Bug 248368 - Enabling Hyperthreading on fc6 kernel 2.6.20-1.2952.fc6 or 2.6.20-1.2962.fc6 on IBM eseries x345 Xeon hardware makes kernel clock loose time like a bandit
Summary: Enabling Hyperthreading on fc6 kernel 2.6.20-1.2952.fc6 or 2.6.20-1.2962.fc6 ...
Keywords:
Status: CLOSED INSUFFICIENT_DATA
Alias: None
Product: Fedora
Classification: Fedora
Component: kernel
Version: 6
Hardware: i686
OS: Linux
low
high
Target Milestone: ---
Assignee: Kernel Maintainer List
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2007-07-16 14:34 UTC by James Robinson
Modified: 2007-11-30 22:12 UTC (History)
0 users

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2007-07-16 15:15:18 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)
time drift measurements with ntpd off, varying HT enabled versus disabled. (13.46 KB, text/plain)
2007-07-16 14:34 UTC, James Robinson
no flags Details

Description James Robinson 2007-07-16 14:34:49 UTC
Description of problem:

Turing on HT on our IBM eServer xSeries 345: Type 8670 Model 31X machines
makes the kernel loose about 1.5 seconds per each 10 seconds. Turing off HT
in the BIOS makes everything happy again. Seems to have been introduced
around kernel 2.6.20-1.2952.fc6, continues to exist in 2.6.20-1.2962.fc6.



Version-Release number of selected component (if applicable):
2.6.20-1.2952.fc6, 2.6.20-1.2962.fc6

How reproducible:
Definitely reproducable on our two HT-capable boxes -- both IBM eseries x345
machines. Our other non-HT intel / PPC boxes running same kernels don't
exhibit time issues.

Steps to Reproduce:
1. Take BM eseries x345 (or possibly other era older xeon HT machine)
2. Enable HT
3. Boot
4. Watch clock loose time, regardless of ntpd.

Then disable HT, and all's happy.

Actual results:
ntpd can't keep clock adjusted, gettimeofday gets worse and worse off.


Expected results:
kernel keeps proper time just like it did before with HT enabled, or like it
does with HT disabled.

Additional info:
See attachment for time drift measurements with ntpd off, varying HT enabled
versus disabled.

Comment 1 James Robinson 2007-07-16 14:34:49 UTC
Created attachment 159325 [details]
time drift measurements with ntpd off, varying HT enabled versus disabled.

Comment 2 James Robinson 2007-07-16 14:44:10 UTC
Oh yeah -- these machines are SMP -- 2 2.4Ghz Xeons. So with HT enabled, kernel
shows 'em as having 4 CPUs.

Here's what /proc/cpuinfo has to say about 'em with HT off, if that helps in
finding a machine to test against:

[root@build-db02 ~]# cat /proc/cpuinfo 
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Xeon(TM) CPU 2.40GHz
stepping        : 7
cpu MHz         : 2602.628
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat
pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
bogomips        : 7186.29
clflush size    : 64

processor       : 1
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Xeon(TM) CPU 2.40GHz
stepping        : 7
cpu MHz         : 2602.628
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat
pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
bogomips        : 7189.95
clflush size    : 64



Comment 3 James Robinson 2007-07-16 15:14:49 UTC
Grr. Spoke too soon. Disabling HT doesn't help at all -- time still drifts
uncontrollably. Will close bug out and gather better data and then resubmit.


Note You need to log in before you can comment on or make changes to this bug.