Unneeded kernel threads (xfs, jfs, gfs2)

Unneeded kernel threads (xfs, jfs, gfs2)

[Georgi Chorbadzhiyski]

Hi guys, I know that not all kernel thread users are converted to
start only when needed but the following is little annoying specially
when I'm not even using jfs, xfs and gfs2.

Look at the ps output from a standard distro 2.6.24.5 kernel after a
fresh boot:

>   PID TTY      STAT   TIME COMMAND
>     2 ?        S<     0:00 [kthreadd]
>     3 ?        S<     0:00  \_ [migration/0]
>     4 ?        S<     0:00  \_ [ksoftirqd/0]
>     5 ?        S<     0:00  \_ [migration/1]
>     6 ?        S<     0:00  \_ [ksoftirqd/1]
>     7 ?        S<     0:00  \_ [events/0]
>     8 ?        S<     0:00  \_ [events/1]
>     9 ?        S<     0:00  \_ [khelper]
>   102 ?        S<     0:00  \_ [kblockd/0]
>   103 ?        S<     0:00  \_ [kblockd/1]
>   106 ?        S<     0:00  \_ [kacpid]
>   107 ?        S<     0:00  \_ [kacpi_notify]
>   230 ?        S<     0:00  \_ [ata/0]
>   231 ?        S<     0:00  \_ [ata/1]
>   232 ?        S<     0:00  \_ [ata_aux]
>   233 ?        S<     0:00  \_ [ksuspend_usbd]
>   239 ?        S<     0:00  \_ [khubd]
>   242 ?        S<     0:00  \_ [kseriod]
>   284 ?        S      0:00  \_ [pdflush]
>   285 ?        S      0:00  \_ [pdflush]
>   286 ?        S<     0:00  \_ [kswapd0]
>   326 ?        S<     0:00  \_ [aio/0]
>   327 ?        S<     0:00  \_ [aio/1]
>   349 ?        S<     0:00  \_ [jfsIO]
>   350 ?        S<     0:00  \_ [jfsCommit]
>   351 ?        S<     0:00  \_ [jfsCommit]
>   352 ?        S<     0:00  \_ [jfsSync]

Not needed here, because no jfs partition is in use.

>   354 ?        S<     0:00  \_ [xfslogd/0]
>   355 ?        S<     0:00  \_ [xfslogd/1]
>   356 ?        S<     0:00  \_ [xfsdatad/0]
>   357 ?        S<     0:00  \_ [xfsdatad/1]
>   361 ?        S<     0:00  \_ [xfs_mru_cache]

Not needed here, because no xfs partition is in use.

>   366 ?        S<     0:00  \_ [gfs2_scand]
>   367 ?        S<     0:00  \_ [glock_workqueue]
>   368 ?        S<     0:00  \_ [glock_workqueue]

I dont use gfs2

>  1047 ?        S<     0:00  \_ [scsi_tgtd/0]
>  1048 ?        S<     0:00  \_ [scsi_tgtd/1]
>  1118 ?        S<     0:00  \_ [scsi_eh_2]
>  1120 ?        S<     0:00  \_ [scsi_eh_3]
>  1122 ?        S<     0:00  \_ [scsi_eh_4]
>  1124 ?        S<     0:00  \_ [scsi_eh_5]
>  1133 ?        S<     0:00  \_ [exec-osm/0]
>  1134 ?        S<     0:00  \_ [exec-osm/1]
>  1140 ?        S<     0:00  \_ [block-osm/0]
>  1141 ?        S<     0:00  \_ [block-osm/1]

What is osm? Can't find anything in kernel config?

>  1148 ?        S<     0:00  \_ [khpsbpkt]

What is this?

>  1193 ?        S<     0:00  \_ [ksnapd]
>  1201 ?        S<     0:00  \_ [rpciod/0]
>  1202 ?        S<     0:00  \_ [rpciod/1]
>  1204 ?        S<     0:00  \_ [kjournald]
>  2448 ?        S<     0:00  \_ [kjournald]
>  2649 ?        S      0:00  \_ [lockd]

Kernel config can be found here:
  http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config

-- 
Georgi Chorbadzhiyski
http://georgi.unixsol.org/

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Jan Engelhardt]

On Tuesday 2008-05-13 00:55, Georgi Chorbadzhiyski wrote:
>
>Not needed here, because no jfs partition is in use.
>Not needed here, because no xfs partition is in use.
>I dont use gfs2

You miss the best of Linux. So, why not unload the modules?

>>  1141 ?        S<     0:00  \_ [block-osm/1]
>
>What is osm? Can't find anything in kernel config?

drivers/message/i2o/i2o_block.c:#define OSM_NAME        "block-osm"

>>  1148 ?        S<     0:00  \_ [khpsbpkt]
>
>What is this?

ieee1394/ieee1394_core.c:static struct task_struct *khpsbpkt_thread;

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Georgi Chorbadzhiyski]

Jan Engelhardt mumbled something about, On 5/13/08 2:21 AM:
> On Tuesday 2008-05-13 00:55, Georgi Chorbadzhiyski wrote:
>> Not needed here, because no jfs partition is in use.
>> Not needed here, because no xfs partition is in use.
>> I dont use gfs2
> 
> You miss the best of Linux. So, why not unload the modules?

ext3 is working just fine for now, thanks (:

>>>  1141 ?        S<     0:00  \_ [block-osm/1]
>> What is osm? Can't find anything in kernel config?
> 
> drivers/message/i2o/i2o_block.c:#define OSM_NAME        "block-osm"
> 
>>>  1148 ?        S<     0:00  \_ [khpsbpkt]
>> What is this?
> 
> ieee1394/ieee1394_core.c:static struct task_struct *khpsbpkt_thread;

Strange names :)

-- 
Georgi Chorbadzhiyski
http://georgi.unixsol.org/

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Stefan Richter]

Georgi Chorbadzhiyski wrote:
> Jan Engelhardt mumbled something about, On 5/13/08 2:21 AM:
>> On Tuesday 2008-05-13 00:55, Georgi Chorbadzhiyski wrote:
>>>>  1148 ?        S<     0:00  \_ [khpsbpkt]
>>> What is this?
>> 
>> ieee1394/ieee1394_core.c:static struct task_struct *khpsbpkt_thread;
> 
> Strange names :)

Fixed in drivers/firewire/.  :-)

PS:
"k" = kernel
"hpsb" = High Performance Serial Bus ( = IEEE 1394)
"pkt" = packet  (The thread completes split transactions associated with
outbound request packets.)
-- 
Stefan Richter
-=====-==--- -=-= -===-
http://arcgraph.de/sr/

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Donald Douwsma]

Georgi Chorbadzhiyski wrote:
> Hi guys, I know that not all kernel thread users are converted to
> start only when needed but the following is little annoying specially
> when I'm not even using jfs, xfs and gfs2.
> 
> Look at the ps output from a standard distro 2.6.24.5 kernel after a
> fresh boot:
...

> 
> Kernel config can be found here:
>   http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config
> 

Your distro is building all of these modules into the kernel.
	CONFIG_XFS_FS=y
	CONFIG_JFS_FS=y
	CONFIG_GFS2_FS=y

This isnt exactly standard practice, normally they'd be set to =m and only
used if required to mount a filesystem. You may want to ask the slackware
people why they chose to do this for their hugexxx.s kernels.

Don

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[FD Cami]

On Tue, 13 May 2008 15:29:38 +1000
Donald Douwsma <donaldd@sgi.com> wrote:

> Georgi Chorbadzhiyski wrote:
> > Hi guys, I know that not all kernel thread users are converted to
> > start only when needed but the following is little annoying specially
> > when I'm not even using jfs, xfs and gfs2.
> > 
> > Look at the ps output from a standard distro 2.6.24.5 kernel after a
> > fresh boot:
> ...
> 
> > 
> > Kernel config can be found here:
> >   http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config
> > 
> 
> Your distro is building all of these modules into the kernel.
> 	CONFIG_XFS_FS=y
> 	CONFIG_JFS_FS=y
> 	CONFIG_GFS2_FS=y
> 
> This isnt exactly standard practice, normally they'd be set to =m and only
> used if required to mount a filesystem. You may want to ask the slackware
> people why they chose to do this for their hugexxx.s kernels.

To be able to use those filesystems on root partitions without requiring
an initrd. 

Georgi might be interested to use the generic kernel instead of the huge
one, it is available there :
http://mirrors.unixsol.org/slackware/slackware-12.1/slackware/a/kernel-generic-smp-2.6.24.5_smp-i686-2.tgz

Documentation :
http://mirrors.unixsol.org/slackware/slackware-12.1/README.TXT
http://mirrors.unixsol.org/slackware/slackware-12.1/README.initrd
http://mirrors.unixsol.org/slackware/slackware-12.1/RELEASE_NOTES

Cheers

Francois

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Georgi Chorbadzhiyski]

FD Cami mumbled something about, On 5/13/08 8:51 AM:
> On Tue, 13 May 2008 15:29:38 +1000
> Donald Douwsma <donaldd@sgi.com> wrote:
> 
>> Georgi Chorbadzhiyski wrote:
>>> Hi guys, I know that not all kernel thread users are converted to
>>> start only when needed but the following is little annoying specially
>>> when I'm not even using jfs, xfs and gfs2.
>>>
>>> Look at the ps output from a standard distro 2.6.24.5 kernel after a
>>> fresh boot:
>> ...
>>
>>> Kernel config can be found here:
>>>   http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config
>>>
>> Your distro is building all of these modules into the kernel.
>> 	CONFIG_XFS_FS=y
>> 	CONFIG_JFS_FS=y
>> 	CONFIG_GFS2_FS=y
>>
>> This isnt exactly standard practice, normally they'd be set to =m and only
>> used if required to mount a filesystem. You may want to ask the slackware
>> people why they chose to do this for their hugexxx.s kernels.
> 
> To be able to use those filesystems on root partitions without requiring
> an initrd. 
> 
> Georgi might be interested to use the generic kernel instead of the huge
> one, it is available there :
> http://mirrors.unixsol.org/slackware/slackware-12.1/slackware/a/kernel-generic-smp-2.6.24.5_smp-i686-2.tgz

I'm well aware of this kernels but I prefer to not use
initrd when possible and still have most of the functionality
available.

-- 
Georgi Chorbadzhiyski
http://georgi.unixsol.org/

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Georgi Chorbadzhiyski]

Donald Douwsma mumbled something about, On 5/13/08 8:29 AM:
> Georgi Chorbadzhiyski wrote:
>> Hi guys, I know that not all kernel thread users are converted to
>> start only when needed but the following is little annoying specially
>> when I'm not even using jfs, xfs and gfs2.
>>
>> Look at the ps output from a standard distro 2.6.24.5 kernel after a
>> fresh boot:
> ...
> 
>> Kernel config can be found here:
>>   
>> http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config 
> 
> Your distro is building all of these modules into the kernel.
>     CONFIG_XFS_FS=y
>     CONFIG_JFS_FS=y
>     CONFIG_GFS2_FS=y
> 
> This isnt exactly standard practice, normally they'd be set to =m and only
> used if required to mount a filesystem. You may want to ask the slackware
> people why they chose to do this for their hugexxx.s kernels.

I know that they are compiled in the kernel, but since they
are not used isn't starting their own kthreads kind of
unnecessary? Surely the threads can be started on demand
only when xfs/etc volume is mounted.

-- 
Georgi Chorbadzhiyski
http://georgi.unixsol.org/

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[David Chinner]

On Tue, May 13, 2008 at 11:30:59AM +0300, Georgi Chorbadzhiyski wrote:
> >>http://mirrors.unixsol.org/slackware/slackware-12.1/kernels/hugesmp.s/config 
> >
> >Your distro is building all of these modules into the kernel.
> >    CONFIG_XFS_FS=y
> >    CONFIG_JFS_FS=y
> >    CONFIG_GFS2_FS=y
> >
> >This isnt exactly standard practice, normally they'd be set to =m and only
> >used if required to mount a filesystem. You may want to ask the slackware
> >people why they chose to do this for their hugexxx.s kernels.
> 
> I know that they are compiled in the kernel, but since they
> are not used isn't starting their own kthreads kind of
> unnecessary? Surely the threads can be started on demand
> only when xfs/etc volume is mounted.

Sure - XFS will start another three kernel threads per filesystem
that gets mounted. And for good measure, it cleans them up again
on unmount. :)

The other threads are per-cpu workqueue threads that are shared
across all XFS filesystems in the system and hence are started
when XFS is initialised rather than when a mount occurs.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: [Jfs-discussion] Unneeded kernel threads (xfs, jfs, gfs2)

[Christoph Hellwig]

On Tue, May 13, 2008 at 07:03:11PM +1000, David Chinner wrote:
> Sure - XFS will start another three kernel threads per filesystem
> that gets mounted. And for good measure, it cleans them up again
> on unmount. :)
> 
> The other threads are per-cpu workqueue threads that are shared
> across all XFS filesystems in the system and hence are started
> when XFS is initialised rather than when a mount occurs.

Well, we could refcount the number of active xfs instances and
start/stop the global threads based on that.  Not really worth my
time IHMO, but if someone comes up with a clean enough patch it should
go in.

Re: [Jfs-discussion] Unneeded kernel threads (xfs, jfs, gfs2)

[Christoph Hellwig]

On Tue, May 13, 2008 at 05:21:56AM -0400, Christoph Hellwig wrote:
> Well, we could refcount the number of active xfs instances and
> start/stop the global threads based on that.  Not really worth my
> time IHMO, but if someone comes up with a clean enough patch it should
> go in.

Actually doing it in the VFS might be even better.  Add ->init and
->exit methods to struct file_system_type and then the filesystems can
move most of module_init/exit into the new methods.

Re: [Jfs-discussion] Unneeded kernel threads (xfs, jfs, gfs2)

[Eric Sandeen]

Christoph Hellwig wrote:
> On Tue, May 13, 2008 at 05:21:56AM -0400, Christoph Hellwig wrote:
>> Well, we could refcount the number of active xfs instances and
>> start/stop the global threads based on that.  Not really worth my
>> time IHMO, but if someone comes up with a clean enough patch it should
>> go in.
> 
> Actually doing it in the VFS might be even better.  Add ->init and
> ->exit methods to struct file_system_type and then the filesystems can
> move most of module_init/exit into the new methods.

Hmm I like that idea! :)

-Eric

Re: Unneeded kernel threads (xfs, jfs, gfs2)

[Lennart Sorensen]

On Tue, May 13, 2008 at 03:29:38PM +1000, Donald Douwsma wrote:
> Your distro is building all of these modules into the kernel.
> 	CONFIG_XFS_FS=y
> 	CONFIG_JFS_FS=y
> 	CONFIG_GFS2_FS=y
> 
> This isnt exactly standard practice, normally they'd be set to =m and only
> used if required to mount a filesystem. You may want to ask the slackware
> people why they chose to do this for their hugexxx.s kernels.

Because they have always done it that way so it has to be right? :)

-- 
Len Sorensen