[Bug 2002429] Re: Patch needs to be removed ("debian/patches/gcc-ibmz-plt-revert.diff: Revert PLT changes from the gcc-11 branch")

[Dimitri John Ledkov]

On Fri, 19 May 2023 at 00:30, Chris Halse Rogers
<2002429 at bugs.launchpad.net> wrote:
>
> Hm, maybe I've been unclear.

You are trying to apply mechanical reasoning without looking at the
fix in question.

The combination of users you are alluding to who may experience a
regression is zero, in this instance:
- userspace code is unaffected
- kinetic+ requires 5.19+ kernels,
- 5.19+ kinetic kernels can only be built with gcc-12 due to security
features that require newer toolchain
- the only affected combination of users is jammy building v5.15 kernel

Separately, we ensure no regressions of default versions of toolchains
between release upgrades, and we frequently do not ship neither
security or fixes to non-default toolchains.

>
> What I mean is: while the launchpad buildds don't use gcc-11 for
> anything in kinetic onwards, launchpad buildds aren't the only users of
> a compiler and the Ubuntu kernel is not the only thing that users of a
> compiler might compile.
>
> gcc-11 is a package available in the archive, and users can use it¹,
> regardless of the particular rationale that *we* have for including it
> in the archive. If a jammy user is deliberately using gcc-11 I would
> expect that upgrading to kinetic would not regress any fixes we've made
> to the jammy package.
>
> I don't *think* we have a policy that non-default compilers are not
> subject to the usual SRU policy? If we do, maybe that should be better
> communicated :)

Nature of the bug, and the fix, guaratee no regression to any users
upon dist-upgrade, even if they choose to use gcc-11 by force.

>
> Again, I'm OK with deciding to tell users of gcc-11 that to get this fix
> they need to either stay on Jammy or upgrade all the way to Lunar. I
> just want to make sure that we've actively decided to tell users this is
> unsupported, rather than just verifying that it doesn't affect *us*.
>
> ¹: Maybe they're working around a gcc-12 miscompilation, or don't want
> to revalidate gcc-12 just now, or just like the number 11.
>

SRU fixes can be released in LTS release without being prepared or
released yet in the interim releases, especially since they are fixed
in the devel release already or even non-applicable. We do this all
the time.

For example python3.8 is default in main in some releases, but not
others. The one that is in main gets security fixes in the Ubuntu
Archive, the one not in universe doesn't receive neither SRUs nor
security fixes, and only receives Pro updates.

Similar for all compilers we ever shipped.

Note, your reasoning is flawed, and doesn't stand the test of time.
Demanding a fix for kinetic is pointless, given it is old release
which is about to go eol, and the package in question is unused is not
affected, and the regression is not present in the default compiler.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to gcc-11-cross in Ubuntu.
https://bugs.launchpad.net/bugs/2002429

Title:
  Patch needs to be removed ("debian/patches/gcc-ibmz-plt-revert.diff:
  Revert PLT changes from the gcc-11 branch")

Status in Ubuntu on IBM z Systems:
  New
Status in gcc-11 package in Ubuntu:
  Fix Released
Status in gcc-11-cross package in Ubuntu:
  New
Status in gcc-11-cross-ports package in Ubuntu:
  New
Status in gcc-11 source package in Jammy:
  Fix Committed
Status in gcc-11-cross source package in Jammy:
  Fix Committed
Status in gcc-11-cross-ports source package in Jammy:
  Fix Committed
Status in gcc-11 source package in Kinetic:
  New
Status in gcc-11 source package in Lunar:
  Fix Released
Status in gcc-11 package in Debian:
  Fix Released

Bug description:
  [ Impact ]

   * Bug #1954676 (https://bugs.launchpad.net/bugs/1954676) was opened
  because of kernel soft lockup issues with Jammy 5.15. A temporary
  patch to gcc-11 was identified and released under debian/patches/gcc-
  ibmz-plt-revert.diff, which reverted gcc commit 0990d93 ("IBM Z: Use
  @PLT symbols for local functions in 64-bit mode"). An eventual fix to
  the kernel lockup was released under the upstream Linux kernel, commit
  f3b7e73b2c66.

   * To this end, the debian/patches/gcc-ibmz-plt-revert.diff patch
  should be removed from gcc-11 as it presents issues with object file
  symbols. An attempt at this is in my PPA -
  https://launchpad.net/~john-cabaj/+archive/ubuntu/gcc-11). This
  allowed me to get past my issues with the object file symbols.

  [ Test Plan ]

   * use kpatch-build testcases to build and load a livepatch on s390x
  kernel built with gcc-11

   * it should be successful

  [ Where problems could occur ]

   * Whilst the kernel issues was worked-around by reverting compiler
  changes, the actual issue was in the kernel code; which has since been
  fixed. In ubuntu the kernel has been fixed since Ubuntu-5.15.0-20.20

  [ Other Info ]
   
   * A bit sad that this was not detected earlier, and compiler fixed up earlier.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu-z-systems/+bug/2002429/+subscriptions