On Wed, Apr 12, 2006 at 11:15:49PM -0500, Linda Xie wrote: > On Wed, Apr 12, 2006 at 05:50:49PM -0500, Linda Xie wrote: > > I am running lhcs_regression-1.1 on my Power 5 partition. The partition > is > > booted with 2 cpus: cpu0 and cpu1 > > (cpu0/online =1 and cpu1/online =1): > > > > When CPU_TO_TEST is set to 1, I get the following results: > > > -------------------------------------------------------------------------------------- > Do you mean set to 0 here? > > Linda> Yes, you are right. > > > beta:/home/HOTPLUG-CPU-OSDL/lhcs_regression-1.1 # grep PASS output3 > > hotplug01: PASS: Loops left 1 > > hotplug04 PASS: Could not shutdown cpu0 > > hotplug06.top PASS: PID 13291 still running. > > hotplug06.sar PASS: CPU was found after turned on. > > > > beta:/home/HOTPLUG-CPU-OSDL/lhcs_regression-1.1 # grep FAIL output3 > > hotplug02 FAIL: process did not change from CPU 0 > > hotplug03 FAIL: CPU0 cannot be offlined > > > > When CPU_TO_TEST is set to 1, I get the following results: > > > -------------------------------------------------------------------------------------- > > > > > beta:/home/HOTPLUG-CPU-OSDL/lhcs_regression-1.1 # grep PASS output2 > > hotplug01: PASS: Loops left 1 > > hotplug02 PASS - turned off CPU 1, process migrated to CPU 0 > > hotplug03 PASS: do_spin_loop found on CPU1 > > hotplug04 PASS: Could not shutdown cpu0 > > hotplug06.top PASS: PID 12170 still running. > > > > beta:/home/HOTPLUG-CPU-OSDL/lhcs_regression-1.1 # grep FAIL output2 > > hotplug06.sar FAIL: CPU1 Not Found on SAR! > > > > So it seems that CPU_TO_TEST in runtests.sh should be set to 1 (not 0), > Am > > I right? > > Some architectures do not allow CPU0 to be offlined, in which case the > test should return results such as what you've seen. > > Linda> So CPU_TO_TEST in runtests.sh should be set to 1 for Power > architecture. Yes > > And why hotplug06.sar passed when CPU_TO_TESTis set to 0 and failed when > > CPU_TO_TESTis set to 1? > > Hmm, this does seem odd. Perhaps the sar output is different on your > platform. Try this: > > 1. Offline cpu1 > 2. sar -P ALL 1 0 > /tmp/log & > 3. cat /tmp/log | grep -i nan > > Linda> After offlining cpu1, sar -P ALL 1 0 returned "Not an SMP machine... > " > But after I did "echo 1 > cpu2/online, sar -P ALL 1 0 returned the > following: > 20:32:08 CPU %user %nice %system %iowait %idle > 20:32:09 all 0.00 0.00 0.00 0.00 100.00 > 20:32:09 0 0.00 0.00 0.00 0.00 100.00 > 20:32:09 1 nan nan nan nan nan > 20:32:09 2 0.00 0.00 0.00 0.00 100.00 > > So it seems that 'sar -P ALL x y' won't work unless there are at > least two cpus online. Hmm, that's interesting; when you offline one of your cpu's, 'sar -P ALL' thinks you're not SMP and is returning an error message. That's probably why you're getting strange results on this test. I wonder if that is correct behavior or not? > sar should be outputting 'nan' for the offlined cpu. If not, then > there is an issue with sar. > > Bryce > > > > > Bryce Harrington > > <bryce@xxxxxxxx> > > > To > > 02/17/2006 04:25 Joel Schopp > > PM <jschopp@xxxxxxxxxxxxxx> > > > cc > > Linda Xie/Austin/IBM@IBMUS, > Raymond > > Harney/Rochester/IBM@IBMUS > > > Subject > > Re: hotplug testing > > > > > > > > > > > > > > > > > > > > > > On Fri, Feb 17, 2006 at 04:20:43PM -0600, Joel Schopp wrote: > > > Bryce, > > > > > > Linda is moving into the role of testing liaison in the immediate > future > > > here at IBM. I'm not clear what exactly her new role entails, but I'm > > > confident that she will be able to share some of our tests with you and > > > can aid in integrating some of your tests into our test environment. > > > > Hi Linda, > > > > The hotplug cpu test that we created is available here: > > > > http://developer.osdl.org/dev/hotplug/tests/ > > > > Mary has been experimenting with MemToy, and possibly we make create a > > test around that at some point. > > > > > Linda, > > > > > > Bryce works at OSDL and has some hotplug tests that it would be useful > > > for us to integrate into ABAT. He also does a lot of automated testing > > > and might have use for some of our tests in his environment. > > > > Yes, I'd love to hear about what tests you've run. I'm especially > > interested in tests that'll validate the memory hotplug features. > > > > Thanks, > > Bryce > > > >
