Re: [PATCH v4 4/5] firmware: arm_scmi: Create debugfs files for counts

[Cristian Marussi <cristian.marussi@arm.com>]

On Wed, Jul 31, 2024 at 01:11:22PM +0100, Sudeep Holla wrote:
> On Tue, Jul 30, 2024 at 10:33:41AM +0100, Luke Parkin wrote:
> > Create debugfs files for the metrics in the debug_counters array
> > 
> > Signed-off-by: Luke Parkin <luke.parkin@arm.com>
> > ---
> > v3->v4
> > Use new locations for debug array
> > Use counter instead of stats
> > v2->v3
> > Add extra statistics also added in v3
> > v1->v2
> > Only create stats pointer if stats are enabled
> > Move stats debugfs creation into a seperate helper function
> > ---
> >  drivers/firmware/arm_scmi/driver.c | 38 ++++++++++++++++++++++++++++++
> >  1 file changed, 38 insertions(+)
> > 
> > diff --git a/drivers/firmware/arm_scmi/driver.c b/drivers/firmware/arm_scmi/driver.c
> > index 5acd3d324def..ec6434692d1a 100644
> > --- a/drivers/firmware/arm_scmi/driver.c
> > +++ b/drivers/firmware/arm_scmi/driver.c
> > @@ -2847,6 +2847,41 @@ static int scmi_device_request_notifier(struct notifier_block *nb,
> >  	return NOTIFY_OK;
> >  }
> >  
> > +static void scmi_debugfs_counters_setup(struct scmi_debug_info *dbg,
> > +					struct dentry *trans)
> > +{
> > +	struct dentry *counters;
> > +
> > +	counters = debugfs_create_dir("counters", trans);
> > +
> > +	debugfs_create_atomic_t("sent_ok", 0400, counters,
> > +				&dbg->counters[SENT_OK]);
> > +	debugfs_create_atomic_t("sent_fail", 0400, counters,
> > +				&dbg->counters[SENT_FAIL]);
> > +	debugfs_create_atomic_t("sent_fail_polling_unsupported", 0400, counters,
> > +				&dbg->counters[SENT_FAIL_POLLING_UNSUPPORTED]);
> > +	debugfs_create_atomic_t("sent_fail_channel_not_found", 0400, counters,
> > +				&dbg->counters[SENT_FAIL_CHANNEL_NOT_FOUND]);
> > +	debugfs_create_atomic_t("response_ok", 0400, counters,
> > +				&dbg->counters[RESPONSE_OK]);
> > +	debugfs_create_atomic_t("notif_ok", 0400, counters,
> > +				&dbg->counters[NOTIF_OK]);
> > +	debugfs_create_atomic_t("dlyd_resp_ok", 0400, counters,
> > +				&dbg->counters[DLYD_RESPONSE_OK]);
> > +	debugfs_create_atomic_t("xfers_resp_timeout", 0400, counters,
> > +				&dbg->counters[XFERS_RESPONSE_TIMEOUT]);
> > +	debugfs_create_atomic_t("response_polled_ok", 0400, counters,
> > +				&dbg->counters[RESPONSE_POLLED_OK]);
> > +	debugfs_create_atomic_t("err_msg_unexpected", 0400, counters,
> > +				&dbg->counters[ERR_MSG_UNEXPECTED]);
> > +	debugfs_create_atomic_t("err_msg_invalid", 0400, counters,
> > +				&dbg->counters[ERR_MSG_INVALID]);
> > +	debugfs_create_atomic_t("err_msg_nomem", 0400, counters,
> > +				&dbg->counters[ERR_MSG_NOMEM]);
> > +	debugfs_create_atomic_t("err_protocol", 0400, counters,
> > +				&dbg->counters[ERR_PROTOCOL]);
> 
> Just curious and wonder if we can keep all these read-only and have another
> debugfs file which is write only to just reset the counters ?
> 
> Cristian,
> 
> Thoughts ? Or am I missing somthing ?

Do you mean creating a bunch of additional reset_sent_ok
reset_<your_counters> that are just WO and used to reset each single
counter ?
(...I suppose because a global WO reset-all is already there...)

...I think it should be doable easily...but with a lot of duplication so
my next questions would be why ? what would be the gain ?
is it worth the effort being debug feats ?

Thanks,
Cristian