Skip to contentSkip to conversation

iio: accel: bmc150: use IIO_DECLARE_BUFFER_WITH_TS

PATCH Superseded Yash Suthar 4 participants 8 messages Aug 8 – Aug 12 IIO

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iio: accel: bmc150: use IIO_DECLARE_BUFFER_WITH_TS

Replace bmc150_accel_data plain buffer with
IIO_DECLARE_BUFFER_WITH_TS() that also keep
timestamp aligned.

---
 drivers/iio/accel/bmc150-accel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/accel/bmc150-accel.h +1 -1
index e8f26198359f..e0773533efeb 100644
@@ -64,7 +64,7 @@ struct bmc150_accel_data {
 	struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS];
 	struct mutex mutex;
 	u8 fifo_mode, watermark;
-	s16 buffer[8];
+	IIO_DECLARE_BUFFER_WITH_TS(s16, buffer, 3);
 	/*
 	 * Ensure there is sufficient space and correct alignment for
 	 * the timestamp if enabled
Signature · 1 line
2.43.0
reviewer
#1 lore raw
Quoted 3 lines · Yash Suthar

Replace bmc150_accel_data plain buffer with
IIO_DECLARE_BUFFER_WITH_TS() that also keep
timestamp aligned.

Please don't wrap the lines so short. In git commits we usually aim for 76 columns, IIRC.

Quoted 11 lines
---
 drivers/iio/accel/bmc150-accel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/accel/bmc150-accel.h +0 -0
index e8f26198359f..e0773533efeb 100644

drivers/iio/accel/bmc150-accel.h struct bmc150_accel_data { · line 70

Quoted 8 lines
drivers/iio/accel/bmc150-accel.h +1 -1
@@ -64,7 +64,7 @@ struct bmc150_accel_data {
 	struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS];
 	struct mutex mutex;
 	u8 fifo_mode, watermark;
-	s16 buffer[8];
+	IIO_DECLARE_BUFFER_WITH_TS(s16, buffer, 3);
 	/*
 	 * Ensure there is sufficient space and correct alignment for
 	 * the timestamp if enabled

Probably also needs fixes tag since the old array was not properly aligned.

And as a follow-up task (in a separate patch), we can look at replacing iio_push_to_buffers_with_timestamp() with iio_push_to_buffers_with_ts() in this driver. Should be trivial in this case.

#2 lore raw

PATCH v2 1/1 iio: accel: bmc150: use IIO_DECLARE_BUFFER_WITH_TS

In reply to Yash Suthar

Replace bmc150_accel_data plain buffer with IIO_DECLARE_BUFFER_WITH_TS() that also keep timestamp aligned.

Fixes bd7fe5b71918 ("iio: accel: BMC150 accel support")

---
v2:
- Rewrap commit message.
- Add Fixes tag.

 drivers/iio/accel/bmc150-accel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/accel/bmc150-accel.h +1 -1
index e8f26198359f..e0773533efeb 100644
@@ -64,7 +64,7 @@ struct bmc150_accel_data {
 	struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS];
 	struct mutex mutex;
 	u8 fifo_mode, watermark;
-	s16 buffer[8];
+	IIO_DECLARE_BUFFER_WITH_TS(s16, buffer, 3);
 	/*
 	 * Ensure there is sufficient space and correct alignment for
 	 * the timestamp if enabled
Signature · 1 line
2.43.0
maintainer
#3 lore raw
Quoted 5 lines · On Sun, 9 Aug 2026 11:24:42 +0530 Yash Suthar <[email protected]> wrote:

Replace bmc150_accel_data plain buffer with IIO_DECLARE_BUFFER_WITH_TS() that also keep timestamp aligned.

Fixes bd7fe5b71918 ("iio: accel: BMC150 accel support")

For future reference, please never reply to an existing thread with a new version of a patch. I'm a bit confused why this one keeps coming up as I'm not aware of any part of the kernel that requests doing it this way. Reasons not to do this:
1) Confusing threads once they get sufficient numbers of replies, including making it harder for tooling to work out what is going on.
2) Reviewers and maintainers tend to use mail clients that put replies to old threads, somewhere back in history, so the chances of getting a review is reduced.

Anyhow, don't resend existing patches to 'fix this' but make sure to do new threads, if you send out any new versions.

This looks fine to me so applied to the fixes-togreg branch of iio.git Note that branch will be rebased on rc1 once it is available.

Thanks,

Jonathan

Quoted 21 lines

---
v2:
- Rewrap commit message.
- Add Fixes tag.

 drivers/iio/accel/bmc150-accel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/accel/bmc150-accel.h +1 -1
index e8f26198359f..e0773533efeb 100644
@@ -64,7 +64,7 @@ struct bmc150_accel_data {
 	struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS];
 	struct mutex mutex;
 	u8 fifo_mode, watermark;
-	s16 buffer[8];
+	IIO_DECLARE_BUFFER_WITH_TS(s16, buffer, 3);
 	/*
 	 * Ensure there is sufficient space and correct alignment for
 	 * the timestamp if enabled
maintainer
#4 lore raw
Quoted 19 lines · On Sun, 9 Aug 2026 23:59:34 +0100 Jonathan Cameron <[email protected]> wrote:
Quoted 5 lines · On Sun, 9 Aug 2026 11:24:42 +0530 Yash Suthar <[email protected]> wrote:

Replace bmc150_accel_data plain buffer with IIO_DECLARE_BUFFER_WITH_TS() that also keep timestamp aligned.

Fixes bd7fe5b71918 ("iio: accel: BMC150 accel support")

For future reference, please never reply to an existing thread with a new version of a patch. I'm a bit confused why this one keeps coming up as I'm not aware of any part of the kernel that requests doing it this way. Reasons not to do this:
1) Confusing threads once they get sufficient numbers of replies, including making it harder for tooling to work out what is going on.
2) Reviewers and maintainers tend to use mail clients that put replies to old threads, somewhere back in history, so the chances of getting a review is reduced.

Quoted 5 lines

Anyhow, don't resend existing patches to 'fix this' but make sure to do new threads, if you send out any new versions.

This looks fine to me so applied to the fixes-togreg branch of iio.git Note that branch will be rebased on rc1 once it is available.

Actually no. I've backed that out.

Given the data alignment is fixed, if we were going to do this it would be clearer as

struct {
	s16 chans[3]; //see later, I believe this should be __le16
	aligned_s64 timestamp;
};

But that is very similar to the structure that follows immediately after this, but that has __le16 chans[]

So what is going on here?

bmc150_accel_trigger_handler() Does a bulk read into data->buffer (so the array this patch is touching). That is then pushed to the buffers. Note this should be DMA safe, so we need to ensure nothing shares cache line with it that might be edited concurrently.

data->scan is used in __bmc150_accel_fifo_flush() by memcpying from a local buffer into this to marshal data.

That local buffer isn't dma safe, so good to fix that as a separate issue. The buffer is fairly small (32 x 3 x 2 bytes) so one choice would be to put an aligned buffer at the end of the iio_priv() structure. Anyhow, that's a different issue.

So we already have two very similar structures.
The chan spec is little endian so the scan one is more correct.

So I think the fix for this issue is move the scan element to the end of struct bmc150_accel_data and mark it with __aligned(IIO_DMA_MINALIGN);

Then use that for both the bmc150_accel_trigger_handler() and __bmc150_accel_fifo_flush() paths. A second fix will resolve the local buffer in __bmc150_accel_fifo_flush() that is being used for a bulk regmap transfer that may need a dma safe buffer.

I'd place that after the moved scan element in struct bmc150_accel_data. Do that change as a separate fix. As we have two dma buffer issues in this driver take a look for any other bulk regmap accesses that may have similar problems!

Please combine these two fixes (if you agree with my analysis) with a follow up to do the iio_push_to_buffers_with_ts() all in one series as they will be touching the same code.

Jonathan

Quoted 30 lines

Thanks,

Jonathan

Quoted 21 lines

---
v2:
- Rewrap commit message.
- Add Fixes tag.

 drivers/iio/accel/bmc150-accel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/accel/bmc150-accel.h +1 -1
index e8f26198359f..e0773533efeb 100644
@@ -64,7 +64,7 @@ struct bmc150_accel_data {
 	struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS];
 	struct mutex mutex;
 	u8 fifo_mode, watermark;
-	s16 buffer[8];
+	IIO_DECLARE_BUFFER_WITH_TS(s16, buffer, 3);
 	/*
 	 * Ensure there is sufficient space and correct alignment for
 	 * the timestamp if enabled  
#5 lore raw
Quoted 4 lines · Jonathan Cameron
Quoted 2 lines · On Sun, 9 Aug 2026 23:59:34 +0100 Jonathan Cameron <[email protected]> wrote:

On Sun, 9 Aug 2026 11:24:42 +0530
Yash Suthar <[email protected]> wrote:

...

Quoted 9 lines

Actually no. I've backed that out.

Given the data alignment is fixed, if we were going to do this it would be clearer as

struct {
	s16 chans[3]; //see later, I believe this should be __le16
	aligned_s64 timestamp;
};
Quoted 4 lines

But that is very similar to the structure that follows immediately after this, but that has __le16 chans[]

So what is going on here?

AI slop?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
maintainer
#6 lore raw
Quoted 17 lines · On Mon, 10 Aug 2026 12:00:05 +0300 Andy Shevchenko…
Quoted 4 lines · Jonathan Cameron
Quoted 2 lines · On Sun, 9 Aug 2026 23:59:34 +0100 Jonathan Cameron <[email protected]> wrote:

On Sun, 9 Aug 2026 11:24:42 +0530
Yash Suthar <[email protected]> wrote:

...

Quoted 9 lines

Actually no. I've backed that out.

Given the data alignment is fixed, if we were going to do this it would be clearer as

struct {
	s16 chans[3]; //see later, I believe this should be __le16
	aligned_s64 timestamp;
};
Quoted 6 lines
Quoted 4 lines

But that is very similar to the structure that follows immediately after this, but that has __le16 chans[]

So what is going on here?

AI slop?

It's the original code I'm complaining about. Been around a while! So for once, nope (I think!)

Jonathan

Quoted 1 line
#7 lore raw
In reply to Jonathan Cameron
Quoted 3 lines · Andy Shevchenko
Quoted 1 line

So what is going on here?

AI slop?

No. I only intended to convert buffer to IIO_DECLARE_BUFFER_WITH_TS() and dont want to mix it with the existing scan member. Really sorry for that i should have asked.

Quoted 17 lines · Jonathan Cameron

Given the data alignment is fixed, if we were going to do this it would be clearer as

struct {
	s16 chans[3]; //see later, I believe this should be __le16
	aligned_s64 timestamp;
};

But that is very similar to the structure that follows immediately after this, but that has __le16 chans[]

So what is going on here?
...
The chan spec is little endian so the scan one is more correct.

So I think the fix for this issue is move the scan element to the end of struct bmc150_accel_data and mark it with __aligned(IIO_DMA_MINALIGN);

Quoted 8 lines

Then use that for both the bmc150_accel_trigger_handler() and __bmc150_accel_fifo_flush() paths. A second fix will resolve the local buffer in __bmc150_accel_fifo_flush() that is being used for a bulk regmap transfer that may need a dma safe buffer.

Please combine these two fixes (if you agree with my analysis) with a follow up to do the iio_push_to_buffers_with_ts() all in one series as they will be touching the same code.

Thanks, Jonathan.
I agree with your analysis.

On type, kept s16 because that was the originally used.
I'll keep __le16 and will merge buffer and scan together as you suggested. Will move scan to the end with __aligned(IIO_DMA_MINALIGN) and use it for both trigger handler and fifo flush.
A separate patch for the FIFO stack buffer and keep the iio_push_to_buffers_with_ts() change in the same.

Yash Suthar