linux-kernel - Re: 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-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 27 May 2011 09:04:35 -0600
From:	David Ahern <dsahern@...il.com>
To:	Ingo Molnar <mingo@...e.hu>
CC:	Arnaldo Carvalho de Melo <acme@...stprotocols.net>,
	LKML <linux-kernel@...r.kernel.org>,
	Peter Zijlstra <a.p.zijlstra@...llo.nl>
Subject: Re: perf: h/w counters not counted and no error/info reported to
 user



On 05/27/2011 03:50 AM, Ingo Molnar wrote:
> 
> * David Ahern <dsahern@...il.com> wrote:
> 
>> By default (tip-perf-core) perf-stat enables a number of H/W counters:
>>
>> perf stat  -- sleep 1
> 
> did you mean to do 'perf stat -a sleep 1' ?

Adding the -a has its own problems. It blocks reading from fd=0 (stdin);
I have to type Ctrl-D to see the results. The FD array gets initialized
to 0 and some case is not erroring out or setting the fd.

perf stat -a -- sleep 1

 Performance counter stats for 'sleep 1':

       2008.600377 task-clock
               545 context-switches
                31 CPU-migrations
               198 page-faults
       101,938,898 cycles
140,734,186,560,480 stalled-cycles-frontend
                 0 stalled-cycles-backend
       115,406,129 instructions

        26,361,321 branches
         1,068,804 branch-misses

       1.003842524 seconds time elapsed

The stalled-cycles-frontend is off the chats, but shows a value.
Similarly, backed shows 0 versus <not counted>.


> 
> 'perf stat sleep 1' in itself will only measure that very simple 
> command - which does not do much.

Simple program for a simple test. I just wanted to know which counters
are available for a given processor.

Per your comment and Corey's I tried a program that does some work:

perf stat -- openssl speed aes

Performance counter stats for 'openssl speed aes':

      44964.148608 task-clock
             4,593 context-switches
                48 CPU-migrations
               487 page-faults
   124,134,470,217 cycles
     <not counted> stalled-cycles-frontend
     <not counted> stalled-cycles-backend
   319,440,401,206 instructions
     6,710,107,937 branches
         1,638,412 branch-misses

      45.012092601 seconds time elapsed
This time I get data for 4 events -- which I did not get with sleep --
but I get <not counted> for the stalled cycles.

How is the <not counted> supposed to be interpreted?

Also, back to the 'perf stat -- sleep 1' example: If I run it with
individual -e <event> args I get values -- up to 3. Sure sleep is a
simple program, but some counter value comes out. What's magical about 3?

David


> 
> Thanks,
> 
> 	Ingo
--
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/