BZ 20822 :powerpc: race condition in __lll_unlock_elision

[Torvald Riegel]

On Tue, 2016-11-22 at 16:02 +0100, Florian Weimer wrote:
> On 11/22/2016 02:45 PM, Torvald Riegel wrote:
> 
> > Lock elision relies on this kind of mixed access.  While it would be
> > nice to have a common formal model for the various HTMs out there from
> > the perspective of a C11/C++11 memory model setting, I don't think it's
> > a big problem right now that we don't (AFAIK) have such a formal model.
> > Lock elision should be well understood, so I'm not worried about any
> > surprises regarding this use case.
> 
> You are the expert.  I'm just surprised we embrace a wild-west approach 
> to P&C again, while trying to convince others to argue from the 
> definitions instead based on gut feeling.

To avoid misunderstandings: I'm not arguing in favor of embracing any
wild-west approach.  It would be great if we had a formal model that can
cover more than one architecture's semantics.  Nonetheless, we don't
have it, and I don't see this as a big enough problem to prevent us from
trying to use lock elision.  Once someone comes up with a model, I
suppose we'd certainly try to use it if possible.