Re: memory barrier question

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Sep 16, 2010 at 03:30:56PM +0100, David Howells wrote:
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
> > Is the rmb() really needed?
> > 
> > Take this code from fs/namei.c for example:
> > 
> > 		inode = next.dentry->d_inode;
> > 		if (!inode)
> > 			goto out_dput;
> > 
> > 		if (inode->i_op->follow_link) {
> > 
> > It happily dereferences dentry->d_inode without a barrier after
> > checking it for non-null, while that d_inode might have just been
> > initialized on another CPU with a freshly created inode.  There's
> > absolutely no synchornization with that on this side.
> 
> Perhaps it's not necessary; once set, how likely is i_op to be changed once
> I_NEW is cleared?

Are the path_get()s protecting this?

If there is no protection, then something like rcu_dereference() is
needed for the assignment from next.dentry->d_inode.

> > Isn't the fact that we check the pointer for being non-null (together
> > with locking/barrier on the other side) enough to ensure that it's
> > safe to dereference it?
> 
> It's possible that since there's a dependency between the variables on the
> reading CPU that the barrier is not required.
> 
> I think I have to refer that question to Paul.

We would need either one of the rcu_dereference() or smp_read_barrier_depends()
APIs to enforce the dependency, for example, against the compiler.

							Thanx, Paul