From: "Liang, Kan" <kan.liang@intel.com>
To: Jiri Olsa <jolsa@redhat.com>
CC: "acme@kernel.org" <acme@kernel.org>,
        "peterz@infradead.org" <peterz@infradead.org>,
        "mingo@redhat.com" <mingo@redhat.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "wangnan0@huawei.com" <wangnan0@huawei.com>,
        "jolsa@kernel.org" <jolsa@kernel.org>,
        "namhyung@kernel.org" <namhyung@kernel.org>,
        "ak@linux.intel.com" <ak@linux.intel.com>,
        "yao.jin@linux.intel.com" <yao.jin@linux.intel.com>
Subject: RE: [PATCH V3 08/12] perf top: check per event overwrite term
Thread-Topic: [PATCH V3 08/12] perf top: check per event overwrite term
Thread-Index: AQHTeohZVR6bK+Uv8UC/ikJPKdDyPaNuVPqAgADsXwCAAHg4AIAAsIRg
Date: Fri, 12 Jan 2018 14:58:29 +0000
Message-ID: <37D7C6CF3E00A74B8858931C1DB2F077537FFFB9@SHSMSX103.ccr.corp.intel.com>
References: <1513879734-237492-1-git-send-email-kan.liang@intel.com>
 <1513879734-237492-9-git-send-email-kan.liang@intel.com>
 <20180111142549.GD16655@krava>
 <37D7C6CF3E00A74B8858931C1DB2F077537FFC1F@SHSMSX103.ccr.corp.intel.com>
 <20180112114206.GA25413@krava>
In-Reply-To: <20180112114206.GA25413@krava>
Accept-Language: zh-CN, en-US
Content-Language: en-US
dlp-product: dlpe-windows
dlp-version: 11.0.0.116
dlp-reaction: no-action
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 8BIT
MIME-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org


> On Thu, Jan 11, 2018 at 09:29:21PM +0000, Liang, Kan wrote:
> > > On Thu, Dec 21, 2017 at 10:08:50AM -0800, kan.liang@intel.com wrote:
> > > > From: Kan Liang <kan.liang@intel.com>
> > > >
> > > > Per event overwrite term is not forbidden in perf top, which can
> > > > bring problems. Because perf top only support non-overwrite mode.
> > > >
> > > > Check and forbid inconsistent per event overwrite term in the evlist.
> > > > Make it possible to support either non-overwrite or overwrite mode.
> > > > The overwrite mode is forbidden now, which will be removed when
> > > > the overwrite mode is supported later.
> > > >
> > > > Signed-off-by: Kan Liang <kan.liang@intel.com>
> > > > ---
> > > >  tools/perf/builtin-top.c | 60
> > > ++++++++++++++++++++++++++++++++++++++++++++++++
> > > >  1 file changed, 60 insertions(+)
> > > >
> > > > diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> > > > index c6ccda5..4b85e7b 100644
> > > > --- a/tools/perf/builtin-top.c
> > > > +++ b/tools/perf/builtin-top.c
> > > > @@ -881,6 +881,56 @@ static void perf_top__mmap_read(struct
> > > > perf_top
> > > *top)
> > > >  		perf_top__mmap_read_idx(top, i);  }
> > > >
> > > > +/*
> > > > + * Check per event overwrite term.
> > > > + * perf top supports consistent mode for all events.
> > > > + * Return -1 if the per event terms set but not consistent.
> > >
> > > please list the rules for overwrite in the comment are they just top
> > > specific?
> >
> > Yes, it's only for perf top.
> > It doesn't support that some events are overwrite mode and other
> > events are non-overwrite mode.
> > I will refine the comments.
> >
> > >
> > > SNIP
> > >
> > > > +			if (evsel == perf_evlist__first(evlist))
> > > > +				overwrite = set;
> > > > +			else
> > > > +				return -1;
> > > > +		}
> > > > +	}
> > > > +
> > > > +	if ((overwrite >= 0) && (opts->overwrite != overwrite))
> > > > +		opts->overwrite = overwrite;
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > >  static int perf_top__start_counters(struct perf_top *top)  {
> > > >  	char msg[BUFSIZ];
> > > > @@ -890,6 +940,16 @@ static int perf_top__start_counters(struct
> > > perf_top *top)
> > > >
> > > >  	perf_evlist__config(evlist, opts, &callchain_param);
> > >
> > > so perf_evlist__config uses opts->overwrite, which you set
> > > in your perf_top_overwrite_check call.. I'd think you need
> > > to call it sooner
> > >
> >
> > User may set per-event mode, which is processed in perf_evlist__config
> > It's possible that the per-event mode is not same as opts->overwrite.
> > If so, perf_evlist__config actually uses per-event mode.
> >
> > perf_top_overwrite_check will do the check. If the per-event mode is
> > inconsistent as opts->overwrite, updating the opts->overwrite.
> >
> > It has to be called after the perf_evlist__config.
> 
> let's see the rules then first, also wrt opts->overwrite,
> which is used later on I assume
>

The rules for perf_top_overwrite_check is quite simple.
 - All events have same per-event mode.
   E.g. "cpu/cpu-cycles,no-overwrite/,cpu/instructions,no-overwrite/
   Using the per-event setting to replace the opts->overwrite if they are different.
 - Events have different per-event mode.
   E.g. "cpu/cpu-cycles,overwrite/,cpu/instructions,no-overwrite/"
   Error out.
 - Some of the event set per-event mode, but some not.
   E.g. "cpu/cpu-cycles, /,cpu/instructions,no-overwrite/"
   Error out.

There is no --overwrite option support for perf top.
We have a default opts->overwrite.
In perf_evlist__config, attr->write_backward will be set according to
opts->overwrite at first. But if there is per-event mode, the
attr->write_backward will be updated accordingly.
There may be inconsistent between opts->overwrite and
attr->write_backward.

perf_top_overwrite_check will be called right after perf_evlist__config.
It can fix the inconsistence by updating the opts->overwrite or error out.
The opts->overwrite will be used by fallback and _mmap_read() later.
 
Thanks,
Kan