Hyperthreading problem: only interrupts on CPU1, never applications?

William M. Shubert wms at igoweb.org
Fri Apr 9 04:02:51 UTC 2004

Hi. I just got a new HT system, so I installed
kernel-smp-2.4.22-1.2174.nptl onto my Fedora hard disk and plugged it
in. The dual CPUs are detected, and reading /proc/interrupts shows that
interrupts are indeed being balanced between the two, but application
code is only being run on CPU0, never on CPU1!!! I have tried running a
bunch of all-cpu, no-io processes, but no good...always CPU 0 gets 100%
load, but CPU1 sits there at 0% (unless there are a lot of interrupts,
then it's system time might creep up to 0.1%).

Some other strange things are happening (some of these will be
comparisons with my other HT systems, which run RHEL3, and properly use
both logical CPUs):
      * "top" on an unloaded system shows "idle" to be the sum of CPU0
        and CPU1, not the average. Since cpu1 is always 100% idle, that
        means that the "total" line is always in the 100%..200% area.
      * The "migrate/0" and "migrate/1" kernel threads do not appear in
        "ps aux". On the RHEL3 systems, these threads appear; does the
        FC1 kernel require these threads? If so, why wouldn't they be
      * "x86info" shows "Number of logical processors supported within
        the physical package: 0" for the FC1 system. On the RHEL3
        system, it properly shows "2" instead.
I'd like to try plugging the RHEL3 disks into the FC1 system to see if
it is a hardware issue, but they are at a remote site so that isn't
possible. Anybody know what this could be?

More info in case it helps:
$ uname -r
$ cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 2.80GHz
stepping        : 9
  ... etc ...
processor       : 1
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 2.80GHz
stepping        : 9
  ... etc ...

BIOS says HT is enabled. Those model/stepping parameters are exactly the
same as seen on the RHEL3 systems which work fine. Please let me know
what could be causing this, I'd like to get to actually use the HT for
more than IRQ balancing!
        Bill Shubert (wms at igoweb.org)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.fedoraproject.org/pipermail/users/attachments/20040408/06c853ad/attachment-0002.bin 

More information about the users mailing list