Re: [tip:perf/timer] perf: Add per event clockid support

From: Peter Zijlstra <peterz@infradead.org>
To: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Arnaldo Carvalho de Melo <acme@redhat.com>,
	Jiri Olsa <jolsa@redhat.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	LKML <linux-kernel@vger.kernel.org>,
	John Stultz <john.stultz@linaro.org>,
	"H. Peter Anvin" <hpa@zytor.com>, David Ahern <dsahern@gmail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Ingo Molnar <mingo@kernel.org>
Subject: Re: [tip:perf/timer] perf: Add per event clockid support
Date: Fri, 27 Mar 2015 17:57:39 +0100	[thread overview]
Message-ID: <20150327165739.GJ23123@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <CABPqkBT-H8T40FcDZAWahMAajP4NBbMOVs9Awghh598xaoiz9Q@mail.gmail.com>

On Fri, Mar 27, 2015 at 09:52:25AM -0700, Stephane Eranian wrote:
> Ok, I see. But on architectures which do not have NMI, they would all
> be safe. And that would work if they set the PERF_PMU_CAP_NO_NMI flag
> on their pmu->capabilities.

Indeed so. I have not audited them all as the current 'default' is safe.

> Next, I am trying to understand how perf is going to expose this. I am
> thinking about this in the context of my jitted code patches.

I send you some preliminary patches for this.

> With this approach, the jit runtime and the perf tool need to agree on
> the clock they are using. That mean they need to advertise or document
> the clock they use

This is indeed still missing, I put an XXX question in one of the
emails to this effect. I suppose we should store the used clockid in the
data file. I just haven't figured out how all that code works.

> and there needs to be an option in perf record to pass that same
> clockid.

I did indeed post a patch to do that.