On Sat, Apr 18, 2020 at 07:37:08AM -0500, Eric W. Biederman wrote:
<gregkh@xxxxxxxxxxxxxxxxxxx> writes:
The patch below does not apply to the 5.6-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable@xxxxxxxxxxxxxxx>.
So for anyone who cares the fix I refer to in the commit comment is the
workaround that keeps the past implementation from being a real problem.
I just see this as code cleanup so I can remove the old workaround. The
old workaround will cause posix_cpu_timers_exit_group to be called early
on particular variants of multi-threaded exec, resulting in the
corresponding cpu clock stopping. So this does represent a real fix.
However using a cpu timer of another process to signal things in your
process is rare, and the case is breaks is only in certain obscure
variations of a multi-threaded exec. Further no one has to my knowledge
complained in over a decade.
If someone sees that fix as important, and something that needs to be
backported, it will be easiest to backport my earlier cleanup patches
in the same series.
For 5.6, 5.5, and 5.4 it was enough to take:
60f2ceaa8111 ("posix-cpu-timers: Remove unnecessary locking around cpu_clock_sample_group")
as a dependency. Based on the commit message there it should be safe so
I did just that.
For older branches, as you point out, we need quite a few more cleanups,
so I haven't done that.
--
Thanks,
Sasha
