iio: accel: bmc150: DMA-safe buffers and use iio_push_to_buffers_with_ts()
v3 cover iio: accel: bmc150: DMA-safe buffers and use iio_push_to_buffers_with_ts()
The trigger path used a plain s16 buffer while the FIFO path already had a properly typed scan member (__le16 + aligned timestamp). Those two buffers did the same job, and s16 does not represent little-endian data. Drop the duplicate, keep scan at the end of struct bmc150_accel_data with IIO_DMA_MINALIGN, and use it for both the trigger and FIFO paths.
The FIFO and read_raw buffers are passed to regmap bulk/raw reads, which is not DMA-safe. Move them into struct bmc150_accel_data after the aligned scan buffer. Because all buffers are serialized by data->mutex, placing them after scan ensures DMA safety without needing separate alignment.
Finally, replace the deprecated iio_push_to_buffers_with_timestamp() with iio_push_to_buffers_with_ts() so the push sites pass an explicit buffer size.
This supersedes the earlier standalone patches:
Link to v1: https://lore.kernel.org/all/[email protected]/
Link to v2: https://lore.kernel.org/linux-iio/[email protected]/
Changes in v3:
- patch 1: hold mutex for both the read and push to guarantee DMA safety
- patch 1: drop unused AXIS_MAX and use sizeof(data->scan.channels)
- patch 2: remove redundant __aligned(IIO_DMA_MINALIGN) from fifo_buff and regval, as the mutex serializes access
- patch 2: fix comma typo in commit message
Changes in v2:
- patch 1: restore early error check flow
- patch 2: keep fifo_buff declaration in one line
Yash Suthar (3):
iio: accel: bmc150: use aligned scan buffer for both trigger and fifo
iio: accel: bmc150: use DMA-safe buffers for regmap bulk reads
iio: accel: bmc150: use iio_push_to_buffers_with_ts()
drivers/iio/accel/bmc150-accel-core.c | 28 +++++++++++++-------------- drivers/iio/accel/bmc150-accel.h | 19 +++++++++--------- 2 files changed, 23 insertions(+), 24 deletions(-)
Signature · 1 line
2.43.0
PATCH v3 1/3 iio: accel: bmc150: use aligned scan buffer for both trigger and fifo
Drop buffer as duplicate and s16 is not correct we needed __le16 as little endian, keep scan at the end of the bmc150_accel_data struct with IIO_DMA_MINALIGN, and use it for both paths.
In trigger handle, replaced AXIS_MAX with sizeof(data->scan.channels). Additionally, ensure mutex is held across both read and push to get DMA safety for the scan buffer.
--- drivers/iio/accel/bmc150-accel-core.c | 8 ++++---- drivers/iio/accel/bmc150-accel.h | 14 +++++--------- 2 files changed, 9 insertions(+), 13 deletions(-)
drivers/iio/accel/bmc150-accel-core.c +4 -4
@@ -131,7 +131,6 @@ enum bmc150_accel_axis {AXIS_X ,AXIS_Y , AXIS_Z, - AXIS_MAX, }; enum bmc150_power_modes {
@@ -1191,14 +1190,15 @@ static irqreturn_t bmc150_accel_trigger_handler(int irq, void *p)mutex_lock (&data->mutex); ret =regmap_bulk_read (data->regmap,BMC150_ACCEL_REG_XOUT_L , - data->buffer,AXIS_MAX *2 ); -mutex_unlock (&data->mutex); + data->scan.channels, +sizeof (data->scan.channels));if (ret <0 ) goto err_read; -iio_push_to_buffers_with_timestamp (indio_dev, data->buffer, +iio_push_to_buffers_with_timestamp (indio_dev, &data->scan, pf->timestamp); err_read: +mutex_unlock (&data->mutex);iio_trigger_notify_done (indio_dev->trig);return IRQ_HANDLED ;
drivers/iio/accel/bmc150-accel.h +5 -9
@@ -64,15 +64,6 @@ struct bmc150_accel_data {struct bmc150_accel_trigger triggers[BMC150_ACCEL_TRIGGERS ];struct mutex mutex;u8 fifo_mode, watermark; -s16 buffer[8 ]; -/* -* Ensure there is sufficient space and correct alignment for -* the timestamp if enabled -*/ -struct { -__le16 channels[3 ]; -aligned_s64 ts; - } scan;u8 bw_bits;u32 slope_dur;u32 slope_thres;
@@ -85,6 +76,11 @@ struct bmc150_accel_data {void (*resume_callback)(struct device *dev);struct delayed_work resume_work;struct iio_mount_matrix orientation; +/* Ensure correct alignment of timestamp and DMA safety */ +struct { +__le16 channels[3 ]; +aligned_s64 ts; + } scan__aligned(IIO_DMA_MINALIGN) ; }; int bmc150_accel_core_probe(struct device *dev, struct regmap *regmap, int irq,
Signature · 1 line
2.43.0
Re: [PATCH v3 1/3] iio: accel: bmc150: use aligned scan buffer for both trigger and fifo
Quoted 7 lines · Yash Suthar
Drop buffer as duplicate and s16 is not correct we needed
__le16as little endian, keep scan at the end of thebmc150_accel_datastruct withIIO_DMA_MINALIGN, and use it for both paths.In trigger handle, replaced
AXIS_MAXwithsizeof(data->scan.channels). Additionally, ensure mutex is held across both read and push to get DMA safety for the scan buffer.
...
drivers/iio/accel/bmc150-accel-core.c static irqreturn_t bmc150_accel_trigger_handler(int irq, void *p)
Quoted 7 lines
drivers/iio/accel/bmc150-accel-core.c +2
-2mutex_lock (&data->mutex); ret =regmap_bulk_read (data->regmap,BMC150_ACCEL_REG_XOUT_L , - data->buffer,AXIS_MAX *2 ); -mutex_unlock (&data->mutex); + data->scan.channels, +sizeof (data->scan.channels));
I would put this on a single line (it is 81 characters only).
drivers/iio/accel/bmc150-accel-core.c static irqreturn_t bmc150_accel_trigger_handler(int irq, void *p) · line 1199
Quoted 6 lines
drivers/iio/accel/bmc150-accel-core.c +1
-1if (ret <0 ) goto err_read; -iio_push_to_buffers_with_timestamp (indio_dev, data->buffer, +iio_push_to_buffers_with_timestamp (indio_dev, &data->scan, pf->timestamp);
I would put this on a single line (it is 82 characters only).
Quoted 3 lines
err_read:drivers/iio/accel/bmc150-accel-core.c +1
-0+mutex_unlock (&data->mutex);iio_trigger_notify_done (indio_dev->trig);
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH v3 2/3 iio: accel: bmc150: use DMA-safe buffers for regmap bulk reads
In reply to Yash SutharThe FIFO and read_raw buffers are passed to regmap bulk/raw reads, which is not DMA-safe. Moved them into struct bmc150_accel_data after scan, each buffer are protected by mutex, placing them after ensure DMA safety.
--- drivers/iio/accel/bmc150-accel-core.c | 14 ++++++-------- drivers/iio/accel/bmc150-accel.h | 5 +++++ 2 files changed, 11 insertions(+), 8 deletions(-)
drivers/iio/accel/bmc150-accel-core.c +6 -8
@@ -125,7 +125,6 @@#define BMC150_ACCEL_REG_FIFO_CONFIG0 0x30#define BMC150_ACCEL_REG_FIFO_CONFIG1 0x3E#define BMC150_ACCEL_REG_FIFO_DATA 0x3F -#define BMC150_ACCEL_FIFO_LENGTH 32enum bmc150_accel_axis {AXIS_X ,
@@ -621,7 +620,6 @@ static int bmc150_accel_get_axis(struct bmc150_accel_data *data,struct device *dev =regmap_get_device (data->regmap);int ret;int axis = chan->scan_index; -__le16 raw_val;mutex_lock (&data->mutex); ret =bmc150_accel_set_power_state (data, true);
@@ -631,14 +629,14 @@ static int bmc150_accel_get_axis(struct bmc150_accel_data *data, } ret =regmap_bulk_read (data->regmap,BMC150_ACCEL_AXIS_TO_REG (axis), - &raw_val,sizeof (raw_val)); + &data->regval,sizeof (data->regval));if (ret <0 ) {dev_err (dev,"Error reading axis %d\n" , axis);bmc150_accel_set_power_state (data, false);mutex_unlock (&data->mutex);return ret; } - *val =sign_extend32 (le16_to_cpu (raw_val) >> chan->scan_type.shift, + *val =sign_extend32 (le16_to_cpu (data->regval) >> chan->scan_type.shift, chan->scan_type.realbits -1 ); ret =bmc150_accel_set_power_state (data, false);mutex_unlock (&data->mutex);
@@ -917,7 +915,7 @@ static int bmc150_accel_set_watermark(struct iio_dev *indio_dev, unsigned val)* frame data is discarded. */ static int bmc150_accel_fifo_transfer (struct bmc150_accel_data *data, -char *buffer,int samples) +void *buffer,int samples) {struct device *dev =regmap_get_device (data->regmap);int sample_length =3 *2 ;
@@ -940,7 +938,6 @@ static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev,struct device *dev =regmap_get_device (data->regmap);int ret, i;u8 count; -u16 buffer[BMC150_ACCEL_FIFO_LENGTH *3 ];int64_t tstamp;uint64_t sample_period;unsigned int val;
@@ -992,7 +989,7 @@ static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, count =min_t (u8, count,BMC150_ACCEL_FIFO_LENGTH ); - ret =bmc150_accel_fifo_transfer (data, (u8 *)buffer, count); + ret =bmc150_accel_fifo_transfer (data, data->fifo_buff, count);if (ret)return ret;
@@ -1007,7 +1004,8 @@ static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, j =0 ;iio_for_each_active_channel (indio_dev, bit) -memcpy (&data->scan.channels[j++], &buffer[i *3 + bit], +memcpy (&data->scan.channels[j++], + &data->fifo_buff[i *3 + bit],sizeof (data->scan.channels[0 ])); iio_push_to_buffers_with_timestamp(indio_dev, &data->scan,
drivers/iio/accel/bmc150-accel.h +5 -0
@@ -56,6 +56,8 @@ enum bmc150_accel_trigger_id { BMC150_ACCEL_TRIGGERS, }; +#define BMC150_ACCEL_FIFO_LENGTH 32 +struct bmc150_accel_data {struct regmap *regmap;int irq;
@@ -81,6 +83,9 @@ struct bmc150_accel_data {__le16 channels[3 ];aligned_s64 ts; } scan__aligned(IIO_DMA_MINALIGN) ; +/* DMA-safe buffers for bulk/raw reads */ +__le16 fifo_buff[BMC150_ACCEL_FIFO_LENGTH *3 ]; +__le16 regval; }; int bmc150_accel_core_probe(struct device *dev, struct regmap *regmap, int irq,
Signature · 1 line
2.43.0
Re: [PATCH v3 2/3] iio: accel: bmc150: use DMA-safe buffers for regmap bulk reads
Quoted 4 lines · Yash Suthar
The FIFO and
read_rawbuffers are passed to regmap bulk/raw reads, which is not DMA-safe. Moved them into structbmc150_accel_dataafter scan, each buffer are protected by mutex, placing them after ensure DMA safety.
...
drivers/iio/accel/bmc150-accel-core.c static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, · line 1009
Quoted 6 lines
drivers/iio/accel/bmc150-accel-core.c +2
-1j =0 ;iio_for_each_active_channel (indio_dev, bit) -memcpy (&data->scan.channels[j++], &buffer[i *3 + bit], +memcpy (&data->scan.channels[j++], + &data->fifo_buff[i *3 + bit],sizeof (data->scan.channels[0 ]));
While this works, I find a bit confusing usage of [j++] and [0] together.
IIUC the below is an equivalent
iio_for_each_active_channel(indio_dev, bit) {
__le16 *val = &data->scan.channels[j++];
memcpy(val, &data->fifo_buff[i * 3 + bit], sizeof(val));
}and if go further we may use asm/byteorder.h macros instead of plain memcpy().
...
Maybe this is for a separate change, I dunno. I leave it to others to decide.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Jonathan Cameron
maintainerRe: [PATCH v3 2/3] iio: accel: bmc150: use DMA-safe buffers for regmap bulk reads
Quoted 17 lines · On Thu, 27 Aug 2026 10:05:19 +0300 Andy Shevchenko…
Quoted 4 lines · Yash Suthar
The FIFO and
read_rawbuffers are passed to regmap bulk/raw reads, which is not DMA-safe. Moved them into structbmc150_accel_dataafter scan, each buffer are protected by mutex, placing them after ensure DMA safety....
drivers/iio/accel/bmc150-accel-core.c static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, · line 1009
Quoted 6 lines
drivers/iio/accel/bmc150-accel-core.c +2
-1j =0 ;iio_for_each_active_channel (indio_dev, bit) -memcpy (&data->scan.channels[j++], &buffer[i *3 + bit], +memcpy (&data->scan.channels[j++], + &data->fifo_buff[i *3 + bit],sizeof (data->scan.channels[0 ]));While this works, I find a bit confusing usage of [j++] and [0] together.
Quoted 7 lines
IIUC the below is an equivalent
iio_for_each_active_channel(indio_dev, bit) { __le16 *val = &data->scan.channels[j++]; memcpy(val, &data->fifo_buff[i * 3 + bit], sizeof(val)); }
This bit I fully agree with. Much more readable.
Quoted 3 lines
and if go further we may use
asm/byteorder.hmacros instead of plainmemcpy().
Here I'm lost. We are just moving little endian values around.
For that matter if it is a single value, why can't see just assign it?
iio_for_each_active_channel(indio_dev, bit) data->scan.channels[j++] = data->fifo_buf[i * 3 + bit];
Quoted 4 lines
...
Maybe this is for a separate change, I dunno. I leave it to others to decide.
Re: [PATCH v3 2/3] iio: accel: bmc150: use DMA-safe buffers for regmap bulk reads
Quoted 3 lines · Jonathan Cameron
Quoted 1 line · On Thu, 27 Aug 2026 10:05:19 +0300 Andy Shevchenko…
On Tue, Aug 25, 2026 at 02:21:55PM +0530, Yash Suthar wrote:
...
Quoted 19 lines
Quoted 9 lines
drivers/iio/accel/bmc150-accel-core.c static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, · line 1009
Quoted 6 lines
drivers/iio/accel/bmc150-accel-core.c +2
-1j =0 ;iio_for_each_active_channel (indio_dev, bit) -memcpy (&data->scan.channels[j++], &buffer[i *3 + bit], +memcpy (&data->scan.channels[j++], + &data->fifo_buff[i *3 + bit],sizeof (data->scan.channels[0 ]));While this works, I find a bit confusing usage of [j++] and [0] together.
Quoted 7 lines
IIUC the below is an equivalent
iio_for_each_active_channel(indio_dev, bit) { __le16 *val = &data->scan.channels[j++]; memcpy(val, &data->fifo_buff[i * 3 + bit], sizeof(val)); }This bit I fully agree with. Much more readable.
Quoted 1 line
Quoted 8 lines
Quoted 2 lines
and if go further we may use
asm/byteorder.hmacros instead of plainmemcpy().Here I'm lost. We are just moving little endian values around.
For that matter if it is a single value, why can't see just assign it?
iio_for_each_active_channel(indio_dev, bit) data->scan.channels[j++] = data->fifo_buf[i * 3 + bit];
It depends on the alignment. I haven't checked the assignee and the source for that. That's why I mentioned byteorder.h (and implying also unaligned.h) for the cases like put_unaligned_le16(le16_to_cpup(val)). If both sides are aligned, then yours suggestion will work.
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH v3 3/3 iio: accel: bmc150: use iio_push_to_buffers_with_ts()
In reply to Yash SutharReplace deprecated iio_push_to_buffers_with_timestamp() with iio_push_to_buffers_with_ts() to allow source size runtime checks.
--- drivers/iio/accel/bmc150-accel-core.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/iio/accel/bmc150-accel-core.c +4 -4
@@ -1008,8 +1008,8 @@ static int __bmc150_accel_fifo_flush(struct iio_dev *indio_dev, &data->fifo_buff[i *3 + bit], sizeof(data->scan.channels[0])); -iio_push_to_buffers_with_timestamp (indio_dev, &data->scan, - tstamp); +iio_push_to_buffers_with_ts (indio_dev, &data->scan, +sizeof (data->scan), tstamp); tstamp += sample_period; }
@@ -1193,8 +1193,8 @@ static irqreturn_t bmc150_accel_trigger_handler(int irq, void *p)if (ret <0 ) goto err_read; -iio_push_to_buffers_with_timestamp (indio_dev, &data->scan, - pf->timestamp); +iio_push_to_buffers_with_ts (indio_dev, &data->scan, +sizeof (data->scan), pf->timestamp); err_read:mutex_unlock (&data->mutex);iio_trigger_notify_done (indio_dev->trig);
Signature · 1 line
2.43.0
Re: [PATCH v3 3/3] iio: accel: bmc150: use iio_push_to_buffers_with_ts()
Quoted 2 lines · Yash Suthar
Replace deprecated
iio_push_to_buffers_with_timestamp()withiio_push_to_buffers_with_ts()to allow source size runtime checks.
With this you might want to use IIO_DECLARE_DMA_BUFFER_WITH_TS() in the definition of the scan buffer.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Jonathan Cameron
maintainerRe: [PATCH v3 3/3] iio: accel: bmc150: use iio_push_to_buffers_with_ts()
On Thu, 27 Aug 2026 10:08:07 +0300
Andy Shevchenko <[email protected]> wrote:
Hi Andy,
Quoted 7 lines · Yash Suthar
Quoted 2 lines
Replace deprecated
iio_push_to_buffers_with_timestamp()withiio_push_to_buffers_with_ts()to allow source size runtime checks.With this you might want to use
IIO_DECLARE_DMA_BUFFER_WITH_TS()in the definition of the scan buffer.
Why? The structure used is:
struct { __le16 channels[3];
aligned_s64 ts;
} scan __aligned(IIO_DMA_MINALIGN);
So given we never let things get enabled with just timestamp, the timestamp and channels always end up in the places the structure describes.
Where that is the case, I much prefer this to the magic buffer with enough elements of the channel type on the end.
Jonathan
Re: [PATCH v3 3/3] iio: accel: bmc150: use iio_push_to_buffers_with_ts()
Quoted 7 lines · Jonathan Cameron
Quoted 5 lines · On Thu, 27 Aug 2026 10:08:07 +0300 Andy Shevchenko…
Quoted 2 lines · Yash Suthar
Replace deprecated
iio_push_to_buffers_with_timestamp()withiio_push_to_buffers_with_ts()to allow source size runtime checks.With this you might want to use
IIO_DECLARE_DMA_BUFFER_WITH_TS()in the
(For the record, I put it as "might".)
Quoted 10 lines
Quoted 2 lines
definition of the scan buffer.
Why? The structure used is:
struct {(quoted diff) +0
-0__le16 channels[3]; aligned_s64 ts; } scan __aligned(IIO_DMA_MINALIGN);So given we never let things get enabled with just timestamp, the timestamp and channels always end up in the places the structure describes.
OK! I personally consider the macro a bit more robust in the cases when we _might_ have new HW to support where it will be the case.
Quoted 2 lines
Where that is the case, I much prefer this to the magic buffer with enough elements of the channel type on the end.
No objections.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Jonathan Cameron
maintainerRe: [PATCH v3 3/3] iio: accel: bmc150: use iio_push_to_buffers_with_ts()
Quoted 19 lines · On Mon, 31 Aug 2026 10:00:47 +0300 Andy Shevchenko…
Quoted 7 lines · Jonathan Cameron
Quoted 5 lines · On Thu, 27 Aug 2026 10:08:07 +0300 Andy Shevchenko…
Quoted 2 lines · Yash Suthar
Replace deprecated
iio_push_to_buffers_with_timestamp()withiio_push_to_buffers_with_ts()to allow source size runtime checks.With this you might want to use
IIO_DECLARE_DMA_BUFFER_WITH_TS()in the(For the record, I put it as "might".)
Quoted 8 lines
Quoted 2 lines
definition of the scan buffer.
Why? The structure used is:
struct {(quoted diff) +0
-0__le16 channels[3]; aligned_s64 ts; } scan __aligned(IIO_DMA_MINALIGN);
Quoted 5 lines
Quoted 2 lines
So given we never let things get enabled with just timestamp, the timestamp and channels always end up in the places the structure describes.
OK! I personally consider the macro a bit more robust in the cases when we _might_ have new HW to support where it will be the case.
Ah, I kind of get that but doesn't worry me too much.
They will need to modify the structure and then we'll spot it. It isn't a source of bugs if it does go wrong, just inelegant as the structure kind of implies the timestamp is in a different place from where ends up.
Maybe this is one for a maintainers profile that hopefully the AI bots will then use to monitor this for us ;)
Jonathan
Quoted 6 lines
Quoted 3 lines
Where that is the case, I much prefer this to the magic buffer with enough elements of the channel type on the end.
No objections.
Quoted 11 lines · Yash Suthar
The trigger path used a plain s16 buffer while the FIFO path already had a properly typed scan member (
__le16+ aligned timestamp). Those two buffers did the same job, and s16 does not represent little-endian data. Drop the duplicate, keep scan at the end of structbmc150_accel_datawithIIO_DMA_MINALIGN, and use it for both the trigger and FIFO paths.The FIFO and
read_rawbuffers are passed to regmap bulk/raw reads, which is not DMA-safe. Move them into structbmc150_accel_dataafter the aligned scan buffer. Because all buffers are serialized bydata->mutex, placing them after scan ensures DMA safety without needing separate alignment.
Quoted 7 lines
Finally, replace the deprecated
iio_push_to_buffers_with_timestamp()withiio_push_to_buffers_with_ts()so the push sites pass an explicit buffer size.This supersedes the earlier standalone patches:
In this case it's better not to use Link: tag. b4 can't distinguish them from the important ones.
Quoted 2 lines
Link to v1: https://lore.kernel.org/all/[email protected]/
Link to v2: https://lore.kernel.org/linux-iio/[email protected]/
Signature · 2 lines
With Best Regards, Andy Shevchenko