Re: [PATCH 1/6] trace: add tracepoints to timekeeping code - xtime changes

[David Ahern <dsahern@gmail.com>]

Thomas/Peter:

On 06/17/2011 08:13 AM, David Ahern wrote:
> 
> 
> On 06/17/2011 07:23 AM, Frederic Weisbecker wrote:
>> On Tue, Jun 07, 2011 at 05:55:03PM -0600, David Ahern wrote:
>>> Trace points in timekeeping.c where xtime is modified by a user
>>> or ntp.
>>>
>>> Signed-off-by: David Ahern <dsahern@gmail.com>
>>> ---
>>>  include/trace/events/timekeeping.h |   43 ++++++++++++++++++++++++++++++++++++
>>>  kernel/time/timekeeping.c          |    8 ++++++
>>>  2 files changed, 51 insertions(+), 0 deletions(-)
>>>  create mode 100644 include/trace/events/timekeeping.h
>>>
>>> diff --git a/include/trace/events/timekeeping.h b/include/trace/events/timekeeping.h
>>> new file mode 100644
>>> index 0000000..3d5d083
>>> --- /dev/null
>>> +++ b/include/trace/events/timekeeping.h
>>> @@ -0,0 +1,43 @@
>>> +#undef TRACE_SYSTEM
>>> +#define TRACE_SYSTEM timekeeping
>>> +
>>> +#if !defined(_TRACE_TIMEKEEP_H) || defined(TRACE_HEADER_MULTI_READ)
>>> +#define _TRACE_TIMEKEEP_H
>>> +
>>> +#include <linux/tracepoint.h>
>>> +#include <linux/time.h>
>>> +
>>> +DECLARE_EVENT_CLASS(tod_template,
>>> +
>>> +	TP_PROTO(const struct timespec *tv),
>>> +
>>> +	TP_ARGS(tv),
>>> +
>>> +	TP_STRUCT__entry(
>>> +		__field( __kernel_time_t,	tv_sec)
>>> +		__field( long,			tv_nsec)
>>> +	),
>>> +
>>> +	TP_fast_assign(
>>> +		__entry->tv_sec  = tv->tv_sec;
>>> +		__entry->tv_nsec = tv->tv_nsec;
>>> +	),
>>> +
>>> +	TP_printk("tv_sec=%ld tv_nsec=%ld", __entry->tv_sec, __entry->tv_nsec)
>>> +);
>>> +
>>> +DEFINE_EVENT(tod_template, settimeofday,
>>> +	TP_PROTO(const struct timespec *tv),
>>> +	TP_ARGS(tv));
>>> +
>>> +DEFINE_EVENT(tod_template, timekeeping_inject_offset,
>>> +	TP_PROTO(const struct timespec *tv),
>>> +	TP_ARGS(tv));
>>> +
>>> +DEFINE_EVENT(tod_template, timekeeping_inject_sleeptime,
>>> +	TP_PROTO(const struct timespec *tv),
>>> +	TP_ARGS(tv));
>>
>> Does the fact it's any of the three way of updating xtime make any
>> difference from the user point of view?
> 
> This use case only cares that xtime is updated.
> 
>>
>> If not can we rather factorize that in a single settimeofday tracepoint?
>> Or update_time_of_day if we don't want to confuse the user with the
>> syscall.
> 
> Peter and Thomas expressed interest in timekeeping tracepoints. How the
> update happens might be wanted. If desired I can consolidate xtime = *tv
> into a single update function and put the trace point there.

Any comments on the proposed trace points in the timekeeping code? I am
working on a new spin of the patches and would like to know if there are
changes needed here too.

Thanks,
David