Re: [RFC][PATCH 1/3] tracing: Added hardware latency tracer

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 08/05/2016 04:44 PM, Steven Rostedt wrote:
> On Fri, 5 Aug 2016 16:25:21 +0200
> Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote:
> 
>> * Steven Rostedt | 2016-08-04 10:57:09 [-0400]:
>>
>>> diff --git a/kernel/trace/trace_hwlat.c b/kernel/trace/trace_hwlat.c
>>> new file mode 100644
>>> index 000000000000..08dfabe4e862
>>> --- /dev/null
>>> +++ b/kernel/trace/trace_hwlat.c  
>> …
>>> +/* Macros to encapsulate the time capturing infrastructure */
>>> +#define time_type	u64
>>> +#define time_get()	trace_clock_local()
>>> +#define time_to_us(x)	div_u64(x, 1000)
>>> +#define time_sub(a, b)	((a) - (b))
>>> +#define init_time(a, b)	(a = b)
>>> +#define time_u64(a)	a  
>>
>> Do we need a macro for this? In the old code we could choose between
>> CONFIG_TRACING but now we don't.
>>
> 
> Probably not, I kept it for two reasons. 1) to keep the same logic as
> what was in PREEMPT_RT, and 2) in case we can come up with a better
> clock.

I assumed it was a leftover.

> But it's not that important. Should it be nuked? They do somewhat make
> the code easier to read.

that time_get() is close to ktime_get() which is almost u64 nowadays.
So it might not be that cool for upstream. A hwlat prefix makes the
whole thing not prettier.

1. PREEMPT_RT. Do I need any changes? I assumed I could keep this 1:1
   (once it is merged) and throw the current hwlat out.
2. a better clock is an argument. But why would you have a better clock
   for hwlat and not for the whole tracing infrastructure?

If you want to keep it, keep it. I just assumed it was a leftover.
> 
> -- Steve

Sebastian