On Fri, May 31, 2019 at 7:11 PM Eric Dumazet <eduma...@google.com> wrote:
>
> On Fri, May 31, 2019 at 9:29 AM Andrea Parri
> <andrea.pa...@amarulasolutions.com> wrote:
> >
> > On Fri, May 31, 2019 at 08:45:47AM -0700, Eric Dumazet wrote:
> > > On 5/31/19 7:45 AM, Herbert Xu wrote:
> >
> > > > In this case the code doesn't need them because an implicit
> > > > barrier() (which is *stronger* than READ_ONCE/WRITE_ONCE) already
> > > > exists in both places.
> >
> >
> > > I have already explained that the READ_ONCE() was a leftover of the first
> > > version
> > > of the patch, that I refined later, adding correct (and slightly more
> > > complex) RCU
> > > barriers and rules.
> >
> > AFAICT, neither barrier() nor RCU synchronization can be used as
> > a replacement for {READ,WRITE}_ONCE() here (and in tons of other
> > different situations). IOW, you might want to try harder. ;-)
>
> At least the writer side is using queue_rcu_work() which implies many
> full memory barriers,
> it is not equivalent to a compiler barrier() :/
>
> David, Herbert, I really do not care, I want to move on fixing real
> bugs, not arguing with memory barriers experts.
>
> Lets add back the READ_ONCE() and be happy.
We will have get back to this later with LKMM and KTSAN.
