Re: [PATCH v5 8/8] gpiolib: notify user-space about in-kernel line state changes

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Thu, 24 Oct 2024 08:49:30 +0200, Bartosz Golaszewski <brgl@xxxxxxxx> said:
> On Wed, Oct 23, 2024 at 11:05 PM Mark Brown <broonie@xxxxxxxxxx> wrote:
>>
>> On Fri, Oct 18, 2024 at 11:10:16AM +0200, Bartosz Golaszewski wrote:
>> > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
>> >
>> > We currently only notify user-space about line config changes that are
>> > made from user-space. Any kernel config changes are not signalled.
>> >
>> > Let's improve the situation by emitting the events closer to the source.
>> > To that end let's call the relevant notifier chain from the functions
>> > setting direction, gpiod_set_config(), gpiod_set_consumer_name() and
>> > gpiod_toggle_active_low(). This covers all the options that we can
>> > inform the user-space about. We ignore events which don't have
>> > corresponding flags exported to user-space on purpose - otherwise the
>> > user would see a config-changed event but the associated line-info would
>> > remain unchanged.
>>
>> Today's -next is not booting on several of my platforms, including
>> beaglebone-black, i.MX8MP-EVK and pine64plus.  Bisects are pointing at
>> this commit, and i.MX8MP-EVK is actually giving some console output:
>>
>> [    2.502208] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
>> [    2.511036] Mem abort info:
>>
>> ...
>>
>> [    2.679934] Call trace:
>> [    2.682379]  gpiod_direction_output+0x34/0x5c
>> [    2.686745]  i2c_register_adapter+0x59c/0x670
>> [    2.691111]  __i2c_add_numbered_adapter+0x58/0xa8
>> [    2.695822]  i2c_add_adapter+0xa0/0xd0
>> [    2.699578]  i2c_add_numbered_adapter+0x2c/0x38
>> [    2.704117]  i2c_imx_probe+0x2d0/0x640
>>
>> which looks plausible given the change.
>>
>> Full boot log for i.MX8MP-EVK:
>>
>>    https://lava.sirena.org.uk/scheduler/job/887083
>>
>> Bisect log for that, the others look similar (the long run of good/bad
>> tags at the start for random commits is my automation pulling test
>> results it already knows about in the affected range to try to speed up
>> the bisect):
>>
>
> Hi Mark!
>
> Any chance you could post the output of
>
>     scripts/faddr2line drivers/gpio/gpiolib.o gpiod_direction_output+0x34/0x5c
>
> for that build?
>
> Bart
>

While I can't really reproduce it, I've looked at what could be wrong and
figured that the NULL-pointer in question can possibly be the
line_state_notifier.

I realized that for some historical reasons we add the GPIO device to the
global list before it's fully set up - including initializing the notifier.
In some corner cases (devlinks borked?) this could lead to consumers requesting
GPIOs before their provider is ready.

Mark: could you try the following diff and let me know if it works?

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index ae758ba6dc3d..4258acac0162 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -987,45 +987,6 @@ int gpiochip_add_data_with_key(struct gpio_chip
*gc, void *data,

 	gdev->ngpio = gc->ngpio;
 	gdev->can_sleep = gc->can_sleep;
-
-	scoped_guard(mutex, &gpio_devices_lock) {
-		/*
-		 * TODO: this allocates a Linux GPIO number base in the global
-		 * GPIO numberspace for this chip. In the long run we want to
-		 * get *rid* of this numberspace and use only descriptors, but
-		 * it may be a pipe dream. It will not happen before we get rid
-		 * of the sysfs interface anyways.
-		 */
-		base = gc->base;
-		if (base < 0) {
-			base = gpiochip_find_base_unlocked(gc->ngpio);
-			if (base < 0) {
-				ret = base;
-				base = 0;
-				goto err_free_label;
-			}
-
-			/*
-			 * TODO: it should not be necessary to reflect the
-			 * assigned base outside of the GPIO subsystem. Go over
-			 * drivers and see if anyone makes use of this, else
-			 * drop this and assign a poison instead.
-			 */
-			gc->base = base;
-		} else {
-			dev_warn(&gdev->dev,
-				 "Static allocation of GPIO base is deprecated, use dynamic
allocation.\n");
-		}
-
-		gdev->base = base;
-
-		ret = gpiodev_add_to_list_unlocked(gdev);
-		if (ret) {
-			chip_err(gc, "GPIO integer space overlap, cannot add chip\n");
-			goto err_free_label;
-		}
-	}
-
 	ATOMIC_INIT_NOTIFIER_HEAD(&gdev->line_state_notifier);
 	BLOCKING_INIT_NOTIFIER_HEAD(&gdev->device_notifier);

@@ -1103,6 +1064,45 @@ int gpiochip_add_data_with_key(struct gpio_chip
*gc, void *data,
 		if (ret)
 			goto err_remove_irqchip;
 	}
+
+	scoped_guard(mutex, &gpio_devices_lock) {
+		/*
+		 * TODO: this allocates a Linux GPIO number base in the global
+		 * GPIO numberspace for this chip. In the long run we want to
+		 * get *rid* of this numberspace and use only descriptors, but
+		 * it may be a pipe dream. It will not happen before we get rid
+		 * of the sysfs interface anyways.
+		 */
+		base = gc->base;
+		if (base < 0) {
+			base = gpiochip_find_base_unlocked(gc->ngpio);
+			if (base < 0) {
+				ret = base;
+				base = 0;
+				goto err_free_label;
+			}
+
+			/*
+			 * TODO: it should not be necessary to reflect the
+			 * assigned base outside of the GPIO subsystem. Go over
+			 * drivers and see if anyone makes use of this, else
+			 * drop this and assign a poison instead.
+			 */
+			gc->base = base;
+		} else {
+			dev_warn(&gdev->dev,
+				 "Static allocation of GPIO base is deprecated, use dynamic
allocation.\n");
+		}
+
+		gdev->base = base;
+
+		ret = gpiodev_add_to_list_unlocked(gdev);
+		if (ret) {
+			chip_err(gc, "GPIO integer space overlap, cannot add chip\n");
+			goto err_free_label;
+		}
+	}
+
 	return 0;

 err_remove_irqchip:

Thanks
Bartosz