->pin_base + range->npins))) {
-				gpio_num = range->base + (pin - range->pin_base);
-				break;
+			if (range->pins != NULL) {
+				for (int i = 0; i < range->npins; ++i) {
+					if (range->pins[i] == pin) {
+						gpio_num = range->base + i;
+						break;
+					}
+				}
+			} else if ((pin >= range->pin_base) &&
+				   (pin < (range->pin_base + range->npins))) {
+				gpio_num =
+					range->base + (pin - range->pin_base);
 			}
+			if (gpio_num != -1)
+				break;
 		}
 		if (gpio_num >= 0)
 			/*
-- 
2.42.0

[PATCH 2/2] pinctrl: core: reset gpio_device in loop in pinctrl_pins_show()

[Léo DUBOIN]

We were not resetting the pointer to the associated gpio_device once
we are done displaying a pin's information.

This meant that once we reached the end of a gpio-range, if there
were pins right after it that did not belong to any known range,
they would be associated with the previous range's gpio device.

This resulted in those pins appearing as <4294966783:old_gdev> instead
of the expected <0:?> (due to gpio_num being -1).

Signed-off-by: Léo DUBOIN <lduboin@freebox.fr>
---
 drivers/pinctrl/core.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/pinctrl/core.c b/drivers/pinctrl/core.c
index 901f2f9bf850..ad878196ada9 100644
--- a/drivers/pinctrl/core.c
+++ b/drivers/pinctrl/core.c
@@ -1670,6 +1670,7 @@ static int pinctrl_pins_show(struct seq_file *s, void *what)
 		seq_printf(s, "pin %d (%s) ", pin, desc->name);
 
 #ifdef CONFIG_GPIOLIB
+		gdev = NULL;
 		gpio_num = -1;
 		list_for_each_entry(range, &pctldev->gpio_ranges, node) {
 			if (range->pins != NULL) {
-- 
2.42.0

[PATCH RESEND 0/2] pinctrl: core: fix untreated named gpio ranges in pinctrl_pins_show()

[Léo DUBOIN]

This series covers errors I encountered with the pinctrl_pins_show()
function when dealing with named gpio ranges generated through the
device tree using 'gpio-ranges-group-names'.

These errors were introduced with the original implementation in 
f1b206cf7c57561ea156798f323b0541a783bd2f.

Léo DUBOIN (2):
  pinctrl: core: take into account the pins array in pinctrl_pins_show()
  pinctrl: core: reset gpio_device in loop in pinctrl_pins_show()

 drivers/pinctrl/core.c | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)

-- 
2.42.0

[PATCH RESEND 1/2] pinctrl: core: take into account the pins array in pinctrl_pins_show()

[Léo DUBOIN]

We previously only looked at the 'pin_base' of the pinctrl_gpio_ranges
struct for determining if a pin matched a GPIO number.

This value is present only if the 'pins' array is not NULL,
and is 0 otherwise. This means that GPIO ranges declared using
gpiochip_add_pingroup_range(), thus making use of pins, were always matched
by the pins in the range [0-npins] even if they contained pins in a
completely separate range.

Signed-off-by: Léo DUBOIN <lduboin@freebox.fr>
---
 drivers/pinctrl/core.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/pinctrl/core.c b/drivers/pinctrl/core.c
index 6649357637ff..901f2f9bf850 100644
--- a/drivers/pinctrl/core.c
+++ b/drivers/pinctrl/core.c
@@ -1672,11 +1672,20 @@ static int pinctrl_pins_show(struct seq_file *s, void *what)
 #ifdef CONFIG_GPIOLIB
 		gpio_num = -1;
 		list_for_each_entry(range, &pctldev->gpio_ranges, node) {
-			if ((pin >= range->pin_base) &&
-			    (pin < (range->pin_base + range->npins))) {
-				gpio_num = range->base + (pin - range->pin_base);
-				break;
+			if (range->pins != NULL) {
+				for (int i = 0; i < range->npins; ++i) {
+					if (range->pins[i] == pin) {
+						gpio_num = range->base + i;
+						break;
+					}
+				}
+			} else if ((pin >= range->pin_base) &&
+				   (pin < (range->pin_base + range->npins))) {
+				gpio_num =
+					range->base + (pin - range->pin_base);
 			}
+			if (gpio_num != -1)
+				break;
 		}
 		if (gpio_num >= 0)
 			/*
-- 
2.42.0

[PATCH RESEND 2/2] pinctrl: core: reset gpio_device in loop in pinctrl_pins_show()

[Léo DUBOIN]

We were not resetting the pointer to the associated gpio_device once
we are done displaying a pin's information.

This meant that once we reached the end of a gpio-range, if there
were pins right after it that did not belong to any known range,
they would be associated with the previous range's gpio device.

This resulted in those pins appearing as <4294966783:old_gdev> instead
of the expected <0:?> (due to gpio_num being -1).

Signed-off-by: Léo DUBOIN <lduboin@freebox.fr>
---
 drivers/pinctrl/core.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/pinctrl/core.c b/drivers/pinctrl/core.c
index 901f2f9bf850..ad878196ada9 100644
--- a/drivers/pinctrl/core.c
+++ b/drivers/pinctrl/core.c
@@ -1670,6 +1670,7 @@ static int pinctrl_pins_show(struct seq_file *s, void *what)
 		seq_printf(s, "pin %d (%s) ", pin, desc->name);
 
 #ifdef CONFIG_GPIOLIB
+		gdev = NULL;
 		gpio_num = -1;
 		list_for_each_entry(range, &pctldev->gpio_ranges, node) {
 			if (range->pins != NULL) {
-- 
2.42.0

Re: [PATCH RESEND 0/2] pinctrl: core: fix untreated named gpio ranges in pinctrl_pins_show()

[Linus Walleij]

On Fri, May 10, 2024 at 6:26 PM Léo DUBOIN <lduboin@freebox.fr> wrote:

> This series covers errors I encountered with the pinctrl_pins_show()
> function when dealing with named gpio ranges generated through the
> device tree using 'gpio-ranges-group-names'.
>
> These errors were introduced with the original implementation in
> f1b206cf7c57561ea156798f323b0541a783bd2f.
>
> Léo DUBOIN (2):
>   pinctrl: core: take into account the pins array in pinctrl_pins_show()
>   pinctrl: core: reset gpio_device in loop in pinctrl_pins_show()

Patches applied for kernel v6.11.

It's debugfs so per definition no regression, people get to test it
in linux-next.

Thanks for looking into this!

Yours,
Linus Walleij