Re: [diamon-discuss] My experience on perf, CTF and TraceCompass, and some suggection.

On 2015/1/24 4:12, Alexandre Montplaisir wrote:
> 
> On 01/23/2015 04:35 AM, Wang Nan wrote:
>> [...] I prefer to use:
>>
>>   # perf record -a -e sched:* -e syscalls:* sleep 1
>>
>> However there are some bugs and I have to make some patches. They are
>> posted and being disscussed currently, those bugs are still exist
>> upstream.
> 
> Hmm, I can confirm this problem, when trying to convert a trace taken with these events enabled, I get the following error:
> 
> Failed to add field 'nr
> Failed to add event 'syscalls:sys_enter_io_submit'.
> 
> Good to know it's being looked at!
> 
> 
> But when you say:
> 
>> TraceCompass doesn't recognize syscall:* tracepoints as CPU status
>> changing point. I have to also catch raw_syscall:*, and which doubles
>> the number of samples.
> 
> I assume this was that with your patches applied, so that you could take a trace of the syscall:* tracepoints?
> 
> Do you have a link to those patches somewhere? Or just let us know when it gets integrated upstream. I was trying to update the analysis in Trace Compass to handle syscall:* too, but we can't really do it without knowing how it will look like in CTF format ;)
> 
> 

Hi,

You can follow this thread in LKML:

https://lkml.org/lkml/2015/1/20/330

In https://lkml.org/lkml/2015/1/22/24 , I posted 2 RFC patches can solve the problem.
However I'm still waiting response from Steven Rostedt whether he allow us to improve
traceevent.


> Cheers,
> Alexandre
>