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

[Thomas Gleixner <tglx@linutronix.de>]

On Fri, 10 Apr 2015, Sebastian Andrzej Siewior wrote:
> >And: please test it, too. (patch the kernel so that you can trigger
> >this case).
> 
> Why patch? Isn't this triggered if you have a reader waiting and you
> send a message?

Manfred referred to the exit race. Though you can spare that exercise
as it is safe by definition if you hold a ref on the task.

Can you please convert that over to Peters lockless wake queues so we
do not reimplement the same thing open coded here.

> +static struct task_struct *pipelined_send(struct mqueue_inode_info *info,
>  				  struct msg_msg *message,
>  				  struct ext_wait_queue *receiver)
>  {
> +	struct task_struct *r_task;
> +
>  	receiver->msg = message;
>  	list_del(&receiver->list);
> -	receiver->state = STATE_PENDING;
> -	wake_up_process(receiver->task);
> +	r_task = receiver->task;
> +	get_task_struct(r_task);
>  	smp_wmb();

While we are at it. The barrier here and the one in pipelined_receive
are not documented and they are missing a proper pairing on the read
side. The comment which you removed was pretty vague about the purpose
of the barrier as well.

Thanks,

	tglx