Hi,
This is quite a long response. I can summarize here:
- Several new memory barrier pairs were identified.
- The placement of a memory barrier was incorrect.
There are now quite a few changes queued up for v2. I will try to get
this posted soon. Also, I believe we've now identified the
On Thu 2020-02-27 13:04:09, John Ogness wrote:
> On 2020-02-21, Petr Mladek wrote:
> > If I get it correctly, the used cmpxchg_relaxed() variants does not
> > provide full barriers. They are just able to prevent parallel
> > manipulation of the modified variable.
>
> Correct.
>
> I purposely
On 2020-02-21, Petr Mladek wrote:
> If I get it correctly, the used cmpxchg_relaxed() variants does not
> provide full barriers. They are just able to prevent parallel
> manipulation of the modified variable.
Correct.
I purposely avoided the full barriers of a successful cmpxchg() so that
we