Skip to contentSkip to conversation

RT exec for exercising RT kernel capabilities

DISCUSSION Mark Hounschell 5 participants 22 messages Jun 8, 2006 – Oct 3, 2006

lore
Share & tools
lore
#0 lore raw

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Regards
Mark Hounschell

#1 lore raw

Mark,

Quoted 6 lines · Mark Hounschell

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

Quoted 7 lines

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Nice tool.

Some remarks. You can build high resolution timer support without the extra lib package from the HRT sourceforge site. You need a recent glibc and some quirks in the source. See the cyclictest program I'm using.
http://www.tglx.de/projects/misc/cyclictest/cyclictest-v0.8.tar.bz2

It would also be cute to add tests for the PI support for pthread_mutexes.

tglx

#2 lore raw
Quoted 15 lines · Thomas Gleixner

Mark,

Quoted 12 lines · Mark Hounschell

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Quoted 9 lines
Quoted 1 line

Complaints, comments, or suggestions welcome.

Nice tool.

Some remarks. You can build high resolution timer support without the extra lib package from the HRT sourceforge site. You need a recent glibc and some quirks in the source. See the cyclictest program I'm using.
http://www.tglx.de/projects/misc/cyclictest/cyclictest-v0.8.tar.bz2

I didn't realize that. Right you are.

Quoted 4 lines

It would also be cute to add tests for the PI support for pthread_mutexes.

tglx

I'm not sure what one needs to do in user land to actually test that but I'll investigate.

Thanks
Mark

#6 lore raw
In reply to Mark Hounschell

jeudi 8 Juin 2006 21:12, Mark Hounschell wrote/a écrit :

Quoted 8 lines

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

In the README, you say :
...
The exec must be run as root because of the use of mlockall, sched_setscheduler, and sched_setaffinity calls. Sorry but there has been no attempt to use the Linux CAPABILITIES API so that it could be run as regular user.
...

It's false if you use the LSM patch from here :
http://sourceforge.net/projects/realtime-lsm/

Signature · 1 line
Serge Noiraud
#3 lore raw
In reply to Mark Hounschell
Quoted 19 lines · Mark Hounschell
Quoted 18 lines · Thomas Gleixner

Mark,

Quoted 6 lines · Mark Hounschell

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

Quoted 7 lines

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Nice tool.

Quoted 8 lines
Quoted 5 lines

Some remarks. You can build high resolution timer support without the extra lib package from the HRT sourceforge site. You need a recent glibc and some quirks in the source. See the cyclictest program I'm using.
http://www.tglx.de/projects/misc/cyclictest/cyclictest-v0.8.tar.bz2

I didn't realize that. Right you are.

Here is a new one that no longer requires the HRT sourceforge package. Thanks again.

ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.1.tar.bz2

Quoted 8 lines
Quoted 4 lines

It would also be cute to add tests for the PI support for pthread_mutexes.

tglx

I'm not sure what one needs to do in user land to actually test that but I'll investigate.

Still investigating...

Regards
Mark

#7 lore raw
In reply to Serge Noiraud
Quoted 10 lines · Serge Noiraud

In the README, you say :
...
The exec must be run as root because of the use of mlockall, sched_setscheduler, and sched_setaffinity calls. Sorry but there has been no attempt to use the Linux CAPABILITIES API so that it could be run as regular user.
...

It's false if you use the LSM patch from here :
http://sourceforge.net/projects/realtime-lsm/

Realtime LSM is deprecated, with a reasonably recent PAM and glibc (Ubuntu Dapper and FC5 should both be good) it's not necessary.

Just add something like:

*               -    rtprio  99
*               -    nice    -20
*               -    memlock 500000

to /etc/security/limits.conf

Lee

#4 lore raw
In reply to Mark Hounschell
Quoted 32 lines · Mark Hounschell
Quoted 17 lines · Mark Hounschell
Quoted 16 lines · Thomas Gleixner

Mark,

Quoted 13 lines · Mark Hounschell

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Quoted 9 lines
Quoted 7 lines

Nice tool.

Some remarks. You can build high resolution timer support without the extra lib package from the HRT sourceforge site. You need a recent glibc and some quirks in the source. See the cyclictest program I'm using.
http://www.tglx.de/projects/misc/cyclictest/cyclictest-v0.8.tar.bz2

I didn't realize that. Right you are.

Here is a new one that no longer requires the HRT sourceforge package. Thanks again.

ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.1.tar.bz2

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines

It would also be cute to add tests for the PI support for pthread_mutexes.

tglx

I'm not sure what one needs to do in user land to actually test that but I'll investigate.

Still investigating...

Am I even close in assuming that to enable the PI support you have to use

pthread_mutexattr_setprotocol(mutexattr, PTHREAD_PRIO_INHERIT);

I have only glibc 2.3 and 2.4 and neither of them understand what PTHREAD_PRIO_INHERIT is.

Can anyone point me to an some Doc or examples of how to enable PI for pthread_mutexes?

Sorry I'm ignorant on the subject. I do understand the principle of PI and may even be able to figure out how to test it in user land but how do I turn it on?

Thanks
Mark

#5 lore raw
Quoted 6 lines · Mark Hounschell
Quoted 1 line

Still investigating...

Am I even close in assuming that to enable the PI support you have to use

pthread_mutexattr_setprotocol(mutexattr, PTHREAD_PRIO_INHERIT);

Quoted 8 lines

I have only glibc 2.3 and 2.4 and neither of them understand what PTHREAD_PRIO_INHERIT is.

Can anyone point me to an some Doc or examples of how to enable PI for pthread_mutexes?

Sorry I'm ignorant on the subject. I do understand the principle of PI and may even be able to figure out how to test it in user land but how do I turn it on?

Sorry for letting you in the dark. There was a patch against glibc CVS, but it vansihed somehow. The support will go into glibc CVS when the PI futex support hits mainline in the 2.6.18 merge window,

tglx

#8 lore raw
In reply to Mark Hounschell

Hi,

Quoted 8 lines · Mark Hounschell

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

Sorry for the simple question. How do I check if my hardware support HRT? Is it common on i386 desktops (mine is an Athlon XP 2000+)?

Many thanks.

Cheers,

~ Antonio

#9 lore raw
In reply to Mark Hounschell

jeudi 8 Juin 2006 21:12, Mark Hounschell wrote/a écrit :

Quoted 6 lines

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

Quoted 7 lines

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Great tool.

I ran ./go and after 16:50 hours, the tasks 9 and 10 ( type hrt ) jumped respectively to 9787 and 3843usec ! 5 or 10 minutes earlier I did have between 100 and 200 usec. What can cause this problem ? is it a hrt problem ?
Could it be a driver problem ( network ) ?

I send the png in copy.

Signature · 1 line
Serge Noiraud
#10 lore raw
Quoted 21 lines · Serge Noiraud
Quoted 6 lines · Mark Hounschell wrote/

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests well the deterministic real-time capabilities of a computer. Maybe it could useful in some way to the effort or to anyone interested in making this type of determination about their kernel/computer.

Quoted 7 lines

A README describing the rt-exec can be found at
ftp://ftp.compro.net/public/rt-exec/README

It can be downloaded from ftp://ftp.compro.net/public/rt-exec/rt-exec-1.0.0.tar.bz2

Complaints, comments, or suggestions welcome.

Great tool.

I ran ./go and after 16:50 hours, the tasks 9 and 10 ( type hrt ) jumped respectively to 9787 and 3843usec ! 5 or 10 minutes earlier I did have between 100 and 200 usec. What can cause this problem ? is it a hrt problem ?
Could it be a driver problem ( network ) ?

Quoted 6 lines

I send the png in copy.

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

All I can say about those 2 big hits is that your system burped. Why, I can't tell you. The people working on the rt stuff may be able to help you.

What bugs me is task16 is not running at all. Tell what kernel and glibc you are using please?

Mark

#11 lore raw

mardi 13 Juin 2006 16:34, Mark Hounschell wrote/a écrit :

Quoted 4 lines
Quoted 3 lines · Serge Noiraud
Quoted 2 lines · Mark Hounschell wrote/

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests

...

Quoted 5 lines
Quoted 5 lines

I ran ./go and after 16:50 hours, the tasks 9 and 10 ( type hrt ) jumped respectively to 9787 and 3843usec ! 5 or 10 minutes earlier I did have between 100 and 200 usec. What can cause this problem ? is it a hrt problem ?
Could it be a driver problem ( network ) ?

...

Quoted 6 lines

All I can say about those 2 big hits is that your system burped. Why, I can't tell you. The people working on the rt stuff may be able to help you.

What bugs me is task16 is not running at all. Tell what kernel and glibc you are using please?

I forgot to ask you this problem.

I'm using linux 2.6.16 with rt28.
glibc-2.3.3

Quoted 2 lines

Mark

Signature · 1 line
Serge Noiraud
#12 lore raw
Quoted 14 lines · Serge Noiraud
Quoted 4 lines · Mark Hounschell wrote/
Quoted 3 lines · Serge Noiraud
Quoted 2 lines · Mark Hounschell wrote/

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests

...

Quoted 4 lines
Quoted 4 lines

I ran ./go and after 16:50 hours, the tasks 9 and 10 ( type hrt ) jumped respectively to 9787 and 3843usec ! 5 or 10 minutes earlier I did have between 100 and 200 usec. What can cause this problem ? is it a hrt problem ?
Could it be a driver problem ( network ) ?

...

Quoted 3 lines

All I can say about those 2 big hits is that your system burped. Why, I can't tell you. The people working on the rt stuff may be able to help you.

Quoted 8 lines
Quoted 2 lines

What bugs me is task16 is not running at all. Tell what kernel and glibc you are using please?

I forgot to ask you this problem.

I'm using linux 2.6.16 with rt28.
glibc-2.3.3

Quoted 1 line

Mark

This is a bug in the exec. I'll fix it ASAIC.

Mark

#13 lore raw
Quoted 19 lines · Mark Hounschell
Quoted 18 lines · Serge Noiraud
Quoted 4 lines · Mark Hounschell wrote/
Quoted 3 lines · Serge Noiraud
Quoted 2 lines · Mark Hounschell wrote/

With the ongoing work being done to rt kernel enhancements by Ingo and friends, I would like to offer the use of a user land test (rt-exec). The rt-exec tests

...

Quoted 4 lines
Quoted 4 lines

I ran ./go and after 16:50 hours, the tasks 9 and 10 ( type hrt ) jumped respectively to 9787 and 3843usec ! 5 or 10 minutes earlier I did have between 100 and 200 usec. What can cause this problem ? is it a hrt problem ?
Could it be a driver problem ( network ) ?

...

Quoted 5 lines

All I can say about those 2 big hits is that your system burped. Why, I can't tell you. The people working on the rt stuff may be able to help you.

What bugs me is task16 is not running at all. Tell what kernel and glibc you are using please?

I forgot to ask you this problem.

Quoted 8 lines
Quoted 4 lines

I'm using linux 2.6.16 with rt28.
glibc-2.3.3

Quoted 1 line

Mark

This is a bug in the exec. I'll fix it ASAIC.

Mark

There is a new version there that should fix that.

Mark

#14 lore raw
In reply to Mark Hounschell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

CONFIG_HIGH_RES_TIMERS=y, CONFIG_HIGH_RES_RESOLUTION=1000:

# Run 00:40:12:784  Posix-hrt   14:27:38  Work:200   CPU:14 Avg:14 Max:23 Pg:0
# DataPool:SHM      Exec Heart Beat Rate:1000Hz      Exec Revision:1.0.3-1
###############################################################################
#          Task Sched Cpu     Intr   Late         -Interrupt Latencies (usec)- 
# Taskname Type  Pr P Mask     Cnt    Cnt   Spare Current  Best   Worst  Determ
# exec      hrt   5 F  2   2412784      0     871       2     1      18      17
# task1     dth  29 F  2    603196      0    3993      19    17      70      53
# task2     dth  28 F  2    603196      1    3993      19    17     128     111
# task3     sth  25 F  2   1206392      0    1993      18    14      40      26
# task4     sth  24 F  2   1206392      0    1993      18    13      37      24
# task5     sem  23 F  2   2412784      0     993       4     3      28      25
# task6     sem  22 F  2   1206392      0    1993       4     3      10       7
# task7     sig  19 F  2   1206392      0    1993       7     7      17      10
# task8     sig  18 F  2   1206392      0    1993       8     6      18      12
# task9     hrt  17 R  2   1017659      0    1993     763   712    2185    1473
# task10    hrt  17 R  2   1017798      0    1993     771   717   13879   13162
# task11    hrn  14 R  2   1586789      0     993    1060   690   15817   15127
# task12    hrn  14 R  2   1586620      0     993    1024   689   15743   15054
# task13    hru  11 R  2   1557359      0     993    1053   689    2972    2283
# task14    hru  11 R  2   1561189      0     993    1018   690    3010    2320
# task15    bth   7 R  2   2412784      0     994      29    13      55      42
# task16    bth   7 R  2   2412784      0     993      14    13      56      43

CONFIG_HIGH_RES_TIMERS=n:

# Run 00:03:59:384  Posix-hrt   18:24:45  Work:200   CPU:14 Avg:14 Max:19 Pg:0
# DataPool:SHM      Exec Heart Beat Rate:1000Hz      Exec Revision:1.0.3-1
###############################################################################
#          Task Sched Cpu     Intr   Late         -Interrupt Latencies (usec)-
# Taskname Type  Pr P Mask     Cnt    Cnt   Spare Current  Best   Worst  Determ
# exec      hrt   5 F  2    239384      0     870       1     1      13      12
# task1     dth  29 F  2     59847      0    3993      19    17      39      22
# task2     dth  28 F  2     59846      0    3993      19    17      39      22
# task3     sth  25 F  2    119693      0    1993      18    17      37      20
# task4     sth  24 F  2    119692      0    1993      19    14      41      27
# task5     sem  23 F  2    239385      0     993       5     4      26      22
# task6     sem  22 F  2    119693      0    1993       5     4       9       5
# task7     sig  19 F  2    119693      0    1993       8     7      15       8
# task8     sig  18 F  2    119692      0    1993       8     6      15       9
# task9     hrt  17 R  2    104060      0    1992    1108   724   11439   10715
# task10    hrt  17 R  2    104057      0    1993    1102   726   12923   12197
# task11    hrn  14 R  2    161028      0     994    1056   705   13001   12296
# task12    hrn  14 R  2    160951      0     993    1019   704   12456   11752
# task13    hru  11 R  2    158334      0     993    1054   744    3109    2365
# task14    hru  11 R  2    158729      0     993    1020   709    3066    2357
# task15    bth   7 R  2    239385      0     993      29    13      54      41
# task16    bth   7 R  2    239385      0     993      15    13      55      42

Any idea what could be wrong?

Lee

#15 lore raw
Quoted 6 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

tglx

#16 lore raw
Quoted 9 lines · Thomas Gleixner
Quoted 6 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

Sent off-list.

Lee

#17 lore raw
Quoted 4 lines · Lee Revell
Quoted 3 lines · Thomas Gleixner
Quoted 2 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

Quoted 10 lines
Quoted 4 lines
Quoted 3 lines

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

Sent off-list.

Lee

Did you get this sorted out?

Mark

#18 lore raw
Quoted 10 lines · Mark Hounschell
Quoted 9 lines · Lee Revell
Quoted 7 lines · Thomas Gleixner
Quoted 5 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

Quoted 7 lines
Quoted 5 lines

Sent off-list.

Lee

Did you get this sorted out?

No. Thomas, were you able to test my .config?

Lee

#19 lore raw
In reply to Lee Revell
Quoted 5 lines · Lee Revell
Quoted 4 lines · Thomas Gleixner
Quoted 3 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

Sent off-list.

I've discovered the problem - I was running rt-exec as a non-root user.

Is the use of high res timers limited to root?

Lee

#20 lore raw
Quoted 10 lines · Lee Revell
Quoted 9 lines · Lee Revell
Quoted 8 lines · Thomas Gleixner
Quoted 6 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Can you please send me a boot log ?

Sent off-list.

I've discovered the problem - I was running rt-exec as a non-root user.

Is the use of high res timers limited to root?

No. I guess the priority setting is not working as non-root.

tglx

#21 lore raw
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Lee Revell
Quoted 8 lines · Lee Revell
Quoted 7 lines · Thomas Gleixner
Quoted 5 lines · Lee Revell
Quoted 1 line · Mark Hounschell

ftp://ftp.compro.net/public/rt-exec/

The high res timers do not seem to be working. Using 2.6.17-rt3 and glibc 2.3.6, I get the exact same (bad) results whether CONFIG_HIGH_RES_TIMERS is enabled or not.

Can you please send me a boot log ?

Quoted 9 lines
Quoted 4 lines
Quoted 1 line

Sent off-list.

I've discovered the problem - I was running rt-exec as a non-root user.

Is the use of high res timers limited to root?

No. I guess the priority setting is not working as non-root.

tglx

I think I put something in the README about this? Yep..

Mark