Bug 113323 - System Clock runs at 2x speed
Summary: System Clock runs at 2x speed
Status: RESOLVED FIXED
Alias: None
Product: SUSE LINUX 10.0
Classification: openSUSE
Component: Kernel (show other bugs)
Version: Beta 3
Hardware: x86-64 SUSE Other
: P5 - None : Critical
Target Milestone: ---
Assignee: Andreas Kleen
QA Contact: E-mail List
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2005-08-26 14:06 UTC by Chip Daiger
Modified: 2005-12-30 05:57 UTC (History)
4 users (show)

See Also:
Found By: Beta-Customer
Services Priority:
Business Priority:
Blocker: ---
Marketing QA Status: ---
IT Deployment: ---


Attachments
hwinfo (237.72 KB, text/plain)
2005-08-27 00:24 UTC, Chip Daiger
Details
dmesg info (23.93 KB, text/plain)
2005-08-27 00:26 UTC, Chip Daiger
Details
boot.msg with nopmtimer passed while booting (22.00 KB, text/plain)
2005-08-29 21:39 UTC, Chip Daiger
Details
boot.msg with "notsc" passed (21.97 KB, text/plain)
2005-08-30 21:25 UTC, Chip Daiger
Details
boot.msg with 8/30 kotd (21.31 KB, application/octet-stream)
2005-08-31 22:44 UTC, Chip Daiger
Details
boot.msg with 8/30 kotd (21.31 KB, text/plain)
2005-08-31 22:48 UTC, Chip Daiger
Details
lspci output (1.42 KB, text/plain)
2005-09-01 18:49 UTC, Chip Daiger
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Chip Daiger 2005-08-26 14:06:12 UTC
System is an Emachine T6212 (MSI mobo, ATI Radeon Xpress 200 chipset).
Comment 1 Olaf Kirch 2005-08-26 14:17:08 UTC
Can you please provide hwinfo output and a copy of the kernel boot 
messages (dmeg output)? 
 
Thanks! 
Comment 2 Chip Daiger 2005-08-27 00:24:03 UTC
Created attachment 47848 [details]
hwinfo

per request
Comment 3 Chip Daiger 2005-08-27 00:26:06 UTC
Created attachment 47849 [details]
dmesg info
Comment 4 Olaf Kirch 2005-08-29 09:00:08 UTC
The kernel sets up the ACPI PM timer: 
 
time.c: Using 3.579545 MHz PM timer. 
time.c: Detected 1989.851 MHz processor. 
time.c: Using PIT/TSC based timekeeping. 
 
Maybe this machine has a broken PM timer? 
 
Please try booting with nopmtimer 
Comment 5 Chip Daiger 2005-08-29 21:38:12 UTC
Booted with nopmtimer...clock still running exactly 2X...will attach boot.msg
Comment 6 Chip Daiger 2005-08-29 21:39:22 UTC
Created attachment 48071 [details]
boot.msg with nopmtimer passed while booting
Comment 7 Olaf Kirch 2005-08-30 07:34:35 UTC
Alright, next try: what happens if you boot with "notsc"? 
 
Andi, do you have any idea what might be wrong here? 
Comment 8 Chip Daiger 2005-08-30 21:23:27 UTC
Clock still runs at exactly 2X with "notsc"
Comment 9 Chip Daiger 2005-08-30 21:25:50 UTC
Created attachment 48238 [details]
boot.msg with "notsc" passed
Comment 10 Olaf Kirch 2005-08-30 21:36:26 UTC
Alright, I'm running out of ideas.

Andi, can you help with this, please?
Comment 11 Andreas Kleen 2005-08-30 21:44:49 UTC
Does the latest kernel from
ftp://ftp.suse.com/pub/projects/kernel/kotd/x86_64/HEAD/* 
still show the problem? It should have a workaround for this ATI
chipset issue.

Comment 12 Chip Daiger 2005-08-31 22:42:54 UTC
Installed the kotd from 8/30...problem still persists
Comment 13 Chip Daiger 2005-08-31 22:44:28 UTC
Created attachment 48390 [details]
boot.msg with 8/30 kotd

let me know if you want/need more info
Comment 14 Chip Daiger 2005-08-31 22:48:40 UTC
Created attachment 48392 [details]
boot.msg with 8/30 kotd

Sorry...previous was not text
Comment 15 Andreas Kleen 2005-08-31 22:57:35 UTC
Hmm, the check didn't trigger. Can you add lspci ?

The clock is ok when you boot with "no_timer_check", right?

Comment 16 Chip Daiger 2005-09-01 18:49:27 UTC
Created attachment 48515 [details]
lspci output
Comment 17 Chip Daiger 2005-09-01 18:50:09 UTC
clock runs fine with "no_timer_check"
Comment 18 Beau Steward 2005-09-02 16:44:46 UTC
Using "noapictimer" will also bring the clock to normal speed. I've seen this  
problem in other distributions, including gentoo and ubuntu. My research has 
shown that this is particularly visible on later generation amd cpus, but I'm 
not sure if it's only the amd64 or if it also includes sempron. 
 
I'm currently using "noapictimer" to keep the timing correct. 
Comment 19 Bart Symons 2005-09-04 18:59:34 UTC
The clock running twice as fast as normal is a known bug with the Linux kernel
v2.6. See http://bugzilla.kernel.org/show_bug.cgi?id=3927.

It seem to be typical with AMD64 + ATI Radeon chipsets. 

I also have this problem with SuSE 9.3 on an HP Pavillion model 5050.be (this
computer has an MSI motherboard, type MS-7093 aka RS480M2 and RX480M2. See
http://www.msi.com.tw/program/products/mainboard/mbd/pro_mbd_detail.php?UID=639

This problem has also been reported in other places:
http://lists.suse.com/archive/suse-amd64/2005-May/0050.html
http://www.linuxquestions.org/questions/history/349343
http://nixdoc.net/files/forum/ntopic32168.html

There is a patch available (see links above) but I personally don't like to
recompile the kernel.


I really hope SuSE can fix this for 9.3 and 10.0. A clock at double speed makes
the entire OS useless.
Comment 20 Bart Symons 2005-09-06 16:13:25 UTC
I have taken the liberty to increase the severity and priority level of this bug
report. 

1) The bug is very real and judging by Google result is very common + a solution
is available
-> increase in priority is necessary

2) Nobody can use an OS for any real work when the clock is going at double speed.
-> severity should be set to "Blocker"

What us is an OS that runs but cannot be used for any real work?
Comment 21 Olaf Kirch 2005-09-07 10:02:47 UTC
I do not think this qualifies as a blocker, since using noapictimer   
or no_timer_check seems to prevent this. IOW there's a known workaround.   
   
Moreover the patch posted to the kernel.org bugzilla seems to introduce   
new problems, and is not ready for inclusion in SL 10.0. 
 
Adding aj to the cc list in case he would like to document this in the 
release notes. 
Comment 22 Andreas Kleen 2005-09-07 10:08:36 UTC
The kernel.org patch is not usable because it breaks 
other stuff.

Need to figure out why the auto detection patch didn't work.

I disagree on the blockerness of this too.
Comment 23 Bart Symons 2005-09-07 10:35:21 UTC
AFAIK it should be sufficient to specify either naapictimer or no_timer_check as
kernel parameterin GRUB, right?

I'm afraid that neither of the two workarounds (noapictimer and no_timer_check)
is guaranteed to work. 

As I have experienced myself, and found others with the same negative
experience, these options may cause the computer to lock up during the boot process.

Maybe it'll work on some computers, but there is evidence that it is not a
surefire workaround.

In the end, many users will be left with the choice:
- OS boots but clock races at double speed
- OS doesn't boot



Comment 24 Andreas Kleen 2005-09-07 10:54:19 UTC
noapic should always work, doesn't it?

Perhaps we just force that on ATI boards for the release.
Comment 25 Andreas Jaeger 2005-09-07 11:14:14 UTC
Once you agreed on the solution, please open a new bug report with component
"Release Notes" and tell us what to write.
Comment 26 Bart Symons 2005-09-07 11:51:39 UTC
(In reply to comment #24)
> noapic should always work, doesn't it?
> 

noapic also causes the system to hang quite early in the boot process.

MSI motherboard, type MS-7093 aka RS480M2 and RX480M2
SuSE 9.3 x86-64
SuSE kernel 2.6.11.4-21.9-default (20050117)

Comment 27 Olaf Kirch 2005-09-07 11:57:31 UTC
Bart, this bug is about 10.0, not 9.3. Can you test with the current beta 
kernel? 
Comment 28 Andreas Kleen 2005-09-07 12:05:32 UTC
The problem is still there - lots of reports for mainline.

There is one patch that will probably help, but it breaks all other
machines.
Comment 29 Mikhail Grushinskiy 2005-09-08 03:06:36 UTC
I have this problem with HP Pavilion a1130n AMD 64 bit SuSe 9.3.
I'll try options from this bug thread. noapic definitely causes hang during boot.
Comment 30 Mikhail Grushinskiy 2005-09-08 03:27:05 UTC
on HP Pavilion a1130n AMD 64 bit SuSe 9.3 (ATI chipset)
no_timer_check, notsc - do not fix the problem
noapic, noapictimer - cause system hang at boot time
Comment 31 Andreas Kleen 2005-09-09 06:57:15 UTC
Increasing severity to critical again to bring it back onto the radar.
Comment 32 Andreas Kleen 2005-09-12 13:06:21 UTC
I added a new workaround which will hopefully fix the problem. 
It didn't make RC2, but you can test the next kotd with this change

- patches.arch/x86_64-no-timer-check: (#113323) Update ATI timer bug workaround.
Comment 33 Andreas Kleen 2005-09-13 12:04:21 UTC
Anyone was able to test it? 

Comment 34 Bart Symons 2005-09-13 12:27:07 UTC
(In reply to comment #33)
> Anyone was able to test it? 
> 
> 
Andreas,

I'm not able to test it because I can't afford to disrupt my production system
which is currently running SuSE 9.3, because I'm too busy and because I will be
travelling abroad.

Also, the following is complete voodoo to me:
"kotd" and this reference "- patches.arch/x86_64-no-timer-check: (#113323)
Update ATI timer bug workaround".

What are we supposed to do with this? Please explain.



Comment 35 Andreas Kleen 2005-09-13 12:34:00 UTC
kotd is kernel of the day
ftp://ftp.suse.com/pub/suse/projects/kernel/kotd/x86_64/HEAD/ 
(or faster mirror e.g. ftp.gwdg.de) 

Ok anyone else? 

Download the kernel rpm from there and rpm -Uvh it and reboot
and test if the problem is gone.
Comment 36 Bart Symons 2005-09-13 13:44:26 UTC
Can this kernel also be used with SuSE 9.3? If so, I'm willing to give it a try.
Comment 37 Andreas Kleen 2005-09-13 13:57:42 UTC
Hmm - why do we have so many people with 9.3 on this 10.0 bug? Please take
a look at the version number on top of the page.

It'll probably work if you update udev too, but no guarantees.
Comment 38 Chip Daiger 2005-09-13 22:34:34 UTC
Andreas:

I installed the kotd on RC1...took out the "no_timer_check" parameter from
grub... back to 2x clock speed.  The only question I have; did I need to install
a patch or did the kotd already contain it?

Let me know what further info you need...for now, back to "no_timer_check"
Comment 39 Beau Steward 2005-09-14 06:53:02 UTC
This doesn't seem to be in the mirror listed or any of the mirrors I've checked.
Is there a mirror confirmed to have the projects directory? I have a system with
this problem that I don't mind testing on. One thing to note is I'm using
"noapictimer" instead of "no_timer_check".

Just a quick note to those using SuSE 9.3 and having this problem...you may want
to consider submitting a separate bug report for the 9.3 distro or upgrading to
10 (not recommended for production systems).
Comment 40 Chip Daiger 2005-09-14 18:26:39 UTC
Changing back to ASSIGNED per earlier comment #38
Comment 41 Beau Steward 2005-09-17 16:24:36 UTC
I've been doing some testing for another bug and have some information
pertaining to this one. As I mentioned, I use the noapictimer flag, which works.
In the other bug, there is a suspend to disk problem, and this also appears in
2.6.13.1 vanilla sources. In previous distributions, I've not seen 2.6.13 used,
such as with gentoo, who is still using 2.6.12, and some of the problems I'm
having with suse 10 don't exist, so I installed vanilla 2.6.12. Suspend to disk
and resume works, but it seems when resuming, my noapictimer flag is ignored as
the timing seems to revert back to the 2x speed.
Comment 42 Chip Daiger 2005-09-24 16:35:34 UTC
Installed kotd kernel-default-2.6.13-20050921081308.x86_64 and took out
"no_timer_check"...again, back to 2x clock
Comment 43 Chip Daiger 2005-11-05 19:26:58 UTC
Fixed in 10.0 Final
Comment 44 Beau Steward 2005-12-30 05:57:38 UTC
I have this bug bookmarked so I'm gonna go ahead and update it with something I just found from HP concerning this problem. I found a BIOS update from November with this note:

"Fixes issue where system time runs twice as fast as it should when the notebook is running a Linux Operating System with Advanced Programmable Interrupt Controller (APIC) enabled."

I plan to flash my BIOS and see if there are any additional problems as a result of this fix.