答复: oprofile-list Digest, Vol 108, Issue 1

答复: oprofile-list Digest, Vol 108, Issue 1

[蒋雄伟(蒋冲)]

In /xen/arch/x86/oprofile/op_counter.h  there is a definition

#define OP_MAX_COUNTER 8

As far as I know , the number of counter equals the number of events.  As
all we know that the number of events is more than 8. 
So my question is : Is it the maximum concurrent counter or the total
counter ? 

-----邮件原件-----
发件人: oprofile-list-request@lists.sourceforge.net
[mailto:oprofile-list-request@lists.sourceforge.net] 
发送时间: 2015年5月5日 10:50
收件人: oprofile-list@lists.sourceforge.net
主题: oprofile-list Digest, Vol 108, Issue 1

Send oprofile-list mailing list submissions to
	oprofile-list@lists.sourceforge.net

To subscribe or unsubscribe via the World Wide Web, visit
	https://lists.sourceforge.net/lists/listinfo/oprofile-list
or, via email, send a message with subject or body 'help' to
	oprofile-list-request@lists.sourceforge.net

You can reach the person managing the list at
	oprofile-list-owner@lists.sourceforge.net

When replying, please edit your Subject line so it is more specific than
"Re: Contents of oprofile-list digest..."


Today's Topics:

   1. Re: Does oprofile suffer from aliasing? (William Cohen)
   2. Perf event setup failed (Mike Accola)
   3. Re: Perf event setup failed (Michael Petlan)
   4. Error cross compiling oprofile (Paul Fine)


----------------------------------------------------------------------

Message: 1
Date: Thu, 30 Apr 2015 14:18:19 -0400
From: William Cohen <wcohen@redhat.com>
Subject: Re: Does oprofile suffer from aliasing?
To: Grant Edwards <grant.b.edwards@gmail.com>,
	oprofile-list@lists.sf.net
Message-ID: <554271EB.10407@redhat.com>
Content-Type: text/plain; charset=windows-1252

On 04/30/2015 12:43 PM, Grant Edwards wrote:
> On 2015-04-30, William Cohen <wcohen@redhat.com> wrote:
>>
>>> AFAICT the ARM926 core doesn't have a PMU at all -- I can find no 
>>> mention of one in either ARM docs for the ARM926 or Atmel docs for 
>>> the
>>> at91 parts.
> [...]
>>> Is there such a thing as software-only PMU support?
> 
>> The perf command included with newer kernels (since 2.6.31) allow 
>> recording both performance monitoring hardware events and software 
>> events.
> 
> So that would be a sort of software-only PMU support (from the user's 
> POV)?

Hi Grant,

Yes, in theory perf should provide software-only events if no pmu is
available. I have seen that happen on trimslice when the device tree didn't
describe performance monitoring hardware.  However, using perf depends on
having a newer kernel that works on the particular device.  In the past I
have encountered arm devices that only worked with particular kernels with
various patches (for example genesi efika smartbook).  I hope that you have
a reasonable new kernel that works on the arm926.

> 
>> Assuming that perf is working you should be able to do following to 
>> see what events it supports:
>>
>> # perf list
> [...]
>> # perf record --freq 1000 -e cpu-clock -a sleep 10
> [...]
>> # perf report
> 
> OK, thanks.  It's probably time to try to get a newer kernel running. 
> IIRC, Atmel provides kernel config examples for a few different 
> versions up to 3.6 for my platform.
> 
> So, let's say I have a hardware interrupt I can hook stuff to (I do), 
> and I can generate "random enough" edges on the pin in question with a 
> predictable frequency distribution (that might take a bit of 
> soldering, but shouldn't be too difficult).
> 
> Can I stick something in the ISR for that interrupt that will cause a 
> "sample"?
> 

You could do something like that with the "perf probe" described on:

https://lwn.net/Articles/355456/
http://www.brendangregg.com/perf.html

It looks like you would need to have a script to do process to group things
by the argument.

-Will




------------------------------

Message: 2
Date: Mon, 4 May 2015 09:56:50 -0500
From: Mike Accola <maccola@us.ibm.com>
Subject: Perf event setup failed
To: oprofile-list@lists.sourceforge.net
Message-ID:
	
<OFE1EF3D3A.4EEEBF45-ON86257E3B.0051E486-86257E3B.00521D5C@us.ibm.com>
Content-Type: text/plain; charset="us-ascii"

I was trying to use operf using the
br_misp_retired:400009:all_branches_pebs event on a machine with Haswell
processor.

I can run count with this event to get counters just fine.

However, when I try to use it with operf, I get this error:

operf -c -e br_misp_retired:400009:all_branches_pebs -d outdir -s
perf_event_open failed with Operation not supported Caught runtime_error:
Internal Error.  Perf event setup failed.
Error running profiler

Any ideas why?

I am using a version of operf 1.0.0 that we compile ourselves.

Thanks.



Mike Accola
IBM InfoSphere Streams
IBM Analytics

Phone: 720-349-3409
E-mail: maccola@us.ibm.com



 
 
-------------- next part --------------
An HTML attachment was scrubbed...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/gif
Size: 7301 bytes
Desc: not available

------------------------------

Message: 3
Date: Mon, 04 May 2015 18:46:06 +0200
From: Michael Petlan <mpetlan@redhat.com>
Subject: Re: Perf event setup failed
To: Mike Accola <maccola@us.ibm.com>
Cc: oprofile-list@lists.sourceforge.net
Message-ID: <1430757966.5694.16.camel@Rudolf-RHEL-7>
Content-Type: text/plain; charset="UTF-8"

Hi Mike,

could you try the following:

perf stat -e r04c5:pp ./something

and then

perf record -e r04c5:pp ./something
perf report


Could you provide some additional info, like kernel version, which
distro is it and cpuinfo?

Thanks.

Regards,
Michael


On Mon, 2015-05-04 at 09:56 -0500, Mike Accola wrote:
> I was trying to use operf using the
> br_misp_retired:400009:all_branches_pebs event on a machine with
> Haswell processor. 
> 
> I can run count with this event to get counters just fine. 
> 
> However, when I try to use it with operf, I get this error: 
> 
> operf -c -e br_misp_retired:400009:all_branches_pebs -d outdir -s 
> perf_event_open failed with Operation not supported 
> Caught runtime_error: Internal Error.  Perf event setup failed. 
> Error running profiler 
> 
> Any ideas why? 
> 
> I am using a version of operf 1.0.0 that we compile ourselves. 
> 
> Thanks.
> 
> 
> ___________________________________
> Mike Accola 
> IBM InfoSphere Streams 
> IBM Analytics
> 
> Phone: 720-349-3409 
> E-mail: maccola@us.ibm.com 
> 
> 
> 
> 
> 
>   
> 
>   
> 
>
----------------------------------------------------------------------------
--
> One dashboard for servers and applications across Physical-Virtual-Cloud 
> Widest out-of-the-box monitoring support with 50+ applications
> Performance metrics, stats and reports that give you Actionable Insights
> Deep dive visibility with transaction tracing using APM Insight.
> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
> _______________________________________________
> oprofile-list mailing list
> oprofile-list@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/oprofile-list





------------------------------

Message: 4
Date: Mon, 4 May 2015 22:49:36 -0400
From: Paul Fine <pbfine91@gmail.com>
Subject: Error cross compiling oprofile
To: oprofile-list@lists.sourceforge.net
Message-ID:
	<CAC7FuqmEaS5H39HCjGhrBqo=Q_feAwAzThpMWqnxDM+KNnudgw@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

Hi,

I am trying to cross compile oprofile for the ARM on the TI C6A816x
running linux-2.6.37.  The "make" step gave me this error:

Making all in pe_profiling
make[2]: Entering directory
`/home/pfine/oprofile/oprofile-1.0.0/pe_profiling'
arm-none-linux-gnueabi-g++ -DHAVE_CONFIG_H -I. -I..  -I ../libutil -I
../libop -I ../libutil++ -I ../libperf_events -I ../libpe_utils
-I/home/pfine/oprofile/linux_hdr_install/include   -W -Wall -fno-common
-ftemplate-depth-50 -g -O2 -MT operf.o -MD -MP -MF .deps/operf.Tpo -c -o
operf.o operf.cpp
operf.cpp: In function 'int start_profiling()':
operf.cpp:330: error: 'pipe2' was not declared in this scope
make[2]: *** [operf.o] Error 1


I can't find the pipe2() function call in the kernel headers.  Any idea
what I am missing?

Thanks
-------------- next part --------------
An HTML attachment was scrubbed...

------------------------------

----------------------------------------------------------------------------
--
One dashboard for servers and applications across Physical-Virtual-Cloud 
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y

------------------------------

_______________________________________________
oprofile-list mailing list
oprofile-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/oprofile-list


End of oprofile-list Digest, Vol 108, Issue 1
*********************************************


------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud 
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
oprofile-list mailing list
oprofile-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/oprofile-list

Re: 答复: oprofile-list Digest, Vol 108, Issue 1

[William Cohen]

On 05/05/2015 12:44 AM, 蒋雄伟(蒋冲) wrote:
> In /xen/arch/x86/oprofile/op_counter.h  there is a definition
> 
> #define OP_MAX_COUNTER 8
> 
> As far as I know , the number of counter equals the number of events.  As
> all we know that the number of events is more than 8. 
> So my question is : Is it the maximum concurrent counter or the total
> counter ? 
> 
Hi,

You are referring to OP_MAX_COUNTER in:

http://xenbits.xen.org/gitweb/?p=xen.git;a=blob;f=xen/arch/x86/oprofile/op_counter.h;h=b515ac9ebc8e379a36917e6f270653d412a87dd7;hb=refs/heads/master

This is the maximum number of performance monitoring counters that the processors could have and can use concurrently.   However, some x86 processors have fewer performance counters available.  Many AMD processors only have 4 and when hyperthreading is used on some Intel processors the number of counters is halfed with each thread getting 4.  The following checks with OP_MAX_COUNTER look a bit odd as a result allowing writing to things that will never be used:

http://xenbits.xen.org/gitweb/?p=xen.git;a=blob;f=xen/arch/x86/oprofile/xenoprof.c;h=002a6962246598ab150f03acc27dc9753fa436f5;hb=refs/heads/master


-Will

------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud 
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
oprofile-list mailing list
oprofile-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/oprofile-list