Re: [PATCH 3/4] fib_trie: print statistics for multiple tables

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

On Tue, Feb 12, 2008 at 06:35:21PM -0800, Andrew Morton wrote:
> On Tue, 12 Feb 2008 16:50:45 -0800 Stephen Hemminger <shemminger@vyatta.com> wrote:
> 
> > +/**
> > + * hlist_for_each_entry_continue_rcu - iterate over rcu hlist after current point
> > + * @tpos:	the type * to use as a loop cursor.
> > + * @pos:	the &struct hlist_node to use as a loop cursor.
> > + * @member:	the name of the hlist_node within the struct.
> > + */
> > +#define hlist_for_each_entry_continue_rcu(tpos, pos, member)		\
> > +	for (pos = (pos)->next;						\
> > +	     rcu_dereference(pos) && ({ prefetch(pos->next); 1;}) &&	\
> > +		     ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1;}); \
> > +	     pos = pos->next)
> 
> Is the compiler allowed to look at a term such as
> 
> 	({ prefetch(pos->next); 1;})
> 
> and, when it is used as a truth value, say "hey, that's always true" and
> then elide the call to prefetch()?  We've no way of telling because this
> remains gcc-specific territory, afaik.

The prefetch() definitions I found are "asm volatile".  So, as I
understand it, the compiler is not supposed to remove it, just as it
would not be permitted to remove something that could have a side effect.

> (cc Paul for rcu stuff)

Given my track record with simple functions of late :-/ I will beat this
one up a bit...

							Thanx, Paul