On Mon, Sep 30, 2024 at 11:42:11AM +0200, Jonas Oberhauser wrote:
>
>
> Am 9/28/2024 um 11:15 PM schrieb Alan Stern:
> > On Sat, Sep 28, 2024 at 11:55:22AM -0400, Mathieu Desnoyers wrote:
> > > On 2024-09-28 17:49, Alan Stern wrote:
> > > > Isn't it true that on strongly ordered CPUs, a compiler barrier is
> > > > sufficient to prevent the rcu_dereference() problem? So the whole idea
> > > > behind ptr_eq() is that it prevents the problem on all CPUs.
> > >
> > > Correct. But given that we have ptr_eq(), it's good to show how it
> > > equally prevents the compiler from reordering address-dependent loads
> > > (comparison with constant) *and* prevents the compiler from using
> > > one pointer rather than the other (comparison between two non-constant
> > > pointers) which affects speculation on weakly-ordered CPUs.
> >
> > I don't see how these two things differ from each other. In the
> > comparison-with-a-constant case, how is the compiler reordering
> > anything? Isn't it just using the constant address rather than the
> > loaded pointer and thereby breaking the address dependency?
>
> I also currently don't see any major difference between the constant and
> register case. The point is that the address is known before loading into b,
> and hence the compiler + hardware can speculatively load *b before loading
> into b.
>
> The only difference is how far before loading into b the address is known.
In theory, true. In practice, in the register case, you need a little
more bad luck for the compiler to be able to exploit your mistake.
Still, it is indeed far better to attain a state of bliss by keeping
the compiler ignorant.
Thanx, Paul
