Skip to contentSkip to conversation

iio: magnetometer: ak8975: ensure device is awake for buffered capture

PATCH In review Joshua Crofts 3 participants 7 messages May 13 – May 15 IIO

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iio: magnetometer: ak8975: ensure device is awake for buffered capture

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

---
 drivers/iio/magnetometer/ak8975.c | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)
drivers/iio/magnetometer/ak8975.c +23 -1
index dbab4d0bba348be70e5c8b71678de935879bff42..0fb2fd03d11ce195a67949056c000c473a8d3e99 100644
@@ -899,6 +899,28 @@ static irqreturn_t ak8975_handle_trigger(int irq, void *p)
 	return IRQ_HANDLED;
 }
 
+static int ak8975_buffer_preenable(struct iio_dev *indio_dev)
+{
+	struct ak8975_data *data = iio_priv(indio_dev);
+	struct device *dev = &data->client->dev;
+
+	return pm_runtime_resume_and_get(dev);
+}
+
+static int ak8975_buffer_postdisable(struct iio_dev *indio_dev)
+{
+	struct ak8975_data *data = iio_priv(indio_dev);
+	struct device *dev = &data->client->dev;
+
+	pm_runtime_put_autosuspend(dev);
+
+	return 0;
+}
+
+static const struct iio_buffer_setup_ops ak8975_buffer_setup_ops = {
+	.preenable = ak8975_buffer_preenable,
+	.postdisable = ak8975_buffer_postdisable,
+};
 static int ak8975_probe(struct i2c_client *client)
 {
 	const struct i2c_device_id *id = i2c_client_get_device_id(client);
@@ -992,7 +1014,7 @@ static int ak8975_probe(struct i2c_client *client)
 	indio_dev->name = name;
 
 	ret = iio_triggered_buffer_setup(indio_dev, NULL, ak8975_handle_trigger,
-					 NULL);
+					 &ak8975_buffer_setup_ops);
 	if (ret) {
 		dev_err(&client->dev, "triggered buffer setup failed\n");
 		goto power_off;

---
base-commit: 1548c54e9adc32a719499216f63fba14b2fc07c3
change-id: 20260513-ak8975-fix-53b1b02857dd

Best regards,

Signature · 1 line
Joshua Crofts <[email protected]>
#1 lore raw
Quoted 10 lines · On Wed, May 13, 2026 at 04:35:52PM +0200, Joshua Crofts via B4 Relay wrote:

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Quoted 3 lines

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

I think Suggested-by: Jonathan is appropriate here.

...

drivers/iio/magnetometer/ak8975.c static irqreturn_t ak8975_handle_trigger(int irq, void *p) · line 905

Quoted 4 lines
drivers/iio/magnetometer/ak8975.c +4 -0
+static int ak8975_buffer_preenable(struct iio_dev *indio_dev)
+{
+	struct ak8975_data *data = iio_priv(indio_dev);
+	struct device *dev = &data->client->dev;

Not sure if actually may use indio_dev->dev.parent here instead of this, but at least this one is robust against changing parent-child relations).

Quoted 2 lines
(quoted diff) +2 -0
+	return pm_runtime_resume_and_get(dev);
+}
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#2 lore raw
Quoted 20 lines · Andy Shevchenko
Quoted 10 lines · On Wed, May 13, 2026 at 04:35:52PM +0200, Joshua Crofts via B4 Relay wrote:

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Quoted 3 lines

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

I think Suggested-by: Jonathan is appropriate here.

...

Quoted 7 lines

drivers/iio/magnetometer/ak8975.c static irqreturn_t ak8975_handle_trigger(int irq, void *p) · line 905

Quoted 4 lines
drivers/iio/magnetometer/ak8975.c +4 -0
+static int ak8975_buffer_preenable(struct iio_dev *indio_dev)
+{
+     struct ak8975_data *data = iio_priv(indio_dev);
+     struct device *dev = &data->client->dev;

Not sure if actually may use indio_dev->dev.parent here instead of this, but at least this one is robust against changing parent-child relations).

It being robust was my intention, considering this will be backported at some point.

Signature · 3 lines
Kind regards

CJD
#3 lore raw
In reply to Joshua Crofts
Quoted 17 lines · On Wed, 13 May 2026 at 16:35, Joshua Crofts via B4 Relay…

From: Joshua Crofts <[email protected]>

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

Quoted 5 lines

---

FYI, Sashiko reported an additional issue concerning the leaking of uninitialized kernel stack memory.

https://sashiko.dev/#/patchset/20260513-ak8975-fix-v1-1-104ea605dd54%40gmail.com

maintainer
#4 lore raw
In reply to Joshua Crofts
Quoted 12 lines · On Wed, 13 May 2026 16:35:52 +0200 Joshua Crofts via B4 Relay…

From: Joshua Crofts <[email protected]>

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Quoted 8 lines

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

I'm going to take this as a clear improvement but the point sashiko makes about turning runtime PM on only after the interfaces are exposed is possibly valid as a general comment. It applies to all the other paths though, not just the buffered capture.

I believe we are actually fine here because the device is powered up at that point (the set_active making that clear).

Open to other people's views on this though.

This is another place where we really should work out a best practice guide / set of standard patterns to use.

Anyhow, applied this one and this is one I've marked for stable. However given mass of stuff around this driver it'll still go the slow way and we may need to manually do backports. So applied to the testing branch of iio.git.

Thanks,

Jonathan

Quoted 52 lines
---
 drivers/iio/magnetometer/ak8975.c | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)
drivers/iio/magnetometer/ak8975.c +23 -2
index dbab4d0bba348be70e5c8b71678de935879bff42..0fb2fd03d11ce195a67949056c000c473a8d3e99 100644
@@ -899,6 +899,28 @@ static irqreturn_t ak8975_handle_trigger(int irq, void *p)
 	return IRQ_HANDLED;
 }
 
+static int ak8975_buffer_preenable(struct iio_dev *indio_dev)
+{
+	struct ak8975_data *data = iio_priv(indio_dev);
+	struct device *dev = &data->client->dev;
+
+	return pm_runtime_resume_and_get(dev);
+}
+
+static int ak8975_buffer_postdisable(struct iio_dev *indio_dev)
+{
+	struct ak8975_data *data = iio_priv(indio_dev);
+	struct device *dev = &data->client->dev;
+
+	pm_runtime_put_autosuspend(dev);
+
+	return 0;
+}
+
+static const struct iio_buffer_setup_ops ak8975_buffer_setup_ops = {
+	.preenable = ak8975_buffer_preenable,
+	.postdisable = ak8975_buffer_postdisable,
+};
 static int ak8975_probe(struct i2c_client *client)
 {
 	const struct i2c_device_id *id = i2c_client_get_device_id(client);
@@ -992,7 +1014,7 @@ static int ak8975_probe(struct i2c_client *client)
 	indio_dev->name = name;
 
 	ret = iio_triggered_buffer_setup(indio_dev, NULL, ak8975_handle_trigger,
-					 NULL);
+					 &ak8975_buffer_setup_ops);
 	if (ret) {
 		dev_err(&client->dev, "triggered buffer setup failed\n");
 		goto power_off;
 
---

base-commit: 1548c54e9adc32a719499216f63fba14b2fc07c3
change-id: 20260513-ak8975-fix-53b1b02857dd

Best regards,

#5 lore raw
Quoted 34 lines · Jonathan Cameron
Quoted 12 lines · On Wed, 13 May 2026 16:35:52 +0200 Joshua Crofts via B4 Relay…

From: Joshua Crofts <[email protected]>

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Quoted 8 lines

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

I'm going to take this as a clear improvement but the point sashiko makes about turning runtime PM on only after the interfaces are exposed is possibly valid as a general comment. It applies to all the other paths though, not just the buffered capture.

I believe we are actually fine here because the device is powered up at that point (the set_active making that clear).

Open to other people's views on this though.

Quoted 7 lines

This is another place where we really should work out a best practice guide / set of standard patterns to use.

Anyhow, applied this one and this is one I've marked for stable. However given mass of stuff around this driver it'll still go the slow way and we may need to manually do backports. So applied to the testing branch of iio.git.

Ah, I forgot to send the v2 with your Suggested-by tag. Up to you to amend it I guess.

Thanks

Signature · 3 lines
Kind regards

CJD
maintainer
#6 lore raw
Quoted 38 lines · On Fri, 15 May 2026 15:22:42 +0200 Joshua Crofts <[email protected]>…
Quoted 37 lines · Jonathan Cameron
Quoted 12 lines · On Wed, 13 May 2026 16:35:52 +0200 Joshua Crofts via B4 Relay…

From: Joshua Crofts <[email protected]>

Currently, the ak8975_start_read_axis() can be called while the device is autosuspended, causing two issues:

1. I2C transfers in the aforementioned function will fail or timeout because ak8975_runtime_suspend() disables the device regulators.
2. Since ak8975_fill_buffer() does not hold runtime references, ak8975_runtime_suspend() can run concurrently, and since PM callbacks do not use a locking mechanism, it may cause a race accessing the control register via the I2C bus.

Quoted 8 lines

Fix this issue by adding struct iio_buffer_setup_ops that contains preenable and postdisable functions to ensure correct that device is powered on when running a buffered capture.

I'm going to take this as a clear improvement but the point sashiko makes about turning runtime PM on only after the interfaces are exposed is possibly valid as a general comment. It applies to all the other paths though, not just the buffered capture.

I believe we are actually fine here because the device is powered up at that point (the set_active making that clear).

Open to other people's views on this though.

This is another place where we really should work out a best practice guide / set of standard patterns to use.

Quoted 7 lines
Quoted 4 lines

Anyhow, applied this one and this is one I've marked for stable. However given mass of stuff around this driver it'll still go the slow way and we may need to manually do backports. So applied to the testing branch of iio.git.

Ah, I forgot to send the v2 with your Suggested-by tag. Up to you to amend it I guess.

I'm lazy and don't care + don't resend to add tags if there is nothing else. Just rely to the thread with it then b4 will pick it up. I'll have to tweak the order though as it'll put it at the end of the tag block.

Quoted 3 lines

Thanks