On Tue, May 16, 2006 at 01:54:35AM -0500, Protasevich, Natalie wrote:
> > Hi Natalie!
> >
> > > I gave a quick run the suite on the 64x and 128x machines
> > (x86_64) and
> > > it looked good, all passed but the last test (hotplug07.sh) which
> > > errored out due to syntax problems.
> >
> > Hmm, what were the syntax problems? Case 7 had been disabled
> > in previous releases due to a path problem, so it's possible
> > it has additional errors.
>
> Is it possible that the archive doesn't have the latest?
> Error messages:
>
> functional/hotplug07.sh: line 31: [: argument expected
> get_affinity_mask 21437
> functional/hotplug07.sh: line 52: [: cpu_is_online: unary operator
> expected
These two are most likely due to ${CPU_TO_TEST} being undefined. Odd.
> Starting loop '1'
> functional/hotplug07.sh: line 69: offline_cpu{1}: command not found
> Offlining cpu1: Return Code = 127
Okay, this one is syntax error (should be parens instead of brackets).
Attached is a patch to fix the above error, and to print out the value
of $CPU_TO_TEST. Can you apply it to hotplug07.sh and re-run it and
send me the output?
> I tried fixing the include files etc. but still got errors:
>
> Starting loop '1'
> Offlining cpu1: Return Code = 1
> FAIL - PID 24574 no longer running
> PASS - turned off CPU 1, process migrated to CPU
> Onlining cpu1: Return Code = 1
Okay, this one seems to be a legitimate issue. Perhaps your system is
doing something unusual here with the process's PID when migrating it?
You could try manually setting the affinity for a process to a cpu, then
offline that cpu, and see where the process moves to. It should move to
one of the other cpu's.
Or possibly the 'ps' on your system works differently than ours. We're
running it like this: ps --pid=${KCOMPILE_LOOP_PID} -o psr --no-headers
> > Another possible solution would be an option to allow only
> > displaying the top N "most active" CPUs.
> >
> > Are you able to run top at all on this machine, or does it
> > simply terminate on startup?
>
> No it worked fine, but appeared not to update the top correctly while
> number of processors was increasing. I just tried it again to double
> check on that, and turned out that top errors out while offlining
> processors (top: failed /proc/stat read), and still works while onlining
> but shows their number decreasing instead of growing :) Maybe I don't
> have the latest top (although running on SLES10?).
It'd be worth checking against the latest version of top in any case;
then you'd know whether to report this issue to the top maintainers, or
just to SuSE.
Bryce
