On Thu, Jul 16, 2015 at 8:51 PM, Mel Gorman <mgorman@suse.de> wrote:
> On Thu, Jul 16, 2015 at 08:13:38PM +0300, Konstantin Khlebnikov wrote:
>> > @@ -1187,14 +1195,14 @@ void __init page_alloc_init_late(void)
>> >  {pgdat_init_rwsempgdat_init_rwsempgdat_init_rwsem
>> >         int nid;
>> >
>> > +       /* There will be num_node_state(N_MEMORY) threads */
>> > +       atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
>> >         for_each_node_state(nid, N_MEMORY) {
>> > -               down_read(&pgdat_init_rwsem);
>>
>> Rw-sem have special "non-owner" mode for keeping lockdep away.
>> This should be enough:
>>
>
> I think in this case that the completions look nicer though so I think
> I'll keep them.

Ok. Not a big deal, they are anyway in init sections.

BTW there's another option: wait_on_atomic_t / wake_up_atomic_t
like wait_on_bit but atomic_t

>
> --
> Mel Gorman
> SUSE Labs