linux-kernel - Re: [PATCH] perf util: Use target->per_thread and target->system

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CANLsYkxJpXDPUVssR2pQPVmwwV8iDfKPK+9psLzVgL8nvVvZ0g@mail.gmail.com>
Date:   Mon, 22 Jan 2018 16:55:22 -0700
From:   Mathieu Poirier <mathieu.poirier@...aro.org>
To:     "Jin, Yao" <yao.jin@...ux.intel.com>
Cc:     Arnaldo Carvalho de Melo <acme@...nel.org>, jolsa@...nel.org,
        Peter Zijlstra <peterz@...radead.org>,
        Ingo Molnar <mingo@...hat.com>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Linux-kernel@...r.kernel.org, Andi Kleen <ak@...ux.intel.com>,
        kan.liang@...el.com, yao.jin@...el.com
Subject: Re: [PATCH] perf util: Use target->per_thread and target->system_wide flags

On 22 January 2018 at 16:02, Jin, Yao <yao.jin@...ux.intel.com> wrote:
>
>
> On 1/23/2018 5:10 AM, Mathieu Poirier wrote:
>>
>> On 22 January 2018 at 15:15, Jin Yao <yao.jin@...ux.intel.com> wrote:
>>>
>>> Mathieu Poirier reports issue in commit ("73c0ca1eee3d perf thread_map:
>>> Enumerate all threads from /proc") that it has negative impact on
>>> 'perf record --per-thread'. It has the effect of creating a kernel event
>>> for each thread in the system for 'perf record --per-thread'.
>>>
>>> Mathieu Poirier's patch ("perf util: Do not reuse target->per_thread
>>> flag")
>>> can fix this issue by creating a new target->all_threads flag.
>>>
>>> This patch is based on Mathieu Poirier's patch but it doesn't use a new
>>> target->all_threads flag. This patch just uses 'target->per_thread &&
>>> target->system_wide' as a condition to check for all threads case.
>>>
>>> Signed-off-by: Jin Yao <yao.jin@...ux.intel.com>
>>> ---
>>>   tools/perf/util/evlist.c     | 2 +-
>>>   tools/perf/util/thread_map.c | 4 ++--
>>>   tools/perf/util/thread_map.h | 2 +-
>>>   3 files changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>>> index 120efd8..9dff74a 100644
>>> --- a/tools/perf/util/evlist.c
>>> +++ b/tools/perf/util/evlist.c
>>> @@ -1106,7 +1106,7 @@ int perf_evlist__create_maps(struct perf_evlist
>>> *evlist, struct target *target)
>>>          struct thread_map *threads;
>>>
>>>          threads = thread_map__new_str(target->pid, target->tid,
>>> target->uid,
>>> -                                     target->per_thread);
>>> +                                     target->per_thread &&
>>> target->system_wide);
>>
>>
>> At first glance I thought your solution would do the trick but perf
>> record does use target->system_wide when the '-a' switch is used.
>> Moreover specifying the '-a' switch doesn't prevent the '--per-thread'
>> option from being used as well, making both target->perf_thread and
>> target_system_wide equal to true (and that is not good).
>>
>> Although not a fan of adding more to struct target, the advantage of
>> having target->all_threads is that we are guaranteed that it isn't
>> used anywhere else.
>>
>> Let me know what you think,
>> Mathieu
>>
>
> If we specify both '-a' and '--per-thread' to perf record, perf record will
> override'--per-thread'. So now target->per_thread = false, and
> target->system_wide = true.

Just spent time looking at it further and target__validate() will make
sure per thread and cpu wide options are mutually exclusive.  I like
your approach better so let's go with that.

Thanks,
Mathieu

>
> If we specify '--per-thread' only to perf record, target->per_thread = true,
> and target->system_wide = false.
>
> So whatever for any case, target->per_thread && target->system_wide is
> false.
>
> Since the parameter is false, in thread_map__new_str(), it will not execute
> the thread_map__new_all_cpus(). So that will not change perf record previous
> behavior.
>
> In perf stat, it allows the case that target->per_thread and
> target->system_wide are all true. That means we want to collect system-wide
> per-thread metrics.
>
> That's my current thinking.
>
> Thanks
> Jin Yao