Hi Steve,
(Writing from my @kernel.org email to ensure you receive it)
On 14/12/2017 at 12:49:12 -0500, Steven Rostedt wrote:
> On Thu, 14 Dec 2017 13:31:43 +0800
> Baolin Wang <baolin.wang@xxxxxxxxxx> wrote:
>
>
> > @@ -53,6 +56,8 @@ int rtc_read_time(struct rtc_device *rtc, struct rtc_time *tm)
> >
> > err = __rtc_read_time(rtc, tm);
> > mutex_unlock(&rtc->ops_lock);
> > +
> > + trace_rtc_read_time(rtc_tm_to_time64(tm), err);
>
> There's a possibility that gcc may put the call to rt_tm_to_time64()
> outside the tracepoint area that gets nop'd out. Could you just pass in
> the tm to the tracepoint, and have the call to rtc_tm_to_time64() done
> within the TP_fast_assign? This would guarantee that we don't incur
> overhead when tracing is off.
>
Actually, I've asked for that as rtc_time64_to_tm is really heavier than
rtc_tm_to_time64 and both set_time and set_alarm will soon have a call
to rtc_tm_to_time64 anyway. read_time and read_alarm will probably end
up having one in the long term too.
> > diff --git a/include/trace/events/rtc.h b/include/trace/events/rtc.h
> > new file mode 100644
> > index 0000000..621333f
> > --- /dev/null
> > +++ b/include/trace/events/rtc.h
> > @@ -0,0 +1,206 @@
> > +#undef TRACE_SYSTEM
> > +#define TRACE_SYSTEM rtc
> > +
> > +#if !defined(_TRACE_RTC_H) || defined(TRACE_HEADER_MULTI_READ)
> > +#define _TRACE_RTC_H
> > +
> > +#include <linux/rtc.h>
> > +#include <linux/tracepoint.h>
> > +
> > +DECLARE_EVENT_CLASS(rtc_time_alarm_class,
> > +
> > + TP_PROTO(time64_t secs, int err),
> > +
> > + TP_ARGS(secs, err),
> > +
> > + TP_STRUCT__entry(
> > + __field(time64_t, secs)
> > + __field(int, err)
> > + ),
> > +
> > + TP_fast_assign(
> > + __entry->secs = secs;
> > + __entry->err = err;
> > + ),
> > +
> > + TP_printk("UTC (%lld) (%d)",
> > + __entry->secs, __entry->err
Can't TP_printk do the rtc_time64_to_tm conversion to pretty print the
date? Or do we have to implement it in vsprintf?
--
Alexandre Belloni
