Re: [RFC] perf: need to expose sched_clock to correlate user samples with kernel samples

From: John Stultz <john.stultz@linaro.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Stephane Eranian <eranian@google.com>,
	Pawel Moll <pawel.moll@arm.com>,
	LKML <linux-kernel@vger.kernel.org>,
	"mingo@elte.hu" <mingo@elte.hu>,
	Paul Mackerras <paulus@samba.org>,
	Anton Blanchard <anton@samba.org>,
	Will Deacon <Will.Deacon@arm.com>,
	"ak@linux.intel.com" <ak@linux.intel.com>,
	Pekka Enberg <penberg@gmail.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Robert Richter <robert.richter@amd.com>
Subject: Re: [RFC] perf: need to expose sched_clock to correlate user samples with kernel samples
Date: Fri, 22 Feb 2013 22:04:33 -0800	[thread overview]
Message-ID: <51285BF1.2090208@linaro.org> (raw)
In-Reply-To: <1361356160.10155.22.camel@laptop>

On 02/20/2013 02:29 AM, Peter Zijlstra wrote:
> On Tue, 2013-02-19 at 10:25 -0800, John Stultz wrote:
>> So describe how the perf time domain is different then
>> CLOCK_MONOTONIC_RAW.
> The primary difference is that the trace/sched/perf time domain is not
> strictly monotonic, it is only locally monotonic -- that is two time
> stamps taken on the same cpu are guaranteed to be monotonic.

So how would a clock_gettime(CLOCK_PERF,...) interface help you figure 
out which cpu you got your timestamp from?

> Furthermore, to make it useful, there's an actual bound on the inter-cpu
> drift (implemented by limiting the drift to CLOCK_MONOTONIC).

So this sounds like you're already sort of interpolating to 
CLOCK_MONOTONIC, or am I just misunderstanding you?

> Additionally -- to increase use -- we also added a monotonic sync point
> when cpu A queries time of cpu B.

Not sure I'm following this bit. But I'll have to go look at the code on 
Monday.

>
>> My concern here is that we're basically creating a kernel interface
>> that
>> exports implementation-defined semantics (again: whatever perf does
>> right now). And I think folks want to do this, because adding
>> CLOCK_PERF
>> is easier then trying to:
>>
>> 1) Get a lock-free method for accessing CLOCK_MONOTONIC_RAW
>>
>> 2) Having perf interpolate its timestamps to CLOCK_MONOTONIC, or
>> CLOCKMONOTONIC_RAW when it exports the data
> Mostly cheaper, not easier. Given unstable TSC, MONOTONIC will have to
> fall back to another clock source (hpet, acpi_pm and other assorted
> crap).
>
> In order to avoid this, we'd had to relax the requirements. Using
> anything other than TSC is simply not an option.

Right, and this I understand. We can can play a little fast and lose 
with the rules for in-kernel uses, given the variety of hardware and the 
fact that performance is more critical then perfect accuracy. Since 
we're in-kernel we also have more information then userland does about 
what cpu we're running on, so we can get away with only 
locally-monotonic timestamps.

But I want to be careful if we're exporting this out to userland that 
its both useful and that there's an actual specification for how 
CLOCK_PERF behaves, applications can rely upon not changing in the future.

thanks
-john