linux-kernel - Re: [RFC 00/13] perf_env/CPU socket reorg/fixes

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: <55F2C726.4080204@huawei.com>
Date:	Fri, 11 Sep 2015 20:20:54 +0800
From:	"Wangnan (F)" <wangnan0@...wei.com>
To:	Arnaldo Carvalho de Melo <acme@...nel.org>,
	Kan Liang <kan.liang@...el.com>
CC:	Ingo Molnar <mingo@...nel.org>, <linux-kernel@...r.kernel.org>,
	"Adrian Hunter" <adrian.hunter@...el.com>,
	Borislav Petkov <bp@...e.de>, David Ahern <dsahern@...il.com>,
	Frederic Weisbecker <fweisbec@...il.com>,
	Jiri Olsa <jolsa@...hat.com>,
	Namhyung Kim <namhyung@...nel.org>,
	Stephane Eranian <eranian@...gle.com>
Subject: Re: [RFC 00/13] perf_env/CPU socket reorg/fixes

Hi Arnaldo,

I have tested patch 1 to 10. They looks good to me except patch 4/13. Please
see my email in that thread.

However, during the testing I found a limitation related to cpu
online/offline and 'perf top' that, if I offline most of cores before
'perf top', then online them during 'perf top' running, 'perf top'
dooesn't report new CPUs. It still reports the CPUs which are online
when 'perf top' starts consume 100% cycles.  So if CPUs are online and
offlined dynamically and there are many CPUs, user of 'perf top' may get
confusion result if he or she doesn't noticed that 'perf top' doesn't
listed all cores they have.

Here is how I did this:

  # for i in `seq 2 7` ; do echo 0 > 
/sys/devices/system/cpu/cpu$i/online ; done
  # perf top -s cpu,socket

  The result is something like:

  Samples: 28K of event 'cycles', Event count (approx.): 23640606383
  Overhead  Socket CPU
    67.14%     000  000
    32.86%     000  001

Then online them:

  # for i in `seq 2 7` ; do echo 1 > 
/sys/devices/system/cpu/cpu$i/online ; done

After a while, 'perf top' still reports two CPUs.

  Samples: 400K of event 'cycles', Event count (approx.): 38728257939
  Overhead  Socket CPU
    51.02%     000  001
    48.98%     000  000

And another 'perf top' report correct result:

Samples: 28K of event 'cycles', Event count (approx.): 24741565854
Overhead  Socket CPU
   27.26%     000  005
   21.07%     000  002
   13.07%     000  001
   12.69%     000  000
    8.07%     000  006
    6.75%     000  007
    5.64%     000  004
    5.45%     000  003

However, It is relatively a rare case. I don't think we have to fix it 
in this
patchset.

Thank you.

On 2015/9/10 3:50, Arnaldo Carvalho de Melo wrote:
> Hi,
>
> 	Please take a look at these changes to fix the problems reported by
> Wang Nan wrt accesses to the cpu_topology_map information.
>
> 	The fixes are present on these following two csets:
>
>    perf event: Use machine->env to find the cpu -> socket mapping
>    perf report: Do not blindly use env->cpu[al.cpu].socket_id
>
> 	The rest are fixes made while working on this, infrastructure to enable
> the fixes, reverts for things that ended up not being necessary and some
> cleanups.
>
> 	It is available at:
>
>   git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git perf/env
>
> 	Please let me know if I can have your Acked-by, Tested-by or
> Reviewed-by.
>
> - Arnaldo
>
> Arnaldo Carvalho de Melo (13):
>    perf env: Move perf_env out of header.h and session.c into separate object
>    perf env: Rename some leftovers from rename to perf_env
>    perf env: Adopt perf_header__set_cmdline
>    perf env: Introduce read_cpu_topology_map() method
>    perf sort: Set flag stating if the "socket" key is being used
>    perf top: Cache the cpu topology info when "-s socket" is used
>    perf hists browser: Fixup the "cpu" column width calculation
>    perf machine: Add pointer to sample's environment
>    perf event: Use machine->env to find the cpu -> socket mapping
>    perf report: Do not blindly use env->cpu[al.cpu].socket_id
>    Revert "perf evsel: Add a backpointer to the evlist a evsel is in"
>    perf evsel: Remove forward declaration of 'struct perf_evlist'
>    Revert "perf evlist: Add backpointer for perf_env to evlist"


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