mad.c

h内核

	if (mad_snoop_priv->snoop_index < 0) {
		ret = ERR_PTR(mad_snoop_priv->snoop_index);
		goto error2;
	}

	atomic_set(&mad_snoop_priv->refcount, 1);
	return &mad_snoop_priv->agent;

error2:
	kfree(mad_snoop_priv);
error1:
	return ret;
}
EXPORT_SYMBOL(ib_register_mad_snoop);

static void unregister_mad_agent(struct ib_mad_agent_private *mad_agent_priv)
{
	struct ib_mad_port_private *port_priv;
	unsigned long flags;

	/* Note that we could still be handling received MADs */

	/*
	 * Canceling all sends results in dropping received response
	 * MADs, preventing us from queuing additional work
	 */
	cancel_mads(mad_agent_priv);

	port_priv = mad_agent_priv->qp_info->port_priv;
	cancel_delayed_work(&mad_agent_priv->timed_work);
	flush_workqueue(port_priv->wq);

	spin_lock_irqsave(&port_priv->reg_lock, flags);
	remove_mad_reg_req(mad_agent_priv);
	list_del(&mad_agent_priv->agent_list);
	spin_unlock_irqrestore(&port_priv->reg_lock, flags);

	/* XXX: Cleanup pending RMPP receives for this agent */

	atomic_dec(&mad_agent_priv->refcount);
	wait_event(mad_agent_priv->wait,
		   !atomic_read(&mad_agent_priv->refcount));

	if (mad_agent_priv->reg_req)
		kfree(mad_agent_priv->reg_req);
	kfree(mad_agent_priv);
}

static void unregister_mad_snoop(struct ib_mad_snoop_private *mad_snoop_priv)
{
	struct ib_mad_qp_info *qp_info;
	unsigned long flags;

	qp_info = mad_snoop_priv->qp_info;
	spin_lock_irqsave(&qp_info->snoop_lock, flags);
	qp_info->snoop_table[mad_snoop_priv->snoop_index] = NULL;
	atomic_dec(&qp_info->snoop_count);
	spin_unlock_irqrestore(&qp_info->snoop_lock, flags);

	atomic_dec(&mad_snoop_priv->refcount);
	wait_event(mad_snoop_priv->wait,
		   !atomic_read(&mad_snoop_priv->refcount));

	kfree(mad_snoop_priv);
}

/*
 * ib_unregister_mad_agent - Unregisters a client from using MAD services
 */
int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent)
{
	struct ib_mad_agent_private *mad_agent_priv;
	struct ib_mad_snoop_private *mad_snoop_priv;

	/* If the TID is zero, the agent can only snoop. */
	if (mad_agent->hi_tid) {
		mad_agent_priv = container_of(mad_agent,
					      struct ib_mad_agent_private,
					      agent);
		unregister_mad_agent(mad_agent_priv);
	} else {
		mad_snoop_priv = container_of(mad_agent,
					      struct ib_mad_snoop_private,
					      agent);
		unregister_mad_snoop(mad_snoop_priv);
	}
	return 0;
}
EXPORT_SYMBOL(ib_unregister_mad_agent);

static void dequeue_mad(struct ib_mad_list_head *mad_list)
{
	struct ib_mad_queue *mad_queue;
	unsigned long flags;

	BUG_ON(!mad_list->mad_queue);
	mad_queue = mad_list->mad_queue;
	spin_lock_irqsave(&mad_queue->lock, flags);
	list_del(&mad_list->list);
	mad_queue->count--;
	spin_unlock_irqrestore(&mad_queue->lock, flags);
}

static void snoop_send(struct ib_mad_qp_info *qp_info,
		       struct ib_send_wr *send_wr,
		       struct ib_mad_send_wc *mad_send_wc,
		       int mad_snoop_flags)
{
	struct ib_mad_snoop_private *mad_snoop_priv;
	unsigned long flags;
	int i;

	spin_lock_irqsave(&qp_info->snoop_lock, flags);
	for (i = 0; i < qp_info->snoop_table_size; i++) {
		mad_snoop_priv = qp_info->snoop_table[i];
		if (!mad_snoop_priv ||
		    !(mad_snoop_priv->mad_snoop_flags & mad_snoop_flags))
			continue;

		atomic_inc(&mad_snoop_priv->refcount);
		spin_unlock_irqrestore(&qp_info->snoop_lock, flags);
		mad_snoop_priv->agent.snoop_handler(&mad_snoop_priv->agent,
						    send_wr, mad_send_wc);
		if (atomic_dec_and_test(&mad_snoop_priv->refcount))
			wake_up(&mad_snoop_priv->wait);
		spin_lock_irqsave(&qp_info->snoop_lock, flags);
	}
	spin_unlock_irqrestore(&qp_info->snoop_lock, flags);
}

static void snoop_recv(struct ib_mad_qp_info *qp_info,
		       struct ib_mad_recv_wc *mad_recv_wc,
		       int mad_snoop_flags)
{
	struct ib_mad_snoop_private *mad_snoop_priv;
	unsigned long flags;
	int i;

	spin_lock_irqsave(&qp_info->snoop_lock, flags);
	for (i = 0; i < qp_info->snoop_table_size; i++) {
		mad_snoop_priv = qp_info->snoop_table[i];
		if (!mad_snoop_priv ||
		    !(mad_snoop_priv->mad_snoop_flags & mad_snoop_flags))
			continue;

		atomic_inc(&mad_snoop_priv->refcount);
		spin_unlock_irqrestore(&qp_info->snoop_lock, flags);
		mad_snoop_priv->agent.recv_handler(&mad_snoop_priv->agent,
						   mad_recv_wc);
		if (atomic_dec_and_test(&mad_snoop_priv->refcount))
			wake_up(&mad_snoop_priv->wait);
		spin_lock_irqsave(&qp_info->snoop_lock, flags);
	}
	spin_unlock_irqrestore(&qp_info->snoop_lock, flags);
}

static void build_smp_wc(u64 wr_id, u16 slid, u16 pkey_index, u8 port_num,
			 struct ib_wc *wc)
{
	memset(wc, 0, sizeof *wc);
	wc->wr_id = wr_id;
	wc->status = IB_WC_SUCCESS;
	wc->opcode = IB_WC_RECV;
	wc->pkey_index = pkey_index;
	wc->byte_len = sizeof(struct ib_mad) + sizeof(struct ib_grh);
	wc->src_qp = IB_QP0;
	wc->qp_num = IB_QP0;
	wc->slid = slid;
	wc->sl = 0;
	wc->dlid_path_bits = 0;
	wc->port_num = port_num;
}

/*
 * Return 0 if SMP is to be sent
 * Return 1 if SMP was consumed locally (whether or not solicited)
 * Return < 0 if error
 */
static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
				  struct ib_smp *smp,
				  struct ib_send_wr *send_wr)
{
	int ret, alloc_flags, solicited;
	unsigned long flags;
	struct ib_mad_local_private *local;
	struct ib_mad_private *mad_priv;
	struct ib_mad_port_private *port_priv;
	struct ib_mad_agent_private *recv_mad_agent = NULL;
	struct ib_device *device = mad_agent_priv->agent.device;
	u8 port_num = mad_agent_priv->agent.port_num;
	struct ib_wc mad_wc;

	if (!smi_handle_dr_smp_send(smp, device->node_type, port_num)) {
		ret = -EINVAL;
		printk(KERN_ERR PFX "Invalid directed route\n");
		goto out;
	}
	/* Check to post send on QP or process locally */
	ret = smi_check_local_dr_smp(smp, device, port_num);
	if (!ret || !device->process_mad)
		goto out;

	if (in_atomic() || irqs_disabled())
		alloc_flags = GFP_ATOMIC;
	else
		alloc_flags = GFP_KERNEL;
	local = kmalloc(sizeof *local, alloc_flags);
	if (!local) {
		ret = -ENOMEM;
		printk(KERN_ERR PFX "No memory for ib_mad_local_private\n");
		goto out;
	}
	local->mad_priv = NULL;
	local->recv_mad_agent = NULL;
	mad_priv = kmem_cache_alloc(ib_mad_cache, alloc_flags);
	if (!mad_priv) {
		ret = -ENOMEM;
		printk(KERN_ERR PFX "No memory for local response MAD\n");
		kfree(local);
		goto out;
	}

	build_smp_wc(send_wr->wr_id, smp->dr_slid, send_wr->wr.ud.pkey_index,
		     send_wr->wr.ud.port_num, &mad_wc);

	/* No GRH for DR SMP */
	ret = device->process_mad(device, 0, port_num, &mad_wc, NULL,
				  (struct ib_mad *)smp,
				  (struct ib_mad *)&mad_priv->mad);
	switch (ret)
	{
	case IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY:
		/*
		 * See if response is solicited and
		 * there is a recv handler
		 */
		if (solicited_mad(&mad_priv->mad.mad) &&
		    mad_agent_priv->agent.recv_handler) {
			local->mad_priv = mad_priv;
			local->recv_mad_agent = mad_agent_priv;
			/*
			 * Reference MAD agent until receive
			 * side of local completion handled
			 */
			atomic_inc(&mad_agent_priv->refcount);
		} else
			kmem_cache_free(ib_mad_cache, mad_priv);
		break;
	case IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED:
		kmem_cache_free(ib_mad_cache, mad_priv);
		break;
	case IB_MAD_RESULT_SUCCESS:
		/* Treat like an incoming receive MAD */
		solicited = solicited_mad(&mad_priv->mad.mad);
		port_priv = ib_get_mad_port(mad_agent_priv->agent.device,
					    mad_agent_priv->agent.port_num);
		if (port_priv) {
			mad_priv->mad.mad.mad_hdr.tid =
				((struct ib_mad *)smp)->mad_hdr.tid;
			recv_mad_agent = find_mad_agent(port_priv,
						       &mad_priv->mad.mad,
							solicited);
		}
		if (!port_priv || !recv_mad_agent) {
			kmem_cache_free(ib_mad_cache, mad_priv);
			kfree(local);
			ret = 0;
			goto out;
		}
		local->mad_priv = mad_priv;
		local->recv_mad_agent = recv_mad_agent;
		break;
	default:
		kmem_cache_free(ib_mad_cache, mad_priv);
		kfree(local);
		ret = -EINVAL;
		goto out;
	}

	local->send_wr = *send_wr;
	local->send_wr.sg_list = local->sg_list;
	memcpy(local->sg_list, send_wr->sg_list,
	       sizeof *send_wr->sg_list * send_wr->num_sge);
	local->send_wr.next = NULL;
	local->tid = send_wr->wr.ud.mad_hdr->tid;
	local->wr_id = send_wr->wr_id;
	/* Reference MAD agent until send side of local completion handled */
	atomic_inc(&mad_agent_priv->refcount);
	/* Queue local completion to local list */
	spin_lock_irqsave(&mad_agent_priv->lock, flags);
	list_add_tail(&local->completion_list, &mad_agent_priv->local_list);
	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
	queue_work(mad_agent_priv->qp_info->port_priv->wq,
		  &mad_agent_priv->local_work);
	ret = 1;
out:
	return ret;
}

static int ib_send_mad(struct ib_mad_agent_private *mad_agent_priv,
		       struct ib_mad_send_wr_private *mad_send_wr)
{
	struct ib_mad_qp_info *qp_info;
	struct ib_send_wr *bad_send_wr;
	unsigned long flags;
	int ret;

	/* Replace user's WR ID with our own to find WR upon completion */
	qp_info = mad_agent_priv->qp_info;
	mad_send_wr->wr_id = mad_send_wr->send_wr.wr_id;
	mad_send_wr->send_wr.wr_id = (unsigned long)&mad_send_wr->mad_list;
	mad_send_wr->mad_list.mad_queue = &qp_info->send_queue;

	spin_lock_irqsave(&qp_info->send_queue.lock, flags);
	if (qp_info->send_queue.count++ < qp_info->send_queue.max_active) {
		list_add_tail(&mad_send_wr->mad_list.list,
			      &qp_info->send_queue.list);
		spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
		ret = ib_post_send(mad_agent_priv->agent.qp,
				   &mad_send_wr->send_wr, &bad_send_wr);
		if (ret) {
			printk(KERN_ERR PFX "ib_post_send failed: %d\n", ret);
			dequeue_mad(&mad_send_wr->mad_list);
		}
	} else {
		list_add_tail(&mad_send_wr->mad_list.list,
			      &qp_info->overflow_list);
		spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
		ret = 0;
	}
	return ret;
}

/*
 * ib_post_send_mad - Posts MAD(s) to the send queue of the QP associated
 *  with the registered client
 */
int ib_post_send_mad(struct ib_mad_agent *mad_agent,
		     struct ib_send_wr *send_wr,
		     struct ib_send_wr **bad_send_wr)
{
	int ret = -EINVAL;
	struct ib_mad_agent_private *mad_agent_priv;

	/* Validate supplied parameters */
	if (!bad_send_wr)
		goto error1;

	if (!mad_agent || !send_wr)
		goto error2;

	if (!mad_agent->send_handler)
		goto error2;

	mad_agent_priv = container_of(mad_agent,
				      struct ib_mad_agent_private,
				      agent);

	/* Walk list of send WRs and post each on send list */
	while (send_wr) {
		unsigned long			flags;
		struct ib_send_wr		*next_send_wr;
		struct ib_mad_send_wr_private	*mad_send_wr;
		struct ib_smp			*smp;

		/* Validate more parameters */
		if (send_wr->num_sge > IB_MAD_SEND_REQ_MAX_SG)
			goto error2;

		if (send_wr->wr.ud.timeout_ms && !mad_agent->recv_handler)
			goto error2;

		if (!send_wr->wr.ud.mad_hdr) {
			printk(KERN_ERR PFX "MAD header must be supplied "
			       "in WR %p\n", send_wr);
			goto error2;
		}

		/*
		 * Save pointer to next work request to post in case the
		 * current one completes, and the user modifies the work
		 * request associated with the completion
		 */
		next_send_wr = (struct ib_send_wr *)send_wr->next;

		smp = (struct ib_smp *)send_wr->wr.ud.mad_hdr;
		if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
			ret = handle_outgoing_dr_smp(mad_agent_priv, smp,
						     send_wr);
			if (ret < 0)		/* error */
				goto error2;
			else if (ret == 1)	/* locally consumed */
				goto next;
		}

		/* Allocate MAD send WR tracking structure */
		mad_send_wr = kmalloc(sizeof *mad_send_wr,
				      (in_atomic() || irqs_disabled()) ?
				      GFP_ATOMIC : GFP_KERNEL);
		if (!mad_send_wr) {
			printk(KERN_ERR PFX "No memory for "
			       "ib_mad_send_wr_private\n");
			ret = -ENOMEM;
			goto error2;
		}

		mad_send_wr->send_wr = *send_wr;
		mad_send_wr->send_wr.sg_list = mad_send_wr->sg_list;
		memcpy(mad_send_wr->sg_list, send_wr->sg_list,
		       sizeof *send_wr->sg_list * send_wr->num_sge);
		mad_send_wr->send_wr.next = NULL;
		mad_send_wr->tid = send_wr->wr.ud.mad_hdr->tid;
		mad_send_wr->agent = mad_agent;
		/* Timeout will be updated after send completes */
		mad_send_wr->timeout = msecs_to_jiffies(send_wr->wr.
							ud.timeout_ms);
		mad_send_wr->retry = 0;
		/* One reference for each work request to QP + response */
		mad_send_wr->refcount = 1 + (mad_send_wr->timeout > 0);
		mad_send_wr->status = IB_WC_SUCCESS;

		/* Reference MAD agent until send completes */
		atomic_inc(&mad_agent_priv->refcount);
		spin_lock_irqsave(&mad_agent_priv->lock, flags);
		list_add_tail(&mad_send_wr->agent_list,
			      &mad_agent_priv->send_list);
		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);

		ret = ib_send_mad(mad_agent_priv, mad_send_wr);
		if (ret) {
			/* Fail send request */
			spin_lock_irqsave(&mad_agent_priv->lock, flags);
			list_del(&mad_send_wr->agent_list);
			spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
			atomic_dec(&mad_agent_priv->refcount);
			goto error2;
		}
next:
		send_wr = next_send_wr;
	}
	return 0;

error2:
	*bad_send_wr = send_wr;
error1:
	return ret;
}
EXPORT_SYMBOL(ib_post_send_mad);

/*
 * ib_free_recv_mad - Returns data buffers used to receive
 *  a MAD to the access layer
 */
void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc)
{
	struct ib_mad_recv_buf *entry;
	struct ib_mad_private_header *mad_priv_hdr;
	struct ib_mad_private *priv;

	mad_priv_hdr = container_of(mad_recv_wc,
				    struct ib_mad_private_header,
				    recv_wc);
	priv = container_of(mad_priv_hdr, struct ib_mad_private, header);