linux-kernel - perf: h/w counters not counted and no error/info reported to user

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-next>] [day] [month] [year] [list]

Message-ID: <4DDECFC8.7090903@gmail.com>
Date:	Thu, 26 May 2011 16:10:16 -0600
From:	David Ahern <dsahern@...il.com>
To:	Ingo Molnar <mingo@...e.hu>,
	Arnaldo Carvalho de Melo <acme@...stprotocols.net>
CC:	LKML <linux-kernel@...r.kernel.org>
Subject: perf: h/w counters not counted and no error/info reported to user

By default (tip-perf-core) perf-stat enables a number of H/W counters:

perf stat  -- sleep 1

Performance counter stats for 'sleep 1':

     1.346889 task-clock
            1 context-switches
            0 CPU-migrations
          157 page-faults
<not counted> cycles
<not counted> stalled-cycles-frontend
<not counted> stalled-cycles-backend
      493,740 instructions
      137,062 branches
        6,773 branch-misses

In this case 3 of them are not counted and perf does not tell me why.
I'm guessing that my processor (Core2 Duo T9550) does not support the
stalled cycles counters. Should perf not emit a message telling me that?

However, cycles is supported by the processor:

perf stat -e cycles -- sleep 1

Performance counter stats for 'sleep 1':

         1,091,102 cycles

Based on that I am guessing that only the last N requested events are
counted where N is processor specific. In my case N is apparently 3 as
this works:

perf stat -e cycles -e instructions -e branches -- sleep 1

 Performance counter stats for 'sleep 1':

    1,028,331 cycles
      634,101 instructions
      130,253 branches

but adding a fourth causes the first one to roll off -- again without
any message from perf:

perf stat -e cycles -e instructions -e branches -e branch-misses -v --
sleep 1
Performance counter stats for 'sleep 1':

<not counted> cycles
      219,727 instructions
      128,071 branches
        6,625 branch-misses

Which is confusing -- cycles are not counted, yet it should be given
that it is supported.

David
--
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/