Re: [PATCH v5 5/5] iio: ssp_sensors: reuse preallocated RX buffer for SPI transfers

[Sanjay Chitroda]

On 6 April 2026 9:37:49 pm IST, David Lechner <dlechner@xxxxxxxxxxxx> wrote:
>On 4/6/26 3:08 AM, Sanjay Chitroda wrote:
>> From: Sanjay Chitroda <sanjayembeddedse@xxxxxxxxx>
>> 
>> Avoid allocating a temporary DMA buffer in the interrupt context when
>> handling hub-to-AP and AP-to-hub SPI write messages.
>> 
>> Preallocate RX buffer during probe and reuse it for SPI receive
>> operations. This removes repeated kzalloc() calls from the IRQ
>> path, reduces allocation overhead, and avoids potential allocation
>> failures under memory pressure.
>> 
>> The RX buffer size is tracked and allocated using devm_kzalloc(), ensuring
>> proper lifetime management tied to the device.
>> 
>> No functional change intended; this is an internal optimization and
>> robustness improvement.
>
>If we are going to claim this is an optimization, we should have
>some measurements to back that up.
>

Hi David,

Thank you for your inputs and feedback.


>> 
>> Signed-off-by: Sanjay Chitroda <sanjayembeddedse@xxxxxxxxx>
>> ---
>> Changes in v5:
>> - Rebase change on top of latest v5 patch series.
>> Changes in v4:
>> - Use preallocated buffer and stash a buffer that gets reused each time instead of a fresh allocation.
>> - Link to v3: https://lore.kernel.org/all/20260315125509.857195-3-sanjayembedded@xxxxxxxxx/
>> Changes in v3:
>> - prepare series to have all respective cleanup API support for the ssp_sensors following input from Andy Shevchenko
>> - Link to v2 https://lore.kernel.org/all/20260311174151.3441429-1-sanjayembedded@xxxxxxxxx/
>> Changes in v2:
>> - split series to individual patch
>> - address review comment from Andy Shevchenko
>> - Link to v1 https://lore.kernel.org/all/20260310200513.2162018-3-sanjayembedded@xxxxxxxxx/
>> ---
>>  drivers/iio/common/ssp_sensors/ssp.h     |  5 +++++
>>  drivers/iio/common/ssp_sensors/ssp_dev.c | 12 ++++++++++++
>>  drivers/iio/common/ssp_sensors/ssp_spi.c | 19 +++----------------
>>  3 files changed, 20 insertions(+), 16 deletions(-)
>> 
>> diff --git a/drivers/iio/common/ssp_sensors/ssp.h b/drivers/iio/common/ssp_sensors/ssp.h
>> index f649cdecc277..aa125fd1bed5 100644
>> --- a/drivers/iio/common/ssp_sensors/ssp.h
>> +++ b/drivers/iio/common/ssp_sensors/ssp.h
>> @@ -175,6 +175,8 @@ struct ssp_sensorhub_info {
>>   * @sensor_devs:	registered IIO devices table
>>   * @enable_refcount:	enable reference count for wdt (watchdog timer)
>>   * @header_buffer:	cache aligned buffer for packet header
>> + * @rx_buf:		buffer to receive SPI data
>> + * @rx_buf_size:	allocated size of rx_buf
>>   */
>>  struct ssp_data {
>>  	struct spi_device *spi;
>> @@ -222,6 +224,9 @@ struct ssp_data {
>>  	atomic_t enable_refcount;
>>  
>>  	__le16 header_buffer[SSP_HEADER_BUFFER_SIZE / sizeof(__le16)] __aligned(IIO_DMA_MINALIGN);
>> +
>> +	u8 *rx_buf;
>> +	size_t rx_buf_size;
>
>No, these can't be after _aligned(IIO_DMA_MINALIGN); without causing problems.
>
>What would work here though is:
>
>	u8 rx_buf[SSP_DATA_PACKET_SIZE];
>

Okay, packet size is 960 bytes; to avoid using static memory i have used dynamic memory with devm API.

However following your input I will update change in next series with static allocation.

>>  };
>>  
>>  void ssp_clean_pending_list(struct ssp_data *data);
>> diff --git a/drivers/iio/common/ssp_sensors/ssp_dev.c b/drivers/iio/common/ssp_sensors/ssp_dev.c
>> index aab28f2a0f75..2a8d6f040ae4 100644
>> --- a/drivers/iio/common/ssp_sensors/ssp_dev.c
>> +++ b/drivers/iio/common/ssp_sensors/ssp_dev.c
>> @@ -516,6 +516,18 @@ static int ssp_probe(struct spi_device *spi)
>>  		goto err_setup_spi;
>>  	}
>>  
>> +	data->rx_buf_size = SSP_DATA_PACKET_SIZE;
>> +	data->rx_buf = devm_kzalloc(&spi->dev,
>> +				    data->rx_buf_size,
>> +				    GFP_KERNEL | GFP_DMA);
>
>Since this is a fixed size, we don't need a separate alloc here. We can
>just embed the array in the data struct.
>

Yes, understood.

>> +
>> +	if (!data->rx_buf) {
>> +		dev_err(&spi->dev,
>> +			"Failed to allocate memory for rx_buf\n");
>> +		ret = -ENOMEM;
>> +		goto err_setup_spi;
>> +	}
>> +
>>  	for (i = 0; i < SSP_SENSOR_MAX; ++i) {
>>  		data->delay_buf[i] = SSP_DEFAULT_POLLING_DELAY;
>>  		data->batch_latency_buf[i] = 0;
>> diff --git a/drivers/iio/common/ssp_sensors/ssp_spi.c b/drivers/iio/common/ssp_sensors/ssp_spi.c
>> index 7c1780e15acf..2f7445e8b4d1 100644
>> --- a/drivers/iio/common/ssp_sensors/ssp_spi.c
>> +++ b/drivers/iio/common/ssp_sensors/ssp_spi.c
>> @@ -383,19 +383,13 @@ int ssp_irq_msg(struct ssp_data *data)
>>  			 * but the slave should not send such ones - it is to
>>  			 * check but let's handle this
>>  			 */
>> -			buffer = kmalloc(length, GFP_KERNEL | GFP_DMA);
>> -			if (!buffer) {
>> -				ret = -ENOMEM;
>> -				goto _unlock;
>> -			}
>> +			buffer = data->rx_buf;
>
>
>I don't think it is helpful to keep the buffer local variable.
>
I will keep buffer at top of ssp_irq_msg() instead of local to specific switch case.

>>  
>>  			/* got dead packet so it is always an error */
>>  			ret = spi_read(data->spi, buffer, length);
>>  			if (ret >= 0)
>>  				ret = -EPROTO;
>>  
>> -			kfree(buffer);
>> -
>>  			dev_err(SSP_DEV, "No match error %x\n",
>>  				msg_options);
>>  
>> @@ -428,22 +422,15 @@ int ssp_irq_msg(struct ssp_data *data)
>>  		mutex_unlock(&data->pending_lock);
>>  		break;
>>  	case SSP_HUB2AP_WRITE:
>> -		buffer = kzalloc(length, GFP_KERNEL | GFP_DMA);
>> -		if (!buffer)
>> -			return -ENOMEM;
>> +		buffer = data->rx_buf;
>>  
>>  		ret = spi_read(data->spi, buffer, length);
>>  		if (ret < 0) {
>>  			dev_err(SSP_DEV, "spi read fail\n");
>> -			kfree(buffer);
>>  			break;
>>  		}
>>  
>> -		ret = ssp_parse_dataframe(data, buffer, length);
>> -
>> -		kfree(buffer);
>> -		break;
>> -
>> +		return ssp_parse_dataframe(data, buffer, length);
>>  	default:
>>  		dev_err(SSP_DEV, "unknown msg type\n");
>>  		return -EPROTO;
>