Re: [PATCH v2] ipc/mqueue: remove STATE_PENDING

[Peter Zijlstra]

On Tue, Apr 28, 2015 at 09:36:50AM -0700, Davidlohr Bueso wrote:
> On Tue, 2015-04-28 at 14:37 +0200, Peter Zijlstra wrote:
> > On Mon, Apr 27, 2015 at 08:24:53PM -0700, Davidlohr Bueso wrote:
> > > +static inline void pipelined_send(struct wake_q_head *wake_q,
> > > +				  struct mqueue_inode_info *info,
> > >  				  struct msg_msg *message,
> > >  				  struct ext_wait_queue *receiver)
> > >  {
> > >  	receiver->msg = message;
> > >  	list_del(&receiver->list);
> > > +	wake_q_add(wake_q, receiver->task);
> > > +	/*
> > > +	 * Ensure that updating receiver->state is the last
> > > +	 * write operation: As once set, the receiver can continue,
> > > +	 * and if we don't have the reference count from the wake_q,
> > > +	 * yet, at that point we can later have a use-after-free
> > > +	 * condition and bogus wakeup.
> > > +	 */
> > > +	smp_wmb(); /* pairs with smp_rmb() in wq_sleep */
> > 
> > You have this barrier because we cannot rely on a failed cmpxchg()
> > actually being a full barrier, right?
> 
> Failed cmpxchg() calls implies that the task is never added to the queue
> (duplicate, which I cannot see occurring in this patch), so nothing
> wrong with the bogus wakeups mentioned in the comment.
> 
> This barrier is not added by this patch though. Currently we have it
> serializing with the wake_up_process() with STATE_READY, for similar
> reasons. Because there is no task refcounting going on, the task can
> easily disappear underneath us if the state is set before the wakeup. I
> applied the same judgment here.

Well, if you can 'guarantee' the cmpxchg will not fail, you can then
rely on the fact that cmpxchg implies a full barrier, which would
obviate the need for the wmb.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/