[PATCH v2] iio: adc: nxp-sar-adc: harden buffer ISR against per-channel read failure

[PATCH v2] iio: adc: nxp-sar-adc: harden buffer ISR against per-channel read failure

[Stepan Ionichev]

nxp_sar_adc_isr_buffer() bails on the first channel-read failure
without calling iio_trigger_notify_done(), so the trigger use_count
is left incremented and iio_trigger_poll_chained() drops subsequent
dispatches until the device is rebound.

Reaching this path means a state machine has gone wrong (driver bug
or the SAR ADC in an unexpected state) rather than a transient bus
issue, so this is hardening rather than a bug fix. If the underlying
condition persists the device is wedged and needs an unbind anyway.

Call iio_trigger_notify_done() on the error exit too, matching the
success path. The nxp_sar_adc_read_notify() duplication is intentional
and avoids a goto label for a two-line bail-out, as suggested by David.

Signed-off-by: Stepan Ionichev <sozdayvek@gmail.com>
---
v2:
- Drop Fixes: and Cc: stable@; this is a state-machine hardening, not
  a user-visible bug fix (Jonathan)
- Reframe the commit message and explain why the failure is not a bus
  issue
- Inline the iio_trigger_notify_done() in the error exit instead of a
  goto label, accepting the small nxp_sar_adc_read_notify() duplication
  (David)

v1: https://lore.kernel.org/all/20260517162346.189-1-sozdayvek@gmail.com/

 drivers/iio/adc/nxp-sar-adc.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/iio/adc/nxp-sar-adc.c b/drivers/iio/adc/nxp-sar-adc.c
index 9d9f2c76b..1fc839a15 100644
--- a/drivers/iio/adc/nxp-sar-adc.c
+++ b/drivers/iio/adc/nxp-sar-adc.c
@@ -341,6 +341,7 @@ static void nxp_sar_adc_isr_buffer(struct iio_dev *indio_dev)
 		ret = nxp_sar_adc_read_data(info, info->buffered_chan[i]);
 		if (ret < 0) {
 			nxp_sar_adc_read_notify(info);
+			iio_trigger_notify_done(indio_dev->trig);
 			return;
 		}
 
-- 
2.43.0

Re: [PATCH v2] iio: adc: nxp-sar-adc: harden buffer ISR against per-channel read failure

[Jonathan Cameron]

On Mon, 18 May 2026 23:11:38 +0500
Stepan Ionichev <sozdayvek@gmail.com> wrote:

> nxp_sar_adc_isr_buffer() bails on the first channel-read failure
> without calling iio_trigger_notify_done(), so the trigger use_count
> is left incremented and iio_trigger_poll_chained() drops subsequent
> dispatches until the device is rebound.
> 
> Reaching this path means a state machine has gone wrong (driver bug
> or the SAR ADC in an unexpected state) rather than a transient bus
> issue, so this is hardening rather than a bug fix. If the underlying
> condition persists the device is wedged and needs an unbind anyway.
> 
> Call iio_trigger_notify_done() on the error exit too, matching the
> success path. The nxp_sar_adc_read_notify() duplication is intentional
> and avoids a goto label for a two-line bail-out, as suggested by David.
> 
> Signed-off-by: Stepan Ionichev <sozdayvek@gmail.com>
Huh. Seems I picked this up a while ago but neither replied to this thread
nor marked it so in patchwork.

Anyhow, long applied.

Thanks,

Jonathan

> ---
> v2:
> - Drop Fixes: and Cc: stable@; this is a state-machine hardening, not
>   a user-visible bug fix (Jonathan)
> - Reframe the commit message and explain why the failure is not a bus
>   issue
> - Inline the iio_trigger_notify_done() in the error exit instead of a
>   goto label, accepting the small nxp_sar_adc_read_notify() duplication
>   (David)
> 
> v1: https://lore.kernel.org/all/20260517162346.189-1-sozdayvek@gmail.com/
> 
>  drivers/iio/adc/nxp-sar-adc.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/iio/adc/nxp-sar-adc.c b/drivers/iio/adc/nxp-sar-adc.c
> index 9d9f2c76b..1fc839a15 100644
> --- a/drivers/iio/adc/nxp-sar-adc.c
> +++ b/drivers/iio/adc/nxp-sar-adc.c
> @@ -341,6 +341,7 @@ static void nxp_sar_adc_isr_buffer(struct iio_dev *indio_dev)
>  		ret = nxp_sar_adc_read_data(info, info->buffered_chan[i]);
>  		if (ret < 0) {
>  			nxp_sar_adc_read_notify(info);
> +			iio_trigger_notify_done(indio_dev->trig);
>  			return;
>  		}
>