> spin_lock_irqsave(&qp_info->send_queue.lock, flags);
> if (qp_info->send_queue.count < qp_info->send_queue.max_active) {
>+ qp_info->send_queue.count++;
>
>+ spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
>
> ret = ib_post_send(mad_agent->qp, &mad_send_wr->send_wr,
> &bad_send_wr);
>
>+ spin_lock_irqsave(&qp_info->send_queue.lock, flags);
> list = &qp_info->send_queue.list;
> } else {
> ret = 0;
>+ qp_info->send_queue.count++;
> list = &qp_info->overflow_list;
> }
>
> if (!ret)
> list_add_tail(&mad_send_wr->mad_list.list, list);
>+ else
>+ qp_info->send_queue.count--;
It's not quite this simple. Once the lock is released before calling
ib_post_send, another thread could come down and queue a MAD to the overflow
list. If ib_post_send fails, the overflow list must be checked to see if a
queued mad should now be sent.
As for being able to hold a lock when calling ib_post_send, that's something
that should be allowed.
- Sean
_______________________________________________
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
