Clock skips (?) with 2.6 and games

Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Hi,

I'm experiencing skips in games like q3demo and enemy territory on a
dual xeon p4. That means, if I'm walking around, about every 2-3 seconds
I'm skipping a bit of the way. It seems that the clock is running too
slow and the games are trying to catch up every x seconds with the
system time.
System is running 2.6.0-test9-mm1. This effect does not show with
2.4.23pre6aa3, though there are only two processors displayed. Is this
normal? Judging from the temperature sensors that is not just one
processor with its sibling but really the two physical processors. Is
there any way with 2.4 to show all 4 processors?
I've tried booting 2.6 with nosmp, but that results in most interrupts
not working anymore.
What can I try to get test9 working properly?

Thanks,

Jan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/pi3ULqMJRclVKIYRAkORAJ9foIw7SyrGuWzUn1FmkW+uAw3iUwCfVDFm
KfPZTA8XKGLqIc+53z7oQmE=
=0M+7
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Nick Piggin]

Jan Dittmer wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi,
>
> I'm experiencing skips in games like q3demo and enemy territory on a
> dual xeon p4. That means, if I'm walking around, about every 2-3 seconds
> I'm skipping a bit of the way. It seems that the clock is running too
> slow and the games are trying to catch up every x seconds with the
> system time. 


Please ensure that X is running at priority 0. Report back if you still
have the problem.

>
> System is running 2.6.0-test9-mm1. This effect does not show with
> 2.4.23pre6aa3, though there are only two processors displayed. Is this
> normal? Judging from the temperature sensors that is not just one
> processor with its sibling but really the two physical processors. Is
> there any way with 2.4 to show all 4 processors?
> I've tried booting 2.6 with nosmp, but that results in most interrupts
> not working anymore.
> What can I try to get test9 working properly?


nosmp has been broken for quite a while. If you want to try uniprocessor,
you'd have to compile a UP kernel.

You should get as good if not better interactivity with SMP enabled, 
however.

Re: Clock skips (?) with 2.6 and games

[Andrea Arcangeli]

On Mon, Nov 03, 2003 at 11:28:36AM +0100, Jan Dittmer wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hi,
> 
> I'm experiencing skips in games like q3demo and enemy territory on a
> dual xeon p4. That means, if I'm walking around, about every 2-3 seconds
> I'm skipping a bit of the way. It seems that the clock is running too
> slow and the games are trying to catch up every x seconds with the
> system time.
> System is running 2.6.0-test9-mm1. This effect does not show with
> 2.4.23pre6aa3, though there are only two processors displayed. Is this

btw, to make it even better for a desktop multimedia usage like yours,
you can also pass 'desktop' to the 2.4.23pre6aa3 kernel. I'd like to
hear how 'desktop' affects your usage in 2.4.23pre6aa3.

thanks.

Re: Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Andrea Arcangeli wrote:
| On Mon, Nov 03, 2003 at 11:28:36AM +0100, Jan Dittmer wrote:
|>
|>System is running 2.6.0-test9-mm1. This effect does not show with
|>2.4.23pre6aa3, though there are only two processors displayed. Is this
|
|
| btw, to make it even better for a desktop multimedia usage like yours,
| you can also pass 'desktop' to the 2.4.23pre6aa3 kernel. I'd like to
| hear how 'desktop' affects your usage in 2.4.23pre6aa3.

I'll give it a try. Is there a way in 2.4-aa to get the two additional
virtual processors displayed?

Thanks,
Jan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/pjpXLqMJRclVKIYRAinYAJsHswAa6yIHDbAiXpCsl4JbjGaFXwCfYfem
/WM79Hac8QaQ6jlhsR6dpXE=
=sC+Q
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Andrea Arcangeli]

On Mon, Nov 03, 2003 at 12:21:59PM +0100, Jan Dittmer wrote:
> I'll give it a try. Is there a way in 2.4-aa to get the two additional
> virtual processors displayed?

No idea why they're not displayed, they should. my HT 2-way xeon shows 4
cpus not 2 (with 2.4 too).

Re: Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Andrea Arcangeli wrote:
| On Mon, Nov 03, 2003 at 12:21:59PM +0100, Jan Dittmer wrote:
|
|>I'll give it a try. Is there a way in 2.4-aa to get the two additional
|>virtual processors displayed?
|
|
| No idea why they're not displayed, they should. my HT 2-way xeon shows 4
| cpus not 2 (with 2.4 too).

Strange, if I enable Highmem support and set CONFIG_NR_CPUS from 4 to 8,
4 penguins are showing up...

Jan

- --- config-2.4.23pre6aa3-4proc-nohm     2003-11-03 16:42:15.000000000 
+0100
+++ config-2.4.23pre6aa3-8proc-hm       2003-11-03 16:33:08.000000000 +0100
@@ -58,19 +58,20 @@
~ CONFIG_MICROCODE=m
~ CONFIG_X86_MSR=m
~ CONFIG_X86_CPUID=m
- -CONFIG_NOHIGHMEM=y
- -# CONFIG_HIGHMEM4G is not set
+# CONFIG_NOHIGHMEM is not set
+CONFIG_HIGHMEM4G=y
~ # CONFIG_HIGHMEM64G is not set
- -# CONFIG_HIGHMEM is not set
+CONFIG_HIGHMEM=y
~ CONFIG_FORCE_MAX_ZONEORDER=11
~ CONFIG_1GB=y
~ # CONFIG_2GB is not set
~ # CONFIG_3GB is not set
~ # CONFIG_05GB is not set
+CONFIG_HIGHIO=y
~ # CONFIG_MATH_EMULATION is not set
~ CONFIG_MTRR=y
~ CONFIG_SMP=y
- -CONFIG_NR_CPUS=4
+CONFIG_NR_CPUS=8
~ # CONFIG_X86_NUMA is not set
~ # CONFIG_X86_TSC_DISABLE is not set
~ CONFIG_X86_TSC=y
@@ -1133,6 +1134,7 @@
~ # CONFIG_KMSGDUMP is not set
~ # CONFIG_DEBUG_SPINLOCK is not set
~ # CONFIG_FRAME_POINTER is not set
+# CONFIG_HIGHMEM_EMULATION is not set
~ # CONFIG_X86_REMOTE_DEBUG is not set
~ # CONFIG_KERNEL_DEBUGGING is not set
~ CONFIG_LOG_BUF_SHIFT=17
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/pnfWLqMJRclVKIYRAtf2AJ9qojtJejZCHC62wVpuobM8V7tRVgCdFnka
A60HaWa0hQbG9vCz4+nVtA0=
=ySBx
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Nick Piggin wrote:
|
|
| Jan Dittmer wrote:
|
|> I'm experiencing skips in games like q3demo and enemy territory on a
|> dual xeon p4. That means, if I'm walking around, about every 2-3 seconds
|> I'm skipping a bit of the way. It seems that the clock is running too
|> slow and the games are trying to catch up every x seconds with the
|> system time.
|
|
|
| Please ensure that X is running at priority 0. Report back if you still
| have the problem.

It is. I've been running 2.6 on a dual p2 for quite some time, with good
results. It is also running with priority 0 under 2.4.

| nosmp has been broken for quite a while. If you want to try uniprocessor,
| you'd have to compile a UP kernel.
|
| You should get as good if not better interactivity with SMP enabled,
| however.
|

The problem does not seem interactivity, but some clocks which drift
apart. The framerate is constantly above 100 fps, there are no
background daemons/cron jobs running, remote top doesn't show any
unusual activity. For now, after several reboots and switching back and
forth between 2.6 and 2.4 it seems to be gone under 2.6. I'll see if it
comes back again. I guess a 'vmstat 1` snapshot would be good in this case?

Thanks,

Jan

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/pni5LqMJRclVKIYRAu+xAJ0a19pnR40JMDGu+DuKB37CzT3AMQCfZBnH
HHTyLuArV3JnyfNGPbMs40c=
=ARv1
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Stephan von Krawczynski]

On Mon, 03 Nov 2003 16:44:23 +0100
Jan Dittmer <j.dittmer@portrix.net> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Andrea Arcangeli wrote:
> | On Mon, Nov 03, 2003 at 12:21:59PM +0100, Jan Dittmer wrote:
> |
> |>I'll give it a try. Is there a way in 2.4-aa to get the two additional
> |>virtual processors displayed?
> |
> |
> | No idea why they're not displayed, they should. my HT 2-way xeon shows 4
> | cpus not 2 (with 2.4 too).
> 
> Strange, if I enable Highmem support and set CONFIG_NR_CPUS from 4 to 8,
> 4 penguins are showing up...
> 
> Jan

Have a look at /proc/cpuinfo. Possibly your processor numbers are not linear ...

Regards,
Stephan

Re: Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Stephan von Krawczynski wrote:
| On Mon, 03 Nov 2003 16:44:23 +0100
| Jan Dittmer <j.dittmer@portrix.net> wrote:
|
|>Strange, if I enable Highmem support and set CONFIG_NR_CPUS from 4 to 8,
|>4 penguins are showing up...
|>
|>Jan
|
|
| Have a look at /proc/cpuinfo. Possibly your processor numbers are not
linear ...
|

$ cat /proc/cpuinfo | grep processor
processor       : 0
processor       : 1
processor       : 2
processor       : 3

Quite linear it seems...

Jan

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/poXoLqMJRclVKIYRAl35AJ9dblZ5NdfxsKHC0PFnjrBaPRoQNwCdEeF+
HB1xqERdkZDSohEGv6wyhyU=
=NKVi
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Jan Dittmer]

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

Stephan von Krawczynski wrote:
|
|
| Have a look at /proc/cpuinfo. Possibly your processor numbers are not
linear ...
|

Ah, sorry, on boot they are numbered differently:

(from dmesg)

ACPI: LAPIC (acpi_id[0x01] lapic_id[0x00] enabled)
Processor #0 Pentium 4(tm) XEON(tm) APIC version 20
ACPI: LAPIC (acpi_id[0x02] lapic_id[0x06] enabled)
Processor #6 Pentium 4(tm) XEON(tm) APIC version 20
ACPI: LAPIC (acpi_id[0x03] lapic_id[0x01] enabled)
Processor #1 Pentium 4(tm) XEON(tm) APIC version 20
ACPI: LAPIC (acpi_id[0x04] lapic_id[0x07] enabled)
Processor #7 Pentium 4(tm) XEON(tm) APIC version 20

So thanks again and sorry for the noise,

Jan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/pomNLqMJRclVKIYRAhWyAJ9yfFPC/Liumt19sswdDK2PCaC7tgCdH9rT
VyQ0a2d0OX+OZZCMeMVje54=
=FMDH
-----END PGP SIGNATURE-----

Re: Clock skips (?) with 2.6 and games

[Zwane Mwaikambo]

On Mon, 3 Nov 2003, Nick Piggin wrote:

> nosmp has been broken for quite a while. If you want to try uniprocessor,
> you'd have to compile a UP kernel.

Hmm? It works here even with sparse APIC IDs.

Re: Clock skips (?) with 2.6 and games

[Zwane Mwaikambo]

On Mon, 3 Nov 2003, Jan Dittmer wrote:

> Strange, if I enable Highmem support and set CONFIG_NR_CPUS from 4 to 8,
> 4 penguins are showing up...

It should do it with the NR_CPUS change only, sounds like yet another APIC 
ID SMP bootstrap problem.

Re: Clock skips (?) with 2.6 and games

[Nick Piggin]

Zwane Mwaikambo wrote:

>On Mon, 3 Nov 2003, Nick Piggin wrote:
>
>
>>nosmp has been broken for quite a while. If you want to try uniprocessor,
>>you'd have to compile a UP kernel.
>>
>
>Hmm? It works here even with sparse APIC IDs.
>
>

Oh? Maybe its just me then. I lose my network card and IIRC a couple
of other interrupt sources with nosmp.

Re: Clock skips (?) with 2.6 and games

[Zwane Mwaikambo]

On Tue, 4 Nov 2003, Nick Piggin wrote:

> >Hmm? It works here even with sparse APIC IDs.
> 
> Oh? Maybe its just me then. I lose my network card and IIRC a couple
> of other interrupt sources with nosmp.

Oh, mind putting together a bugzilla entry? Then you can assign it to me.

Thanks

Re: Clock skips (?) with 2.6 and games

[Andrea Arcangeli]

On Mon, Nov 03, 2003 at 09:40:28PM -0500, Zwane Mwaikambo wrote:
> On Mon, 3 Nov 2003, Jan Dittmer wrote:
> 
> > Strange, if I enable Highmem support and set CONFIG_NR_CPUS from 4 to 8,
> > 4 penguins are showing up...
> 
> It should do it with the NR_CPUS change only, sounds like yet another APIC 
> ID SMP bootstrap problem.

yes, and now with NR_CPUS == 8 Jan can compare apples to apples. So I
would suggest you to repeat the interactivity test, first w/o desktop
then w/ desktop. My tree has a o1 scheduler, though quite different
from any other version (especially for HT machines).