Description of problem:
numactl does not output nodebind information for all cores in a dual core system.
Version-Release number of selected component (if applicable):
$ for i in `seq 0 7`; do taskset -c $i numactl --show | grep nodebind; done
On this system (dl585), core id's > 0 are not reported.
Also reported to affect other AMD dual core systems.
Problem still exists in numactl-0.9.8-1.33.
Reported as fixed upstream in 0.9.10.
Created attachment 136791 [details]
I think you were testing with a too-early release. There should be a nodebind
patch in numactl-0.9.8-1.35 (the changelog release numbers appear to be a bit off).
Re-tested with 0.9.8-1.35 - same problem.
I've confirmed now that this was fixed upstream between 0.9.9 and 0.9.10 on our
At least that gives a smaller set of changes to look at than from the RHEL4 / 5
packages to 0.9.10.
Created attachment 136862 [details]
correct numa_parse_bitmap for dual core systems
I've narrowed the problem down to a one-line change in numa_parse_bitmap - we
weren't handling the case when the first while loop terminates on a ','
I've tested this myself on the dl585 - building an RPM now based on the RHEL5
package with this change.
Proposed patch taken directly from 0.9.10.
please set the RHEL5 version back to waiting on tech please, and I'll fix this
in both locations
Created attachment 136867 [details]
earlier patch had borked path
fixed in release -1.37. Thanks!
Kent, see comment #7. Backported in -1.37
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.