Re: + do_wait-wakeup-optimization.patch added to -mm tree

[Roland McGrath <roland@redhat.com>]

> Let's suppose the ptracer finds the EXIT_ZOMBIE tracee and notifies its
> ->real_parent which sleeps in do_wait(). In that case the usage of
> eligible_child(task == ptracer) above is bogus, and checking for
> group_leader is not rifgt too.

I had overlooked that do_notify_parent() call.

> > +static int do_wait_wake_function(wait_queue_t *curr, unsigned mode, int sync,
> > +				 void *key)
> > +{
> > +	struct task_struct *task = current;
> 
> I think we can fix (and simplify) this code if we change __wake_up_parent(),
> it should call __wake_up(key => p), so we can do
> 
> 	struct task_struct *task = key;

I had not looked into the bowels of various __wake_up variants, just
assumed it would stay as it is and use wake_up_interruptible_sync.

That would certainly be cleaner.  Then do_wait_wake_function would not need
the second of its special cases, only the one double-check for the
thread_group_leader && task_detached case.

I don't see an exposed __wake_up* variant that both passes a "key" pointer
through and does "sync".  For __wake_up_parent, "sync" is quite desireable.

> > +	if (!needs_wakeup(task, w))
> > +		return 0;
> > +
> > +	return default_wake_function(curr, mode, sync, key);
> 
> perhaps autoremove_wake_function() makes more sense.

Why?  The do_wait loop will have to go through again and still might just
sleep again.  The explicit remove at the end of do_wait seems fine to me.


Thanks,
Roland