OProfile

On Wed, May 15, 2002 at 10:58:06AM -0400, William Cohen wrote:

> I have made a trivial patch to allow oprofpp to sort the entries in 
> reverse order just like op_time for the "-l" option. This allows one to 
> put the interesting information all at the top of the list for: 
> information about the metric being measured and the largest samples.

I suppose there is no harm in taking this now. I'll do it soon.

All the frontends are about to have a re-designed set of options etc.
but we may as well include this now anyway

thanks,
john

-- 
"So what you're saying is "screw the disabled" and you want us to do the
 same ?  No thanks..."
	- Ian Hixie, bug 25537

On Wed, May 15, 2002 at 10:58:06AM -0400, William Cohen wrote:

> I have made a trivial patch to allow oprofpp to sort the entries in 
> reverse order just like op_time for the "-l" option. This allows one to 

I applied your patch, I also added doc for it in the man page, and
disallowed its usage when -l is not used

>  	samples.select_symbols(symbols, sort_by_ctr, 0.0, false, true);
>  
> -	out.Output(cout, symbols, false);
> +	out.Output(cout, symbols, reverse_sort != 0);

I don't think this makes sense, since it is ordered by vma - why would
it be useful ?

regards
john

-- 
"So what you're saying is "screw the disabled" and you want us to do the
 same ?  No thanks..."
	- Ian Hixie, bug 25537

Yes, you are right that it doesn't make much sense for sorting the vmas. 
It would be nice to order things for the "-L" based on the number of 
samples for the function like "-l". I realize that the "-L" didn't work 
the way I expected after I sent the patch. I have been using "-l" option 
to see what is going on.

The front-end is going to be redesigned? Do you have a working document 
describing the options and how things are going to function and the 
general architecture?

-Will


John Levon wrote:

> On Wed, May 15, 2002 at 10:58:06AM -0400, William Cohen wrote:
> 
> 
>>I have made a trivial patch to allow oprofpp to sort the entries in 
>>reverse order just like op_time for the "-l" option. This allows one to 
>>
> 
> I applied your patch, I also added doc for it in the man page, and
> disallowed its usage when -l is not used
> 
> 
>> 	samples.select_symbols(symbols, sort_by_ctr, 0.0, false, true);
>> 
>>-	out.Output(cout, symbols, false);
>>+	out.Output(cout, symbols, reverse_sort != 0);
>>
> 
> I don't think this makes sense, since it is ordered by vma - why would
> it be useful ?
> 
> regards
> john
> 
>

On Wed, May 15, 2002 at 12:07:09PM -0400, William Cohen wrote:

> It would be nice to order things for the "-L" based on the number of 
> samples for the function like "-l". I realize that the "-L" didn't work 

Yes, it would. I expect this in the re-design.

> The front-end is going to be redesigned? Do you have a working document 
> describing the options and how things are going to function and the 
> general architecture?

Not yet, but I will. Me and Phil have discussed a few things, but we
need to draw up a spec and bullet proof it. It will get posted to the
list for comments when an initial draft is ready.

regards
john

-- 
"So what you're saying is "screw the disabled" and you want us to do the
 same ?  No thanks..."
	- Ian Hixie, bug 25537

Bugs item #555276, was opened at 2002-05-13 04:31
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555276&group_id=16191

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: William Cohen (wcohen)
>Assigned to: Philippe Elie (phil_e)
Summary: oprofpp needs samples on executable

Initial Comment:
OProfile allows samples for shared libraries to be
attributed to individual executables with the
"--separate-samples". For a program that spends a
majority of its time in the shared libraries, e.g.
mozilla, there may be samples for the shared libraries,
but not be any samples for the main executable. The
file search logic for oprofpp fails when there are
samples for the shared library, but no samples for the
main executable. When this occurs something like the
following error message is printed.

oprofpp: Opening
/var/lib/oprofile/samples/}usr}lib}mozilla}mozilla-bin#1
failed. No such file or directory

Better behavior would be for oprofpp to assume there
are zero samples on the executable and not give an
error. For some events it would be possible that they
were never observed during the execution of the
program, e.g. use of mmx or floating point instructions.

-Will


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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555276&group_id=16191

Bugs item #555276, was opened at 2002-05-13 04:31
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555276&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 5
Submitted By: William Cohen (wcohen)
Assigned to: Philippe Elie (phil_e)
Summary: oprofpp needs samples on executable

Initial Comment:
OProfile allows samples for shared libraries to be
attributed to individual executables with the
"--separate-samples". For a program that spends a
majority of its time in the shared libraries, e.g.
mozilla, there may be samples for the shared libraries,
but not be any samples for the main executable. The
file search logic for oprofpp fails when there are
samples for the shared library, but no samples for the
main executable. When this occurs something like the
following error message is printed.

oprofpp: Opening
/var/lib/oprofile/samples/}usr}lib}mozilla}mozilla-bin#1
failed. No such file or directory

Better behavior would be for oprofpp to assume there
are zero samples on the executable and not give an
error. For some events it would be possible that they
were never observed during the execution of the
program, e.g. use of mmx or floating point instructions.

-Will


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

>Comment By: Philippe Elie (phil_e)
Date: 2002-05-15 04:03

Message:
Logged In: YES 
user_id=318973

fixed in cvs, in a slightly different way, when you
request from app and shared lib from a profiling
session with --separate-samples and the application
don't get any samples, the application is just ignored
and so on you see report only for shared libs

Phil


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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555276&group_id=16191

Bugs item #555304, was opened at 2002-05-13 07:14
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555304&group_id=16191

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: Philippe Elie (phil_e)
>Assigned to: Philippe Elie (phil_e)
Summary: non-native compiler can fail to compile

Initial Comment:
when using a non-native compiler the module can fail
to compile because configure.in check compiler
features through "gcc" while module is compiled with
cc: on some system cc can point to the native compiler
not the user selected gcc version through PATH.

This is fixed in cvs

Phil


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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555304&group_id=16191

Bugs item #555304, was opened at 2002-05-13 07:14
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555304&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 5
Submitted By: Philippe Elie (phil_e)
Assigned to: Philippe Elie (phil_e)
Summary: non-native compiler can fail to compile

Initial Comment:
when using a non-native compiler the module can fail
to compile because configure.in check compiler
features through "gcc" while module is compiled with
cc: on some system cc can point to the native compiler
not the user selected gcc version through PATH.

This is fixed in cvs

Phil


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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555304&group_id=16191

Bugs item #555302, was opened at 2002-05-13 07:05
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555302&group_id=16191

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: Philippe Elie (phil_e)
>Assigned to: Philippe Elie (phil_e)
Summary: oprofile: can't get RTC I/O Ports

Initial Comment:
redhat-7.2, kernel 2.4.17
ibm thinkpad r30

oprofile: can't get RTC I/O Ports
then the module fail to load with an EBUSY error

I open a bug for clarification only, the doc said
"you cannot have both the oprofile and rtc modules
loaded" but it means you can't have the rtc module
loaded neither RTC compiled in the kernel. The doc is
corrected in cvs.

Phil

P.S: john we need to add the FORCE_RTC patch from bob
it's usefull to check RTC.



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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555302&group_id=16191

Bugs item #555302, was opened at 2002-05-13 07:05
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555302&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 5
Submitted By: Philippe Elie (phil_e)
Assigned to: Philippe Elie (phil_e)
Summary: oprofile: can't get RTC I/O Ports

Initial Comment:
redhat-7.2, kernel 2.4.17
ibm thinkpad r30

oprofile: can't get RTC I/O Ports
then the module fail to load with an EBUSY error

I open a bug for clarification only, the doc said
"you cannot have both the oprofile and rtc modules
loaded" but it means you can't have the rtc module
loaded neither RTC compiled in the kernel. The doc is
corrected in cvs.

Phil

P.S: john we need to add the FORCE_RTC patch from bob
it's usefull to check RTC.



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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555302&group_id=16191

Bugs item #555268, was opened at 2002-05-13 04:08
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555268&group_id=16191

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: Nobody/Anonymous (nobody)
Assigned to: Nobody/Anonymous (nobody)
Summary: oprof_start leaks file descriptors

Initial Comment:
The gui program oprof_start leaks file descriptors in
the 0.2 and the current snapshot (20020512) of
OProfile. oprof_start will continue to open new file
descriptors to "/proc" (about 12 a minute on my
machine) and never close old file descriptors to
"/proc". This can be observed by examining the fd
directory for the oprof_start process in /proc. When
the file descriptor limit for the process is reached an
error is produced:

oprofiled: /proc directory could not be opened. : Too
many open files

oprof_start exits immediately after the error message
is printed.

-Will

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

>Comment By: Philippe Elie (phil_e)
Date: 2002-05-13 06:37

Message:
Logged In: YES 
user_id=318973

this is corrected in the current cvs, it's leak due to a 
missing closedir

Phil

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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555268&group_id=16191

Bugs item #555268, was opened at 2002-05-13 04:08
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555268&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 5
Submitted By: Nobody/Anonymous (nobody)
Assigned to: Nobody/Anonymous (nobody)
Summary: oprof_start leaks file descriptors

Initial Comment:
The gui program oprof_start leaks file descriptors in
the 0.2 and the current snapshot (20020512) of
OProfile. oprof_start will continue to open new file
descriptors to "/proc" (about 12 a minute on my
machine) and never close old file descriptors to
"/proc". This can be observed by examining the fd
directory for the oprof_start process in /proc. When
the file descriptor limit for the process is reached an
error is produced:

oprofiled: /proc directory could not be opened. : Too
many open files

oprof_start exits immediately after the error message
is printed.

-Will

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

Comment By: Philippe Elie (phil_e)
Date: 2002-05-13 06:37

Message:
Logged In: YES 
user_id=318973

this is corrected in the current cvs, it's leak due to a 
missing closedir

Phil

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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555268&group_id=16191

Bugs item #555694, was opened at 2002-05-14 00:47
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555694&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 5
Submitted By: Cliff Woolley (jwoolley)
Assigned to: Nobody/Anonymous (nobody)
Summary: [PATCH] get HEAD to compile with Qt3

Initial Comment:
Hey... I had to make the following change to get HEAD 
from the CVS repository to compile with Qt3:

Index: gui/oprof_start.cpp
=======================================================
============
RCS 
file: /cvsroot/oprofile/oprofile/gui/oprof_start.cpp,v
retrieving revision 1.48
diff -u -d -r1.48 oprof_start.cpp
--- gui/oprof_start.cpp 9 May 2002 20:34:07 -
0000       1.48
+++ gui/oprof_start.cpp 14 May 2002 00:46:14 -0000
@@ -19,6 +19,8 @@
 #include <fstream>
 #include <algorithm>
 
+#include <qlineedit.h>
+#include <qlistview.h>
 #include <qcombobox.h> 
 #include <qlistbox.h>
 #include <qfiledialog.h>


--Cliff Woolley
jwoolley@...

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

>Comment By: John Levon (movement)
Date: 2002-05-14 15:54

Message:
Logged In: YES 
user_id=53034

Thanks !

I have applied (but not tested) your patch now.

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

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=555694&group_id=16191

On Thu, May 09, 2002 at 03:46:08PM +0200, Johan Walles wrote:

> I have just given OProfile a very superficial test run, and I'm very
> impressed.  It built and installed smoothly (on my 2.4.17 based Debian
> Testing system) and the results are very interesting.  You have created a
> great utility.

Thanks !

>   Keep it up!

We intend to :)

regards
john