perfmon2 kernel and related tools developments ()

Corey Ashford | 1 Apr 2008 01:24

Picon

perfmon2 code restructuring news

Hi Stephane,

This is great news!  I will start the testing process on POWER soon.

Assuming the testing goes relatively smoothly, do you have a feeling as
to when you'd be able to post the first of the patches to the LKML
mailing list?

Thanks,

- Corey

> Hello,
> 
> I have spent most of last week restructuring the current perfmon2
> kernel code base to make it easier to isolate
> features. That should make it easier for the merge with mainline given
> that LKML people would like to receive
> small patches each adding one feature at a time.
> 
> I have created a bunch of new files under perfmon/, each one
> implementing a feature or logically similar set of
> features. Similarly, I have also restructured the header files to
> separate the generic interface definition in
> perfmon.h from all the stuff needed by the kernel implementation, now
> in perfmon_kern.h. This model is replicated
> for each arch. Perfmon_const.h is gone.
> 
> I have also completely restructured the interrupt handling code in
> perfmon_intr.c. It is now much more readable

(Continue reading)

Arnd Bergmann | 1 Apr 2008 02:28

X-Face

Picon

perfmon2 code restructuring news

stephane eranian wrote:
>  I have spent most of last week restructuring the current perfmon2
>  kernel code base to make it easier to isolate
>  features. That should make it easier for the merge with mainline given
>  that LKML people would like to receive
>  small patches each adding one feature at a time.

Ah, this looks really good. I haven't really followed the perfmon2
progress over the last few releases. What is your plan for submission
to upstream inclusion? I guess if you want to have a chance of getting
it into 2.6.26, now would be the time to post the patches.
I can certainly offer some help getting it into shape for inclusion
if you're interested.

At some point I think you'll need a rebased git tree that contains
the changesets in exactly the way you want them to be merged,
and then you keep changing those patches, instead of adding new
changes on top.

>  I have created a bunch of new files under perfmon/, each one
>  implementing a feature or logically similar set of
>  features. Similarly, I have also restructured the header files to
>  separate the generic interface definition in
>  perfmon.h from all the stuff needed by the kernel implementation, now
>  in perfmon_kern.h. This model is replicated
>  for each arch. Perfmon_const.h is gone.

I looked at the file structure now, and I think it can be done a little
bit better still:

(Continue reading)

TakashiYamamoto | 1 Apr 2008 04:31

Picon

Re: [PATCH 2/7] : Add cell_spe_follow flag to the context flags

Hello,

> >  > Why do you need to expose this Cell-specific feature into generic code?
> >  >
> >  > Can't you hide this inside the arch code by using the pfm_arch_*() callbacks?
> >
> >  No.
> >  In this patch, I want to avoid setting TIF_PERFMON_CTXSW to the task flag
> >  when cell_spe_follow flag is set.
> >  I couln't hide this change in pfmon_arch_load_context(),
> >  because it is called before set_task_thread_flag().
> >
> Ok, I see your problem now.
> 
> We have 2 solutions:
>    - move pfm_arch_load_context() AFTER the TIF_PERFMON_CTXSW
>    - add a generic flag (context_flags) that pfm_arch_load_context()
> can set to tell the generic code NOT to set TIF_PERFMON_CTXSW
> 
> I'd rather not move pfm_arch_load_context() because there would be
> more  code to undo if it ever fails.
> 
> The 2nd solution is not too pretty either because it relies on the
> position of pfm_arch_load_context(). However, we could move
> setting that flag to pfm_arch_create_context() given that you've added
> the context-level flag for SPE_FOLLOW. Could you try that?

How about this?
Is this the change which you expected?

(Continue reading)

Feng Chen | 1 Apr 2008 05:32

Picon

Perfmon2 does not work correctly for Core Quad

Hello,

  we just purchased a Dell PowerEdge 1900 machine with two Intel Core
Quad 5355 processors, and the OS is RHAS5. We installed the perfmon2
(080225) on the machine. The compilation and installation has no
problem, but we met a problem, only seven events could be listed using
"pfmon -l", they are:

UNHALTED_CORE_CYCLES
INSTRUCTIONS_RETIRED
UNHALTED_REFERENCE_CYCLES
LAST_LEVEL_CACHE_REFERENCES
LAST_LEVEL_CACHE_MISSES
BRANCH_INSTRUCTIONS_RETIRED
MISPREDICTED_BRANCH_RETIRED

pfmon -I shows the CPU has been detected correctly. We also tried
older version 071017 and 070209. We also tried to install it in a
64-bit version of RHAS5, it does work either. Appreciate it if you can
give some hints to solve the problem.

Thanks a lot.

Feng

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace

(Continue reading)

stephane eranian | 1 Apr 2008 08:05

Re: Perfmon2 does not work correctly for Core Quad

Hello,

For some reason, pfmon and libpfm are defaulting to the architected PMU.
Could you send me the output from /proc/cpuinfo (one CPU is enough)?

Thanks.

On Tue, Apr 1, 2008 at 5:32 AM, Feng Chen <fchen1978 <at> gmail.com> wrote:
> Hello,
>
>   we just purchased a Dell PowerEdge 1900 machine with two Intel Core
>  Quad 5355 processors, and the OS is RHAS5. We installed the perfmon2
>  (080225) on the machine. The compilation and installation has no
>  problem, but we met a problem, only seven events could be listed using
>  "pfmon -l", they are:
>
>  UNHALTED_CORE_CYCLES
>  INSTRUCTIONS_RETIRED
>  UNHALTED_REFERENCE_CYCLES
>  LAST_LEVEL_CACHE_REFERENCES
>  LAST_LEVEL_CACHE_MISSES
>  BRANCH_INSTRUCTIONS_RETIRED
>  MISPREDICTED_BRANCH_RETIRED
>
>  pfmon -I shows the CPU has been detected correctly. We also tried
>  older version 071017 and 070209. We also tried to install it in a
>  64-bit version of RHAS5, it does work either. Appreciate it if you can
>  give some hints to solve the problem.
>
>  Thanks a lot.

(Continue reading)

stephane eranian | 1 Apr 2008 09:19

Re: Problems encountered on Pentium 4

Hello,

2008/3/27 J K Rai <jk.anurag <at> yahoo.com>:
>
> Problems encountered while using two events together:
> ========================================================
> ******************************************************************************************************
> The events mentioned first gets the count values while the next one remains
> zero always as shown below
> ******************************************************************************************************
>
> session output:
> ==============
> ashwin <at> SUSE-HT:~> pfmon -ePAGE_WALK_TYPE:DTMISS,PAGE_WALK_TYPE:ITMISS ls
> /dev/null
> /dev/null
> 967 PAGE_WALK_TYPE:DTMISS
>   0 PAGE_WALK_TYPE:ITMISS
>
Could you try measuring ONLY the second event instead? It may be a bug
in libpfm assignment of events to counters. That would
not surprise me on Pentium 4, it is so complicated.

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace

Feng Chen | 1 Apr 2008 09:47

Picon

Re: Perfmon2 does not work correctly for Core Quad

hello,

Thanks for response.
/proc/cpuinfo

processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 15
model name      : Intel(R) Xeon(R) CPU           X5355   <at>  2.66GHz
stepping        : 11
cpu MHz         : 2660.081
cache size      : 4096 KB
physical id     : 0
siblings        : 4
core id         : 0
cpu cores       : 4
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 10
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge
mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx
lm constant_tsc arch_perfmon pebs bts pni monitor ds_cpl vmx est tm2
ssse3 cx16 xtpr dca lahf_lm
bogomips        : 5429.34

(Continue reading)

stephane eranian | 1 Apr 2008 10:25

Re: Perfmon2 does not work correctly for Core Quad

Hello,

On Tue, Apr 1, 2008 at 9:47 AM, Feng Chen <fchen1978 <at> gmail.com> wrote:
> hello,
>
>  Thanks for response.
>  /proc/cpuinfo
>
>  processor       : 0
>  vendor_id       : GenuineIntel
>  cpu family      : 6
>  model           : 15

That part looks good. Intel Core should be detected here
>  ------------------------------------
>  The output of "pfmon -I" is:
>
>  detected host CPUs:  8-way 2660MHz/4.0MB -- Intel(R) Xeon(R) CPU X5355
>   <at>  2.66GHz (stepping 11)
>  detected PMU model: Intel architectural PMU

That's not good. Should be Intel Core. Pfmon says it has support for
it. So I suspect
you are not using the correct version of libpfm. Make sure you have
the latest library
installed (libpfm-3.3).

>  max counters/set: 2
>  supported PMU models: [AMD64] [Pentium 4] [Intel Core] [Intel P6
>  processor] [Intel Pentium II] [Intel Pentium Pro] [Intel Pentium M]

(Continue reading)

stephane eranian | 1 Apr 2008 11:11

Re: perfmon2 code restructuring news

Hello Arnd,

On Tue, Apr 1, 2008 at 2:28 AM, Arnd Bergmann <arnd <at> arndb.de> wrote:
> stephane eranian wrote:
>  >  I have spent most of last week restructuring the current perfmon2
>  >  kernel code base to make it easier to isolate
>  >  features. That should make it easier for the merge with mainline given
>  >  that LKML people would like to receive
>  >  small patches each adding one feature at a time.
>
>  Ah, this looks really good. I haven't really followed the perfmon2
>  progress over the last few releases. What is your plan for submission
>  to upstream inclusion? I guess if you want to have a chance of getting

According to what I have heard from Morton. I need to submit very small
patches each adding one feature. The restructuring I am currently doing
aim at making this easier and will hopefully avoid having to maintain two
seperate code bases. I would like to isolate things a bit more so we would
be able to submit a patch that adds, let's per-thread counting, then system-wide
counting, then per-thread sampling and so on. That is not going to be
necessarily
easy.

>  it into 2.6.26, now would be the time to post the patches.
>  I can certainly offer some help getting it into shape for inclusion
>  if you're interested.
>
I don't think the code is yet ready to start submitting patches. We need
some more work to isolate things more. As I said the area of fous right
now should be the generic to arch-specific interface. There is also another

(Continue reading)

Arnd Bergmann | 1 Apr 2008 11:30

X-Face

Picon

Re: perfmon2 code restructuring news

On Tuesday 01 April 2008, stephane eranian wrote:
> On Tue, Apr 1, 2008 at 2:28 AM, Arnd Bergmann <arnd <at> arndb.de> wrote:
> > stephane eranian wrote:
> >  Ah, this looks really good. I haven't really followed the perfmon2
> >  progress over the last few releases. What is your plan for submission
> >  to upstream inclusion? I guess if you want to have a chance of getting
> 
> According to what I have heard from Morton. I need to submit very small
> patches each adding one feature. The restructuring I am currently doing
> aim at making this easier and will hopefully avoid having to maintain two
> seperate code bases. I would like to isolate things a bit more so we would
> be able to submit a patch that adds, let's per-thread counting, then system-wide
> counting, then per-thread sampling and so on. That is not going to be
> necessarily easy.

Note that splitting the patches for review doesn't mean that you have
to split the implementation into smaller files, although often these
go hand in hand.

Also, not every patch has to be meaningful by itself, the strict requirement
is just that adding one aspect of the code doesn't break anything.
E.g. one patch can add a file that can't be used by itself, as long as
it doesn't get enabled in the Makefile.
If it's possible to add functionality in separate units, that's much
better than just adding all files as separate patches and then adding
the Makefile, but OTOH the main purpose of the splitting is to make
reviewing the different aspects easier.

> >  it into 2.6.26, now would be the time to post the patches.
> >  I can certainly offer some help getting it into shape for inclusion

(Continue reading)

Group

gmane.comp.linux.perfmon2.devel.

Project Web Page

perfmon2 kernel and related tools developments ()

Search Archive

Page

1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 17

Articles in period: 168

April 2008

Language

Change language

Options

Current view: Threads only / Showing only 20 lines / Not hiding cited text.
Change to All messages, whole messages, or hide cited text.

Post a message
NNTP Newsgroup
Classic Gmane web interface
XML

XML

RSS Feed
List Information

About Gmane

Recent entries

pfmon for libpfm-4.3.0 (20 May 2013)
Ivy Bridge Uncore support (3 May 2013)
Ivy Bridge Uncore support (3 May 2013)
[PATCH] Power 7, change the code for the PM_RUN_INST_CMPL and PM_RUN_C (30 Apr 2013)
Ivy Bridge Uncore support (29 Apr 2013)
libpfm4 stats are too low (19 Apr 2013)
libpfm4 stats are too low (19 Apr 2013)
PMU and analysis tools workshop at ISCA 2013 (23 Mar 2013)
Will Intel Haswell support be available for libpfm4 and papi? (19 Mar 2013)
I/O monitoring using pfmon. (6 Mar 2013)
[PATCH] Add support for ARM 1176 Processor (7 Feb 2013)
[PATCH] Add the IBM POWER7+ processor support. (28 Jan 2013)
Task example does not work with PERF_COUNT_SW_* counters (23 Jan 2013)
Coverity results for libpfm sources in Fedora 19 (15 Jan 2013)
[PATCH] POWER 7, add comment to event file with link to documentation (4 Jan 2013)
Perfmon with Python 2.7 on Clarksfield processor. Unpredictable collec (7 Dec 2012)
[PATCH 2/2] Remove BITMODE stuff for ppc64 as it is no longer needed (5 Dec 2012)
[PATCH 1/2] Fix installation to honor PREFIX and LIBDIR (5 Dec 2012)
[PATCH 0/2] Fixes to 'make install' to better conform to standards (5 Dec 2012)
A simple question (pfmon2 is available in the linux kernel above 3.0?) (4 Dec 2012)
Example task_smpl doesn't work while with '-i' option (29 Nov 2012)

Archives

10	May 2013
8	April 2013
4	March 2013
4	February 2013
13	January 2013
11	December 2012
11	November 2012
4	October 2012
27	September 2012
13	August 2012
4	July 2012
13	June 2012
1	May 2012
25	April 2012
7	March 2012
2	February 2012
1	December 2011
3	November 2011
5	October 2011
13	September 2011
26	August 2011
31	July 2011
6	June 2011
21	May 2011
15	April 2011
89	March 2011
91	February 2011
37	January 2011
21	December 2010
82	November 2010
54	October 2010
110	September 2010
40	August 2010
53	July 2010
68	June 2010
117	May 2010
46	April 2010
92	March 2010
94	February 2010
97	January 2010
70	December 2009
67	November 2009
77	October 2009
59	September 2009
112	August 2009
89	July 2009
127	June 2009
55	May 2009
53	April 2009
40	March 2009
71	February 2009
77	January 2009
86	December 2008
33	November 2008
85	October 2008
86	September 2008
79	August 2008
148	July 2008
81	June 2008
151	May 2008
166	April 2008
59	March 2008

Design

Zawodny | Right Menu | Left Menu

Your Own Design

Paste the URL of your CSS below. Download CSS template