Andy Fleming | 1 Apr 21:55 2003

build fails for ppc oprofile

So I have the oprofile from cvs, updated monday morning, CST.  I applied
v0004 of the ppc patch, and I have added preliminary support for the
7455.  I can't tell if it works, though, since I can't seem to get
oprofile to compile.  I get an error:

gcc -D__KERNEL__ -I/home/afleming/linux-2.4.20-8a-perfmon/include -Wall
-Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common
-fomit-frame-pointer -I/home/afleming/linux-2.4.20-8a-perfmon/arch/ppc
-fsigned-char -msoft-float -pipe -ffixed-r2 -Wno-uninitialized
-mmultiple -mstring -DMODULE -DMODVERSIONS -include
/home/afleming/linux-2.4.20-8a-perfmon/include/linux/modversions.h
-DHAVE_LINUX_SPINLOCK_HEADER  -DRTC_LOCK -DEXPECT_OK -D__NO_VERSION__
-I/home/afleming/projects/perfmon/oprofile-ppc/
-I/home/afleming/projects/perfmon/oprofile-ppc/libutil
-I/home/afleming/projects/perfmon/oprofile-ppc/libop
-I/home/afleming/projects/perfmon/oprofile-ppc/module -nostdinc
-iwithprefix include -DKBUILD_BASENAME=op_syscalls  -c -o op_syscalls.o
op_syscalls.c
In file included from
/home/afleming/linux-2.4.20-8a-perfmon/include/linux/sched.h:24,
                 from op_syscalls.c:11:
/home/afleming/linux-2.4.20-8a-perfmon/include/linux/tty.h:346: invalid
suffix on floating constant
/home/afleming/linux-2.4.20-8a-perfmon/include/linux/tty.h:346: parse
error before numeric constant
/home/afleming/linux-2.4.20-8a-perfmon/include/linux/tty.h:346: warning:
function declaration isn't a prototype
op_syscalls.c:236: warning: `my_sys_fork' defined but not used
op_syscalls.c:254: warning: `my_sys_vfork' defined but not used
op_syscalls.c:272: warning: `my_sys_clone' defined but not used
(Continue reading)

Bryan Rittmeyer | 2 Apr 22:54 2003

Re: build fails for ppc oprofile

On Tue, Apr 01, 2003 at 01:55:44PM -0600, Andy Fleming wrote:
> -#define UTS_RELEASE "2.4.20-8a-perfmon"
> +#define UTS_RELEASE "2.4.20-8a"

hmm, I've never seen a double suffix in UTS_RELEASE.
does it compile with "2.4.20-8a"?

-Bryan

-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/
Andy Fleming | 2 Apr 23:14 2003

Re: build fails for ppc oprofile

On Wed, 2003-04-02 at 14:54, Bryan Rittmeyer wrote:
> On Tue, Apr 01, 2003 at 01:55:44PM -0600, Andy Fleming wrote:
> > -#define UTS_RELEASE "2.4.20-8a-perfmon"
> > +#define UTS_RELEASE "2.4.20-8a"
> 
> hmm, I've never seen a double suffix in UTS_RELEASE.
> does it compile with "2.4.20-8a"?
> 
> -Bryan

The kernel compiles either way, but the modules won't compile, including
oprofile.  They also don't compile with 2.4.20-8a, which is what is
installed in /usr/src on my machine.

I'm using gcc version 3.2.2 20030217 (Yellow Dog Linux 3.0 3.2.2-2a)

I'm going to try downloading a fresh release, and applying the patches
manually, and see what happens.  I'm currently running the kernel with
your perfmon patch on my PBG4.  Of course, without oprofile, it's kind
of useless.  Sigh.

Andy

-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/
Bryan Rittmeyer | 2 Apr 23:25 2003

Re: build fails for ppc oprofile

On Tue, Apr 01, 2003 at 01:55:44PM -0600, Andy Fleming wrote:
> -iwithprefix include -DKBUILD_BASENAME=op_syscalls  -c -o op_syscalls.o
> op_syscalls.c
> In file included from
> /home/afleming/linux-2.4.20-8a-perfmon/include/linux/sched.h:24,
>                  from op_syscalls.c:11:
> /home/afleming/linux-2.4.20-8a-perfmon/include/linux/tty.h:346: invalid
> suffix on floating constant

ok, can you e-mail me a couple lines around your tty.h:346, and try the
benh kernel from http://www.penguinppc.org/dev/kernel.shtml (which is what
I'm developing against)

you may also be able to nudge the compile by #if 0'ing out everything except
for empty op_[intercept|save|restore]_syscalls in op_syscall.c

-Bryan

-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/
Abhishek Rai | 5 Apr 08:00 2003
Picon

regarding hash table size

hi all,
i m using oprofile to profile the kernel - more
specifically an nfs client i m evaluating. but i keep
getting these error message -

oprofile: Detected overflow of size ... You mush
increase the hash table size of reduce the interrupt
frequency.

I tried looking for hash table size and the only
seemingly relavant reference i found was in
dae/opd_proc.c as
#define OPD_MAX_PROC_HASH   1024
i changed this to as far as 8192, but no escaping from
those error messages.
My settings go as: 
--ctr0-event=CPU_CLOCK_UNHALTED --ctr0-count=6000

so please tell me wat i should do? change hash table
size or do sth about the frequency. if hash table-then
how.
thanks a lot
abhishek

=====
"They thought to use and shame me but I win out by nature, because a true
freak cannot be made.  A true freak must be born." - K. Dunn
============================================================
Abhishek Rai
Final year,B.Tech, Computer Science and Engineering
(Continue reading)

Philippe Elie | 5 Apr 17:01 2003
Picon

Re: regarding hash table size

Abhishek Rai wrote:
> hi all,
> i m using oprofile to profile the kernel - more
> specifically an nfs client i m evaluating. but i keep
> getting these error message -
> 
> oprofile: Detected overflow of size ... You mush
> increase the hash table size of reduce the interrupt
> frequency.
> 
> I tried looking for hash table size and the only
> seemingly relavant reference i found was in
> dae/opd_proc.c as
> #define OPD_MAX_PROC_HASH   1024
> i changed this to as far as 8192, but no escaping from
> those error messages.
> My settings go as: 
> --ctr0-event=CPU_CLOCK_UNHALTED --ctr0-count=6000

message in module and option defined in user space
are confusing, the message would be:

oprofile: Detected overflow of size %d. You must increase
the *buffer* size or reduce the interrupt frequency.

I'm re-wording this message in cvs.

this parameter is defined in libop/op_config_24.h

/** 65536 * sizeof(op_sample) */
(Continue reading)

SourceForge.net | 5 Apr 19:47 2003
Picon
Picon

[ oprofile-Bugs-715923 ] box freeze when passing incorrect parameters to opcontrol

Bugs item #715923, was opened at 2003-04-05 19:47
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=116191&aid=715923&group_id=16191

Category: None
Group: None
Status: Open
Resolution: None
Priority: 7
Submitted By: Philippe Elie (phil_e)
Assigned to: Philippe Elie (phil_e)
Summary: box freeze when passing incorrect parameters to opcontrol

Initial Comment:
normal operation for opcontrol --ctrx-count=value is to
validate
value and reject runs if value is too low cause it can
freeze
the box due to nmi flood.

2.5/2.4 daemon never validate this value.

2.5 module never validate it.
2.4 validation code has been removed from module
in op_nmi.c Sun Mar 2 2003

this bug is in all oprofile version for 2.5 kernel but only
in 0.5.2cvs for 2.4 module.

work-around: don't pass foolish count value.
(Continue reading)

SourceForge.net | 5 Apr 20:42 2003
Picon
Picon

[ oprofile-Bugs-715923 ] box freeze when passing incorrect parameters to opcontrol

Bugs item #715923, was opened at 2003-04-05 19:47
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=116191&aid=715923&group_id=16191

Category: None
Group: None
Status: Open
>Resolution: Fixed
Priority: 7
Submitted By: Philippe Elie (phil_e)
Assigned to: Philippe Elie (phil_e)
Summary: box freeze when passing incorrect parameters to opcontrol

Initial Comment:
normal operation for opcontrol --ctrx-count=value is to
validate
value and reject runs if value is too low cause it can
freeze
the box due to nmi flood.

2.5/2.4 daemon never validate this value.

2.5 module never validate it.
2.4 validation code has been removed from module
in op_nmi.c Sun Mar 2 2003

this bug is in all oprofile version for 2.5 kernel but only
in 0.5.2cvs for 2.4 module.

work-around: don't pass foolish count value.
(Continue reading)

Abhishek Rai | 5 Apr 21:34 2003
Picon

Re: regarding hash table size

hi all,
i tried Philippe's solution. but when i tried to start
oprofile after having altered the buffersize in
op_config_24.h, i've been getting always the following
message:
Failed to open hash map device: Operation not
permitted
Couldn't start oprofiled.
Check the log file "/var/lib/oprofile/oprofiled.log"
and /var/log/messages

i think this "hash map device" is
/var/lib/oprofile/ophashmapdev
so, what has gone wrong? <<My oprofile worked just
fine earlier>>. i then tried the following command
cat /var/lib/oprofile/ophashmapdev and got the same
"Operation not permitted". Moreover
/var/lib/oprofile/oprofiled.log is zero lenght while
/var/log/messages isn't showing anything related to
oprofile.

Next, i tried Philippe's solution on another machine
and it worked. There even 'cat
/var/lib/oprofile/ophashmapdev' "worked" as i got some
other error msg: some invalid parameter thing

Hence the problem is with some configuration on my
machine. Note that oprofile was working just fine on
my machine previously except for the buffer overflow
thing  fwhich i was just going to fix.
(Continue reading)

John Levon | 5 Apr 21:42 2003

Re: regarding hash table size

On Sat, Apr 05, 2003 at 11:34:34AM -0800, Abhishek Rai wrote:

> i tried Philippe's solution. but when i tried to start
> oprofile after having altered the buffersize in
> op_config_24.h, i've been getting always the following
> message:
> Failed to open hash map device: Operation not
> permitted

This is a (very) occassional problem with the 2.4 module code that
nobody has tracked down yet. It's probably something obvious but the bug
is as yet unfixed, sorry.

regards
john

p.s. please edit your posts appropriately

-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Gmane