[lttng-dev] [PATCH 00/11] Add support for TSAN to liburcu

[Olivier Dion]

On Wed, 24 May 2023, Dmitry Vyukov <dvyukov at google.com> wrote:
> On Tue, 23 May 2023 at 18:05, Olivier Dion <odion at efficios.com> wrote:

> I don't this this is true in the C/C++ memory model:
> "the preceding atomic exchange with sequential consistency already
> acts as an implicit release (in term of memory barrier) for the
> following store".
>
> std::atomic_thread_fence does affect all preceding/subsequent
> operations, but an atomic memory operation only affects ordering on
> that variable, it doesn't also serve as a standalone memory fence.

After reading the standard, we concur with you.  We had to revisit the
memory model used by URCU to understand the conflict.  While doing so,
we had to adapt some of the algorithms that were assuming implicit full
memory barriers with operation such as CAS.

Thank for the insight!

-- 
Olivier Dion
EfficiOS Inc.
https://www.efficios.com