Re: [PATCH] net/core: Use RCU_INIT_POINTER(x, NULL) in netpoll.c

[Neil Horman <nhorman@tuxdriver.com>]

On Mon, Mar 24, 2014 at 06:14:35AM -0700, Eric Dumazet wrote:
> On Mon, 2014-03-24 at 07:11 -0400, Neil Horman wrote:
> > On Mon, Mar 24, 2014 at 12:42:46AM +0530, Monam Agarwal wrote:
> > > This patch replaces rcu_assign_pointer(x, NULL) with RCU_INIT_POINTER(x, NULL)
> > > 
> > > The rcu_assign_pointer() ensures that the initialization of a structure       
> > > is carried out before storing a pointer to that structure. 
> > > And in the case of the NULL pointer, there is no structure to initialize. 
> > > So, rcu_assign_pointer(p, NULL) can be safely converted to RCU_INIT_POINTER(p, NULL)
> > > 
> > > Signed-off-by: Monam Agarwal <monamagarwal123@gmail.com>
> > No, I don't think this is right. __netpoll_cleanup can be called in parallel
> > with bottom half access to dev->npinfo.  rcu_assign pointer is safe against
> > those accesses, but RCU_INIT_POINTER is not.
> 
> Wow, it is always safe for NULL value.
> 

Ok, After reading Paul's explination of what this is doing:
http://mid.gmane.org/20140320150601.GK4405@linux.vnet.ibm.com

This makes a bit more sense to me, and I see that its safe, but I'm still not
sure that I like it, just from a readability standpoint.  Making the conversion
removes a memory barrier and the use of a volatile variable, but this isn't a
hot path, and the use of rcu_assign_pointer seems more readable to me as its
balanced with the rcu_assign_pointer in __netpoll_setup.  Doesn't seem worth the
tradeoff to me.
Neil