Re: [PATCH] counter: microchip-tcb-capture: Add watch validation support

[William Breathitt Gray <wbg@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 3540 bytes --]

On Thu, May 15, 2025 at 10:28:25AM +0530, Dharma Balasubiramani wrote:
> Introduce a watch validation callback to restrict supported event and
> channel combinations. This allows userspace to receive notifications only
> for valid event types and sources. Specifically, enable the following
> supported events on channels RA, RB, and RC:
> 
>   - COUNTER_EVENT_CAPTURE
>   - COUNTER_EVENT_CHANGE_OF_STATE
>   - COUNTER_EVENT_OVERFLOW
>   - COUNTER_EVENT_THRESHOLD
> 
> Signed-off-by: Dharma Balasubiramani <dharma.b@microchip.com>
> ---
>  drivers/counter/microchip-tcb-capture.c | 28 +++++++++++++++++++++++++---
>  1 file changed, 25 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/counter/microchip-tcb-capture.c b/drivers/counter/microchip-tcb-capture.c
> index 1de3c50b9804..179ff5595143 100644
> --- a/drivers/counter/microchip-tcb-capture.c
> +++ b/drivers/counter/microchip-tcb-capture.c
> @@ -337,6 +337,27 @@ static struct counter_comp mchp_tc_count_ext[] = {
>  	COUNTER_COMP_COMPARE(mchp_tc_count_compare_read, mchp_tc_count_compare_write),
>  };
> 
> +static int mchp_tc_watch_validate(struct counter_device *counter,
> +				  const struct counter_watch *watch)
> +{
> +	switch (watch->channel) {
> +	case COUNTER_MCHP_EVCHN_RA:
> +	case COUNTER_MCHP_EVCHN_RB:
> +	case COUNTER_MCHP_EVCHN_RC:

Hello Dharma,

Include COUNTER_MCHP_EVCHN_CV as well for the sake of completeness. I
know COUNTER_MCHP_EVCHN_CV and COUNTER_MCHP_EVCHN_RA have the same
underlying channel id, but we're abstracting this fact so it's good to
maintain the consistency of the abstraction across all callbacks.

> +		switch (watch->event) {
> +		case COUNTER_EVENT_CAPTURE:
> +		case COUNTER_EVENT_CHANGE_OF_STATE:
> +		case COUNTER_EVENT_OVERFLOW:
> +		case COUNTER_EVENT_THRESHOLD:
> +			return 0;

The watch_validate callback is used to ensure that the requested watch
configuration is valid: i.e. the watch event is appropriate for the
watch channel. 

Looking at include/uapi/linux/counter/microchip-tcb-capture.h:

     * Channel 0:
     * - CV register changed
     * - CV overflowed
     * - RA captured
     * Channel 1:
     * - RB captured
     * Channel 2:
     * - RC compare triggered

If I'm understanding correctly, channel 0 supports only the
CHANGE_OF_STATE, OVERFLOW, and CAPTURE events; channel 1 supports only
CAPTURE events; and channel 2 supports only THRESHOLD events.

Adjust the code to ensure those limitations.

> +		default:
> +			return -EINVAL;
> +		}
> +	default:
> +		return -EINVAL;
> +	}
> +}
> +
>  static struct counter_count mchp_tc_counts[] = {
>  	{
>  		.id = 0,
> @@ -351,12 +372,13 @@ static struct counter_count mchp_tc_counts[] = {
>  };
> 
>  static const struct counter_ops mchp_tc_ops = {
> -	.signal_read    = mchp_tc_count_signal_read,
> +	.action_read    = mchp_tc_count_action_read,
> +	.action_write   = mchp_tc_count_action_write,
>  	.count_read     = mchp_tc_count_read,
>  	.function_read  = mchp_tc_count_function_read,
>  	.function_write = mchp_tc_count_function_write,
> -	.action_read    = mchp_tc_count_action_read,
> -	.action_write   = mchp_tc_count_action_write
> +	.signal_read    = mchp_tc_count_signal_read,

It's nice to alphabetize the counter_ops callbacks, but it's also
unrelated to the watch_validate implementation. Move the alphabetization
cleanup to a separate patch so that this patch remains dedicated to
just watch_validate changes.

Thanks,

William Breathitt Gray

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]