2.6.8 Thinkpad T40, clock running too fast

2.6.8 Thinkpad T40, clock running too fast

[Shawn Willden]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I'm having a problem with the clock on my Thinkpad running too fast.  My clock
is gaining about 4 seconds every five minutes of operation.  I'm actually not
sure when this started because NTP has been fairly successful at keeping my
clock under control, so I didn't really notice it until I spent some time
operating disconnected... and then it became very obvious very quickly.

On October 15th, I "fixed" the problem by booting with "noapic" on the command
line.  Or I thought I fixed it anyway.  I had to reboot yesterday for an
unrelated issue (a "stuck" smb share that I couldn't figure out how to
unmount) and the racing clock is back.  I've booted with "noapic", "nolapic",
"noapic nolapic" and no options at all, and the clock doesn't keep the time
with any of them.  I'm not sure if it's always too fast.  I seem to remember
that last month it was running too slow.  I'll check with the various apic
settings and see if it's consistent.

I'm not sure what information I need to provide, but you can find my kernel
config at http://willden.org/~shawn/config-2.6.8.  It's built from the Debian
2.6.8 tree, with a custom config.  The CPU is a 1.6GHz Pentium M in a
"Centrino" chipset.

What should I be looking at to track this down?  I'm building a 2.6.9 kernel
but I haven't found any reports of similar problems corrected by it, so I
don't expect that to make a difference.

Thanks,

 Shawn.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQFBiA+J6d8WxFy/CWcRAjU4AKCnqPFgCTM618fcto+k89MrM3eptwCdHJ1I
oXDUSfaXXcs8FR+rl4Hpzco=
=TPBi
-----END PGP SIGNATURE-----

Re: 2.6.8 Thinkpad T40, clock running too fast

[john stultz]

On Tue, 2004-11-02 at 14:51, Shawn Willden wrote:
> I'm having a problem with the clock on my Thinkpad running too fast.  My clock
> is gaining about 4 seconds every five minutes of operation.  I'm actually not
> sure when this started because NTP has been fairly successful at keeping my
> clock under control, so I didn't really notice it until I spent some time
> operating disconnected... and then it became very obvious very quickly.

Does this go away if you disable cpufreq in your kernel config?

Also, looking at /proc/interrupts, does it look like you're getting more
then ~1000 interrupts per second?

thanks
-john

Re: 2.6.8 Thinkpad T40, clock running too fast

[Shawn Willden]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Thanks for the quick response, John.

On Tuesday 02 November 2004 04:06 pm, john stultz wrote:
> Does this go away if you disable cpufreq in your kernel config?

I'll try that next.

> Also, looking at /proc/interrupts, does it look like you're getting more
> then ~1000 interrupts per second?

I don't think so.  I'm not sure how to tell.  Running the following:

prev=0
while true; do
    cur=`cat /proc/interrupts| grep timer|cut -d' ' -f 6`
    (( diff = $cur - $prev ))
    echo $diff; prev=$cur
    sleep 1
done 

gives interrupt count differences that are between 1003 and 1222 per (rough) 
second.  The mean is 1016 with a std deviation of 16.  Running the same thing 
on another machine -- one without clock problems -- yields similar values.

Is there a better way to measure this?

Thanks

 Shawn
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQFBiCBf6d8WxFy/CWcRAlkFAKCIVLcrXdMENi41SZ4jwSNx5Ukg2ACgjRPW
Ig1kUi1EwRw13ba+MVEVrFk=
=/KWj
-----END PGP SIGNATURE-----

Re: 2.6.8 Thinkpad T40, clock running too fast

[john stultz]

On Tue, 2004-11-02 at 16:03, Shawn Willden wrote:
> On Tuesday 02 November 2004 04:06 pm, john stultz wrote:
> > Does this go away if you disable cpufreq in your kernel config?
> 
> I'll try that next.

Thanks, do let me know if it helps or not.

> > Also, looking at /proc/interrupts, does it look like you're getting more
> > then ~1000 interrupts per second?
> 
> I don't think so.  I'm not sure how to tell.  Running the following:
> 
> prev=0
> while true; do
>     cur=`cat /proc/interrupts| grep timer|cut -d' ' -f 6`
>     (( diff = $cur - $prev ))
>     echo $diff; prev=$cur
>     sleep 1
> done 
> 
> gives interrupt count differences that are between 1003 and 1222 per (rough) 
> second.  The mean is 1016 with a std deviation of 16.  Running the same thing 
> on another machine -- one without clock problems -- yields similar values.
> 
> Is there a better way to measure this?

Not really, the problem is that the sleep call returns after so many
timer ticks, so even if the wrong amount of time has passed, you'll see
the same number of interrupts. It would be best if you checked the time
on your watch, waited 5 minutes and checked again, or better, did
something similar w/ ntpdate.  I just wanted to eyeball it to make sure
you weren't running away w/ way too many timer interrupts. 

thanks
-john

PS: If you wouldn't mind, CC me next time.

Re: 2.6.8 Thinkpad T40, clock running too fast

[Shawn Willden]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Tuesday 02 November 2004 05:27 pm, john stultz wrote:
> Not really, the problem is that the sleep call returns after so many
> timer ticks, so even if the wrong amount of time has passed, you'll see
> the same number of interrupts.

Ah, right.  Nothing like measuring a ruler with itself.

> It would be best if you checked the time 
> on your watch, waited 5 minutes and checked again, or better, did
> something similar w/ ntpdate.  I just wanted to eyeball it to make sure
> you weren't running away w/ way too many timer interrupts.

I'll do that

> PS: If you wouldn't mind, CC me next time.

I am.  BTW, in case you didn't see my other e-mail, I'm now running with 2.6.9 
and cpufreq off.  No significant change.

 Shawn.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQFBiC0o6d8WxFy/CWcRAv2RAJ9ZhrXkIAoiEpZ9PrFK+jAQLMKyhQCfXkai
YhtDAgHFdjIR+WpV3XdG5Ss=
=kO2+
-----END PGP SIGNATURE-----

Re: 2.6.8 Thinkpad T40, clock running too fast

[Shawn Willden]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Tuesday 02 November 2004 04:06 pm, john stultz wrote:
> Does this go away if you disable cpufreq in your kernel config?

Nope.

My 2.6.9 config is at http://willden.org/~shawn/config-2.6.9

cpufreq is turned off.  APIC and local APIC are both turned off on the command 
line (noapic and nolapic).

Any ideas?

Thanks,

 Shawn.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD4DBQFBiCij6d8WxFy/CWcRAoBQAJUWR1c2HyOHHbYq+iM8FRh/n8S3AJwKhdOz
35b/Qmv3LxIDboFfXzoH2g==
=5D2Y
-----END PGP SIGNATURE-----

Re: 2.6.8 Thinkpad T40, clock running too fast

[Stefan Seyfried]

On Tue, Nov 02, 2004 at 05:38:53PM -0700, Shawn Willden wrote:
 
> Any ideas?

remove /etc/adjtime and reboot.
-- 
Stefan Seyfried

"Any ideas, John?"
"Well, surrounding them's out."

2.6.8 Thinkpad T40, clock running too fast

[Shawn Willden]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I'm having a problem with the clock on my Thinkpad running too fast.  My clock 
is gaining about 4 seconds every five minutes of operation.  I'm actually not 
sure when this started because NTP has been fairly successful at keeping my 
clock under control, so I didn't really notice it until I spent some time 
operating disconnected... and then it became very obvious very quickly.

On October 15th, I "fixed" the problem by booting with "noapic" on the command 
line.  Or I thought I fixed it anyway.  I had to reboot yesterday for an 
unrelated issue (a "stuck" smb share that I couldn't figure out how to 
unmount) and the racing clock is back.  I've booted with "noapic", "nolapic", 
"noapic nolapic" and no options at all, and the clock doesn't keep the time 
with any of them.  I'm not sure if it's always too fast.  I seem to remember 
that last month it was running too slow.  I'll check with the various apic 
settings and see if it's consistent.

I'm not sure what information I need to provide, but you can find my kernel 
config at http://willden.org/~shawn/config-2.6.8.  It's built from the Debian 
2.6.8 tree, with a custom config.  The CPU is a 1.6GHz Pentium M in a 
"Centrino" chipset.

What should I be looking at to track this down?  I'm building a 2.6.9 kernel 
but I haven't found any reports of similar problems corrected by it, so I 
don't expect that to make a difference.

Thanks,

 Shawn.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQFBiAep6d8WxFy/CWcRAoXVAKCeEwysrrF/+lcFuRttAx5JDs8RYQCgjf1q
cdK/581EqaX+2oGXlkOLpoE=
=sNMu
-----END PGP SIGNATURE-----