Message-ID: <51DEE880.6070908@gmail.com>
Date: Thu, 11 Jul 2013 11:16:48 -0600
From: David Ahern <dsahern@gmail.com>
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:17.0) Gecko/20130620 Thunderbird/17.0.7
MIME-Version: 1.0
To: Adrian Hunter <adrian.hunter@intel.com>,
        Stephane Eranian <eranian@google.com>
CC: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>,
        linux-kernel@vger.kernel.org, Frederic Weisbecker <fweisbec@gmail.com>,
        Jiri Olsa <jolsa@redhat.com>, Mike Galbraith <efault@gmx.de>,
        Namhyung Kim <namhyung@gmail.com>, Paul Mackerras <paulus@samba.org>,
        Peter Zijlstra <peterz@infradead.org>, Ingo Molnar <mingo@kernel.org>
Subject: Re: [PATCH V5 09/12] perf: make events stream always parsable
References: <1373548341-24119-1-git-send-email-adrian.hunter@intel.com> <1373548341-24119-10-git-send-email-adrian.hunter@intel.com> <51DED2AA.6030307@gmail.com>
In-Reply-To: <51DED2AA.6030307@gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2221
Lines: 48

On 7/11/13 9:43 AM, David Ahern wrote:
> On 7/11/13 7:12 AM, Adrian Hunter wrote:
>> The event stream is not always parsable because the format of a sample
>> is dependent on the sample_type of the selected event.  When there
>> is more than one selected event and the sample_types are not the
>> same then parsing becomes problematic.  A sample can be matched to its
>> selected event using the ID that is allocated when the event is opened.
>> Unfortunately, to get the ID from the sample means first parsing it.
>
> Here's an alternative suggestion -- one that does not involve changing
> the kernel API or requiring a common denominator in sample_type options.
>
> perf handles event streams through an mmap which can be directly tied to
> an evsel (a single event) when the mmap is created. ie., when events are
> read we know exactly which evsel they correspond to. (See
> perf_evlist__mmap_per_cpu and perf_evlist__mmap_per_thread and add
> struct perf_evsel *evsel entry to struct perf_mmap).

Read that code a bit too quickly. All events for a thread are dropped 
into the same buffer so as currently organized there is not an easy 
correlation.

David

>
> Commands like perf-record can inject a user event into the stream and
> hence the data file every time the evsel changes while walking all of
> the mmap's reading events -- very  similar to the way finished round is
> done. The event would only contain a perf_event_header which is 8 bytes
> so this does not add a lot to a data file. As an optimization the evsel
> event could only be injected if the sample_types differ.
>
> Live commands would just use the evsel connected to the mmap -- no
> lookups needed which would simplify things a bit processing the events.
>
> In short, the information to associate event streams to an event (evsel)
> is currently available -- it's just being discarded in the many layers.
>
> I'll try to whip up some code that implements this in the next few days.
>
> David
>

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/