[Bug 1951033] Re: 20.04 SRU

[Chris Halse Rogers]

Expanding on (5): we've had problems in the past with glibc and
graphical snaps on core due to the way the host's userspace drivers are
mapped into the snap environment, and so are loaded with the snap's
glibc. We should check that <a graphical snap> runs successfully with
the core20 built from focal-proposed, on mesa-supported hardware, an
NVIDIA system, (and maybe an AMD system with the proprietary stack
installed? I don't know how supported we claim that is). We *might* want
to test running <a graphical snap> using the focal-proposed core20 on
all supported drivers and all supported releases (so, 18.04, 20.04,
21.10).

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to glibc in Ubuntu.
https://bugs.launchpad.net/bugs/1951033

Title:
  20.04 SRU

Status in glibc package in Ubuntu:
  New

Bug description:
  [impact]
  It is time to update glibc in Focal. This bug is a placeholder for the overall process and a place to link the individual bugs that are being targeted:

   bug #1928508 Performance regression on memcpy() calls for AMD Zen
   bug #1899800 Runtime deadlock: pthread_cond_signal failed to wake up pthread_cond_wait due to a bug in undoing stealing
   bug #1892825 update-locale not perform correctly sanity checks
   bug #1918035 ld-2.31.so is not correctly packaged in libc6-dbg
   bug #1951032 AArch64: Backport memcpy improvements

  [test case]
  Each bug listed above has its own test case, of course. And glibc's own test case and the autopkgtests provide a good deal of assurance that things are working. But there are still some additional things we should test by hand.

  1. We should test upgrades interactively in a container and vm and make sure long running processes still function so you can still ssh in and a desktop session continues to operate as expected, ideally on a variety of architectures.
  2. We should build a core20 snap with the glibc from focal-proposed into a branch and test it with a variety of core20 snaps, on a variety of architectures.
  3. For arm64 specifically, we should test that upgrades of machines that do and do not have libc6-lse from 2.31-0ubuntu9.2 installed go smoothly.
  4. We should create a system that has the problematic 2.31-0ubuntu9.3 installed, and check that the behaviour when the new version is available is sensible (the current plan is to refuse such upgrades but I think some more testing here is sensible -- it may be that such upgrades are actually OK).
  5. We should test graphical snaps with a variety of drivers (needs expansion!)

  [regression potential]
  Rebuilding glibc is always a little risky (toolchain bugs and incompatibilities between the old and new versions can be surprising).

  glibc's own tests and the autopkgtests that will be run should catch
  any regression in the new version of glibc itself.

  However, the biggest source of problems recently has been around
  upgrades and interactions between the old and new libcs, whether that
  is different versions of libc6 in a snap and its base or when an long
  running process has the older version mapped but interacts with
  artefacts from the newer version on disk. The tests in this bug are
  aimed at catching any of these problems before it gets to updates.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/1951033/+subscriptions