Received-SPF: pass (google.com: domain of linux-kernel+bounces-102668-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Wed, 13 Mar 2024 17:00:38 -0700
From: Andi Kleen <ak@linux.intel.com>
To: "Wang, Weilin" <weilin.wang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>, Ian Rogers <irogers@google.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>,
	Alexander Shishkin <alexander.shishkin@linux.intel.com>,
	Jiri Olsa <jolsa@kernel.org>,
	"Hunter, Adrian" <adrian.hunter@intel.com>,
	Kan Liang <kan.liang@linux.intel.com>,
	"linux-perf-users@vger.kernel.org" <linux-perf-users@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"Taylor, Perry" <perry.taylor@intel.com>,
	"Alt, Samantha" <samantha.alt@intel.com>,
	"Biggers, Caleb" <caleb.biggers@intel.com>
Subject: Re: [RFC PATCH v4 2/6] perf stat: Fork and launch perf record when
 perf stat needs to get retire latency value for a metric.
Message-ID: <ZfI-JqTZevIhEa3o@tassilo>
References: <20240312234921.812685-1-weilin.wang@intel.com>
 <20240312234921.812685-3-weilin.wang@intel.com>
 <87le6nm20o.fsf@linux.intel.com>
 <CO6PR11MB56356A213DBE53535E9D802BEE2A2@CO6PR11MB5635.namprd11.prod.outlook.com>
 <87edcflzkr.fsf@linux.intel.com>
 <CO6PR11MB56351D1706A9C46D80982AECEE2A2@CO6PR11MB5635.namprd11.prod.outlook.com>
 <ZfHMYM3iWlsODtjP@tassilo>
 <CO6PR11MB5635D2EAC53C2CD2B36D4BF5EE2A2@CO6PR11MB5635.namprd11.prod.outlook.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CO6PR11MB5635D2EAC53C2CD2B36D4BF5EE2A2@CO6PR11MB5635.namprd11.prod.outlook.com>

> We could use -a -C and cgroup together. I think this could be a useful use case.
> There could be other improvement to the implementation in next step. But I believe

I don't know how you would improve it. A lot of the problems are fairly
fundamental.

> current implementation could provide users the access to our new feature with 
> accurate results 
> and without adding too much overhead. 

perf record increases the overhead a lot over a perf stat! Sampling
is much more expensive than counting.

That should be at least a order of magnitude difference.
Another advantage of doing it separately.

That will also make it inaccurate.

Please do a proper implementation. This one is no good.

You can likely reuse a lot of your code:

- Add a perf calibrate to run the profile step separately that generates the JSON
- Add a --metrics option to perf to read the resulting JSON
- Add some mechanism to read a default JSON as fallback. I guess this
one could be compiled in to avoid a dependency on an installed file.

-Andi