** Description changed:

  [ Impact ]
  
  numactl 2.0.18-1build1 has a bug, wherein trying to set a node for
  memory using membind emits an "Invalid argument" error.
  
  Ubuntu: 24.04.3 LTS (Noble Numbat)
  Kernel: 6.8.0-90.91
  libnuma1: 2.0.18-1build1
  numactl: 2.0.18-1build1
  
  This bugfix should be backported since numactl crashes if
  MPOL_PREFERRED_MANY isn't supported.
  
  The fix is to not clobber errno, if the set_mempolicy syscall fails:
  
https://github.com/numactl/numactl/commit/6c14bd59d438ebb5ef828e393e8563ba18f59cb2
  
  More details of this issue can be found here:
  https://github.com/numactl/numactl/issues/223
  
  [ Test Plan ]
  
  - Launch a Focal VM (kernel old enough to not support MPOL_PREFERRED_MANY)
  - Inside of the VM, launch a Noble container
  - install numactl from the archive
  - run $ numactl --membind=0 echo success and see it fail with the error above
  - install numactl from -proposed
  - run $ numactl --membind=0 echo success and see 'success'
  
  [ Where problems could occur ]
  
-  * Think about what the upload changes in the software. Imagine the
-    change is wrong or breaks something else: how would this show up?
+ This is a backport of a small patch from upstream, claiming that it's
+ the proper fix for the issue, and pre-tested before upload. In the
+ unlikely case upstream is wrong, or we misidentified the issue, the fix
+ may still show corner cases we can't predict. It's again unlikely
+ because it's a change in the specific function that checks for the
+ kernel capability.
  
-  * It is assumed that any SRU candidate patch is well-tested before
-    upload and has a low overall risk of regression, but it's important
-    to make the effort to think about what ''could'' happen in the event
-    of a regression.
- 
-  * This must never be "None" or "Low", or entirely an argument as to why
-    your upload is low risk.
- 
-  * This both shows the SRU team that the risks have been considered,
-    and provides guidance to testers in regression-testing the SRU.
+ This does not affect any kernel after 5.15, where the check will cause a
+ positive response, as the support for MPOL_PREFERRED_MANY was added
+ there. All currently supported Ubuntu releases (not counting ESM), and
+ this inclused Noble (target for this SRU), already have newer kernel
+ versions. This means the suggested test setup above, old-kernel
+ underlying system with Noble container on top, is where this change will
+ show a difference. In case of a regression in this function, the amount
+ of users affected will be low, and the feature is not working anyway.
  
  [ Original Report ]
  
  numactl 2.0.19-1 fixes the bug, and has already been built for questing. I 
have tested it on noble and it works without issues.
  Can we bump up the version of `numactl` on noble too?

** Merge proposal linked:
   
https://code.launchpad.net/~rr/ubuntu/+source/numactl/+git/numactl/+merge/498315

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2136104

Title:
  numactl gives "Invalid argument" error when trying to set a node for
  memory using membind

To manage notifications about this bug go to:
https://bugs.launchpad.net/appmajix/+bug/2136104/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to