[Bug nptl/25847] pthread_cond_signal failed to wake up pthread_cond_wait due to a bug in undoing stealing

[carlos at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=25847

--- Comment #71 from Carlos O'Donell <carlos at redhat dot com> ---
(In reply to Evgeny Morozov from comment #70)
> We seem to have run into this again, after a long pause, or something very
> similar to it. (I can provide the stack traces if anyone is interested.) Now
> it happened on Ubuntu 22.04.5 with glibc 2.35-0ubuntu3.8, which is supposed
> to have the fix from
> https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/1899800.
> 
> Do I understand correctly that the Ubuntu fix is incomplete, so we can still
> run into the bug occasionally with it? Otherwise it must be a different bug.
> 
> What is blocking the proper patch for this bug from being merged?

The solution we ended up with in v5 on the list, the one that simplifies the
code to use just the incremental __g_signals, is one that I'm testing
downstream with direct pre-production deployments to see if we've fixed all the
corner cases. I'm waiting right now to see the results of that pre-production
testing. As you know this isn't a bug where the fix shows up right away, you
have to keep running and not see a failure.

There are a few cases where the v5 fix has a problem, particularly the ABA on
the __g_signals (32-bit roll-over), and the less worrisome ABA on
__g_orig_start (63-bit roll-over). I'm writing some extra code for these with
synthetic test cases to cover.

I don't know if the Ubuntu "fix" is complete or not, you need to raise that
with Canonical. The on-list v5 fix from Malte and Frank is a correct fix.

-- 
You are receiving this mail because:
You are on the CC list for the bug.